⚠ EARNINGS TODAY — READ THIS FIRST

> AMGEN REPORTS SECOND-QUARTER 2026 RESULTS ON 2026-08-04 — THE DATE OF THIS DIVE.

>

> The vendor earnings calendar carries an entry dated 2026-08-04 with epsActual: null and revenueActual: null. The estimate fields are populated: non-GAAP EPS $5.62 and revenue $9,426,191,000. A null actual on the dive date means the print has not landed at the time this was written; Amgen's historical pattern in this file is a release on the report date, and the quote timestamp of 1785873601 = 2026-08-04T20:00:01Z is the 16:00 Eastern close. The overwhelming likelihood is that the number lands after this dive is struck.

>

> What consensus is asking for. $9,426M of revenue is +2.8% on the $9,167M delivered in the June 2025 quarter. $5.62 of non-GAAP EPS is a 6.6% DECLINE on the $6.02 delivered a year ago. Consensus expects Amgen's earnings to go backwards this quarter.

>

> What the record says. Amgen has beaten the consensus EPS estimate in all five prints in this file — but by a decaying margin: +14.8% (May 2025), +14.0% (Aug 2025), +12.4% (Nov 2025), +11.8% (Feb 2026), +8.0% (Apr 2026).

>

> What that does to the entry. You would be buying 0.8% below the 52-week closing high of $393.10, at the 97.5th percentile of the annual range, 9.0% above the 50-day average and 13.1% above the 200-day, with RSI at 71.03 — overbought — on a day the stock rose 2.94%, into a number you cannot see, six days after the company declared a material cybersecurity incident. This is the entry question and the verdict below is set with it as an explicit input.

>

> Nothing in this dive uses any second-quarter 2026 figure. Every number here is from the 10-K filed 2026-02-13 (FY2025), the 10-Q filed 2026-05-01 (March 2026 quarter), the three 8-Ks, or the vendor payload — all of which pre-date tonight.

Reference table

1. What the business is, and how the disclosure actually works

Amgen describes itself in the 10-K as a company that "discovers, develops, manufactures and delivers innovative medicines to fight some of the world's toughest diseases," which "helped launch the biotechnology industry more than 45 years ago." Incorporated in California in 1980, reincorporated in Delaware in 1987, headquartered in Thousand Oaks, California. Chairman and chief executive Robert A. Bradway. The vendor profile lists 31,500 full-time employees; the extracted 10-K text does not carry a headcount figure, so that number is vendor-sourced and labelled as such.

The single most important disclosure-structure fact: the 10-K states plainly, "Amgen operates in one operating segment: human therapeutics. Therefore, our results of operations are discussed on a consolidated basis." There are no reportable segments. Anything in a vendor file labelled seg_prod for Amgen is therefore not a segment table — it is the worldwide product sales table from the MD&A, plus one line of non-product revenue. That distinction matters and Section 8 records what the vendor got wrong in it.

The revenue architecture is two lines, not one. Amgen reports product sales and other revenues, summing to total revenues:

Both legs of the vendor's revenue split reconcile to the 10-K's own stated growth percentages. That is a clean check and it is worth recording, because the product-level table underneath it is not clean (Section 8, finding 4).

Product sales, 2023-2025, from the vendor table relabelled to the 10-K's own vocabulary

Three readings follow from that table and they are the whole business.

First, there is no single point of failure — and that is a genuine strength. The largest product is Prolia at 12.0% of total revenue. Nothing else is above 8.2%. Contrast that with the customer concentration, which is severe: the 10-K states that "product sales to three large wholesalers, McKesson Corporation, Cencora, Inc. and Cardinal Health, Inc., each individually accounted for more than 10% of total revenues" in each of 2025, 2024 and 2023, and "on a combined basis, these wholesalers accounted for 77%, 77% and 79% of worldwide gross revenues." Amgen's product risk is diversified; its distribution risk is not.

Second, the growth is concentrated in exactly six names. Repatha, EVENITY, TEZSPIRE, BLINCYTO, KRYSTEXXA and the "Other products" bucket — which contains the newer launches PAVBLU, UPLIZNA, IMDELLTRA and the biosimilars WEZLANA and BKEMV — supplied essentially all of the 2025 increase. "Other products" alone added $1.633B, the largest single contribution in the table.

Third, the decline is concentrated in the oldest, most valuable franchises. ENBREL −32.9%, XGEVA −6.3%, Prolia flat at +0.9% and about to turn sharply negative. Section 2 is about that.

Geography. The vendor's seg_geo block gives only a two-line United States / non-US split, which is all Amgen discloses:

The mix is stable and the US share is drifting slightly up — which is the opposite of what you want when the US is where Medicare price setting, 340B expansion and Most-Favored-Nation pricing all live. The 2025 lines sum to $36.751B against reported revenue of $36.741B, a $10M overstatement (0.03%) that is almost certainly rounding; 2024, 2023 and 2022 sum exactly.

Pipeline, from the 10-K's own product-candidate section. Named late-stage and notable assets, in the filing's words: MariTide (maridebart cafraglutide), "a differentiated antibody-peptide conjugate that activates the GLP-1 receptor and antagonizes the GIPR," in chronic weight management, cardiovascular disease, heart failure, obstructive sleep apnea and Type 2 diabetes; olpasiran, "a small interfering RNA (siRNA) that lowers lipoprotein(a)," in atherosclerotic cardiovascular disease; xaluritamig, an anti-STEAP1 x anti-CD3 bispecific in prostate cancer; IMDELLTRA/IMDYLLTRA, an anti-DLL3 x anti-CD3 bispecific in small-cell lung cancer; inebilizumab in lupus nephritis; daxdilimab in dermatomyositis and discoid lupus; dazodalibep in Sjögren's disease; LUMAKRAS in colorectal cancer and NSCLC; pegloticase subcutaneous; TEZSPIRE in COPD and eosinophilic esophagitis; TEPEZZA subcutaneous. No trial dates, enrolment figures or readout timings appear in the extracted filing text and none are supplied here.

2. The single most important thing to understand about these numbers: volume up, price down, and the cliff has already started

Every headline number Amgen reports is the net of two very large opposing forces, and the filing quantifies both.

> 10-K, 2026-02-13: "Total product sales increased 10% in 2025, driven by volume growth of 13%, partially offset by declines in net selling price of 3%. U.S. volume grew 13% and ROW volume grew 14%."

> 10-Q, 2026-05-01: "Total product sales increased 4% for the three months ended March 31, 2026, driven by volume growth of 9%, partially offset by declines in net selling price of 2% and 2% from lower wholesaler and end user inventory."

Read those two sentences next to each other and the trajectory is unambiguous: volume growth decelerated from 13% to 9%, and total product sales growth decelerated from 10% to 4%, in one year. Total revenue growth was +5.8% in the March 2026 quarter — the gap between that and the +4% product number is other revenues, which rose 45% on corporate partner and royalty income. A quarter in which non-product revenue does that much of the work is a quarter to look at carefully.

Now the specific cause, in the company's own words, from the 10-K:

> "Our patents for RANKL antibodies, including sequences, for Prolia and XGEVA expired in February 2025 in the United States and in November 2025 in select countries in Europe, and we expect accelerated sales erosion driven by increased competition, as multiple biosimilars have launched in the United States and ROW."

And what that looks like in the first quarter it fully bit, from the 10-Q:

A 17% volume decline in the largest product in the company, in the first full quarter of biosimilar competition, is the single most important fact in this dive. The 10-Q repeats the guidance verbatim: "For 2026, we continue to expect accelerated sales erosion."

And it is not the only price event. ENBREL, the third-largest branded product at $2.226B, fell 32.9% in 2025. The 10-K attributes it to "lower net selling price of 36% resulting from the impact of increased 340B Program mix, U.S. Medicare Part D redesign and higher commercial discounts, partially offset by volume growth of 4%." Volume grew 4% and revenue fell a third. The 10-Q then adds that the March-2026 decline came "primarily driven by unfavorable changes to estimated sales deductions of 18% and lower net selling price of 15% resulting from the impact of U.S. Medicare Part D price setting under the IRA, effective January 1, 2026."

Three further policy facts, all from the 10-K, all dated:

• Otezla ($2.265B) "was selected by CMS for Medicare price setting that will be applicable beginning in 2027," and Amgen took $1.2B of Otezla intangible impairment charges in 2025 as a direct consequence.
• The Colorado Prescription Drug Affordability Board "deemed three of five drugs 'unaffordable,' including ENBREL, and in October 2025 the Colorado PDAB established an Upper Payment Limit (UPL) substantially lower than the wholesale acquisition cost of ENBREL that would be generally applicable to all formulations of ENBREL, effective no earlier than January 1, 2027." Washington state selected ENBREL for affordability review in July 2025.
• Medicare price setting expands "starting with 10 drugs effective January 1, 2026, adding 15 in 2027 and 2028, and adding 20 in 2029 and subsequent years."

This is the point of the section. Amgen's demand is genuinely strong and its portfolio is genuinely diversified. But the pricing environment is one-directional, the two products most exposed to biosimilars are 17.7% of revenue and are eroding now, and the company itself has told you so in plain language. That is why consensus revenue growth is 2.6-2.7% and not 8-10%, and it is why paying the 52-week high requires believing something the filing does not say.

3. Growth and margin trajectory — and a gross margin the vendor gets wrong

Six years of annual results, from inc_a:

The 2023 step is Horizon. Amgen closed the Horizon Therapeutics acquisition on 6 October 2023, funded by $24.0B of senior notes issued 2 March 2023 plus $4.0B drawn under a term loan (10-K). That is why 2024 revenue jumped 18.6%, why gross margin collapsed from 70.1% to 61.5%, and why 2024 GAAP EPS fell to $7.56 — the acquisition brought a very large slug of intangible amortisation into cost of sales, and a fair-value step-up on acquired inventory on top of it.

Now the correction, and it matters. The vendor reports FY2025 cost of revenue of $10.729B, a gross margin of 70.8% and therefore a 930 basis point margin expansion over 2024. The 10-K says something different, in a sentence:

> "Cost of sales decreased to 32.8% of total revenues for 2025, driven by lower amortization expense from acquisition-related assets, including the fair value step-up of inventory acquired from Horizon, and lower manufacturing costs, partially offset by higher profit share expense and changes in our sales mix."

> "Cost of sales increased to 38.5% of total revenues for 2024..."

32.8% of $36.741B is $12.051B of cost of sales, not $10.729B — a $1.322B gap — and it implies a gross margin of 67.2%, not 70.8%. The 2024 figure, by contrast, reconciles exactly: $12.858B / $33.424B = 38.47%, matching the filing's 38.5% to the decimal. The vendor computed the same field on a different basis in adjacent years and manufactured roughly 360 basis points of gross-margin expansion that did not happen. This is a textbook instance of confirmed defect class 4, and per the standing rule the filing wins: we use 67.2% for 2025 and state the true margin move as 61.5% → 67.2%, a 570 basis point recovery, not 930.

A necessary honesty caveat. The numeric statements are stripped from the extracted filing text, so we cannot see WHERE the $1.322B lands. The vendor's FY2025 operating income of $10.676B ties arithmetically to a verified chain — operating income $10.676B less interest expense $2.755B plus other income $1.055B equals income before tax $8.976B, less tax $1.265B equals net income $7.711B, divided by 542M diluted shares equals $14.23, the reported EPS. So either operating income is $1.32B lower and other income $1.32B higher, or some other expense line is misallocated. We cannot resolve it and we do not pretend to. What we do is: use the filing's 67.2% gross margin, flag the gross-profit / operating-expense split as unreliable, and run every valuation conclusion in this dive off EPS and revenue, which are independently verified.

Quarterly, from inc_q, with the Q4-2025 gross margin flagged:

Two things to take from this table.

One: the Q1-2026 gross margin is filing-verified and the Q4-2025 one is not. The 10-Q states "Cost of sales decreased to 31.8% of total revenues for the three months ended March 31, 2026"; the vendor's $2.744B on $8.618B is 31.84% — an exact match. The quarterly series is trustworthy at the margin level except for Q4 2025, where an implied 81.9% gross margin and a 39.8% operating margin are both wild outliers against every other quarter in six years.

Two: revenue growth is decelerating and the beat margin is decaying. Year-on-year revenue: +9.4%, +9.3%, +12.4%, +8.9%, +5.8%. Beat magnitude on non-GAAP EPS: +14.8%, +14.0%, +12.4%, +11.8%, +8.0%. Both series point the same way, and tonight's consensus asks for +2.8% revenue and a 6.6% EPS decline.

Research and development was $6.974B in 2025 per the annual row (the quarterly rows sum to $7.281B — see Section 8, finding 6), 19.0% of revenue on the annual basis and 20.2% on the trailing-twelve-month basis per ratios_ttm. The 10-K attributes the increase to "investments in Later-Stage Clinical Programs, including those related to MariTide, and in Research and Early Pipeline" and adds "we expect to continue to grow our spend on Later-Stage Clinical Programs as we advance our pipeline." Six global Phase 3 obesity studies are not cheap and the R&D line is going to keep rising.

Effective tax rate. Trailing 14.16% per ratios_ttm; the 10-Q gives 12.7% for the March 2026 quarter against 12.3% a year earlier. The 10-K is explicit about where it comes from: "Substantially all of the benefit to our effective tax rate from foreign earnings results from locations in which the Company has significant manufacturing operations, including Singapore, Ireland and Puerto Rico... Our operations in Puerto Rico are subject to tax incentive grants through 2050 and the Company's operations in Singapore are subject to a tax incentive grant through 2036." The 10-Q then notes the company is now subject to a 15% global minimum tax rate on adjusted financial statement income, and that US tax on foreign earnings runs at a reduced 12.6% from 1 January 2026. A tax rate this low, this dependent on two jurisdictions, and this actively litigated (Section 4) is a risk line, not a strength line.

4. Balance sheet and the enterprise-value rebuild

Annual balance sheets from bal_a, with the 10-Q update:

The FY2025 debt figure reconciles to the 10-K exactly and this is worth stating. The 10-K: "As of December 31, 2025, we had outstanding notes with an aggregate carrying value of $52.8 billion" and, of the $4.0B term loan drawn for Horizon, "$1.8 billion of borrowings was outstanding as of December 31, 2025." $52.8B + $1.8B = $54.6B against the vendor's totalDebt of $54.604B. Vendor and filing agree to the decimal.

But bal_a[0] is one quarter stale, and it misses a $4.0B financing. The 10-Q filed 2026-05-01 discloses:

> "During the three months ended March 31, 2026, we issued $4.0 billion of debt consisting of $1.0 billion of the 4.20% 2031 Notes, $1.75 billion of the 4.85% 2036 Notes, $500 million of the 5.50% 2046 Notes and $750 million of the 5.65% 2056 Notes."

Against that, in the same quarter Amgen "repaid the €750 million aggregate principal amount of the 2.00% 2026 euro Notes ($833 million upon settlement of the related cross-currency swap)" and "repurchased an aggregate principal amount of our debt of $324 million... for an aggregate cost of $233 million, which resulted in a $90 million gain on extinguishment of debt." Net: +$4.0B − $0.833B − $0.324B = +$2.84B. The 10-Q's fair-value note confirms the arithmetic from the other side: the carrying value of fixed-rate debt went from $52.8B at 2025-12-31 to $55.5B at 2026-03-31, a $2.7B increase.

So: the vendor reports total debt of $54.604B at 2025-12-31; the 10-Q shows approximately $57.3B at 2026-03-31; we use $57.3B. The term loan is not separately quantified at 31 March 2026 in the extracted text and we carry the $1.8B forward as an assumption, labelled as such.

Correction 1 — is there a hidden short-term-investment balance? NO, and this one is clean. The recurring vendor defect of netting debt against cash only and ignoring marketable securities does not apply to Amgen, because Amgen holds none. The 10-Q's liquidity paragraph is a single line: "Our balance of cash and cash equivalents was $12.0 billion as of March 31, 2026." There is no "and marketable securities" clause. The 2023 cash-flow statement shows "net cash inflows related to marketable securities of $1.7 billion" — Amgen liquidated the portfolio to help pay for Horizon and has not rebuilt it. shortTermInvestments: 0 is correct, not a null masquerading as a zero. Verified clean, and stated as a finding.

Correction 2 — what about longTermInvestments of $6.804B? This is not cash. The 10-K identifies the material component: Amgen's equity investment in BeOne, carried at fair value with changes through earnings, which the MD&A says drove the 2025 increase in other income. It is a real asset and a real source of earnings volatility, but it is a strategic equity stake in a listed collaborator, not a liquidity buffer, and we do not net it against debt. As a disclosed sensitivity: deducting it in full would take enterprise value from $255.80B to $249.0B and EV/EBITDA from 15.3x to 14.9x. We do not use that version.

The rebuilt enterprise value:

> Market cap $210.497B (539.7M shares × $390.02)

> + total debt $57.30B (10-Q, 2026-03-31)

> − cash and equivalents $12.00B (10-Q, 2026-03-31)

> − short-term investments $0 (Amgen holds none — verified)

> = rebuilt enterprise value $255.80B

This is the rare case where the vendor's enterprise value is right and we say so. The bal_a block is stale by a quarter and misses the January-2026 note issue, but km_ttm has evidently been struck off the March-2026 balance sheet and lands within $15M of our hand-rebuilt figure. The EV tiles for AMGN can be used as printed; the annual balance sheet cannot.

What the corrected balance sheet actually says — and it is not comfortable.

• Net debt of $45.3B against trailing EBITDA of $16.674B is 2.72x. That is real leverage. Amgen has taken it down from $53.669B at the end of 2023, retiring $2.3B in 2023, $4.5B in 2024 and $6.0B in 2025, and then put $2.84B back on in a single quarter.
• Interest coverage is 4.36x. Trailing interest expense is $2.703B — 7.3% of revenue, and more than the entire 2025 dividend was in 2020.
• Book equity is $8.658B on $90.586B of assets. That is a 9.6% equity ratio and a 10.07x financial leverage ratio. Retained earnings is negative $25.107B — an accumulated deficit created by decades of buybacks and dividends, which the 10-K addresses directly: "Our accumulated deficit is not anticipated to affect our future ability to operate, repurchase stock, pay dividends or repay our debt given our expected continued profitability and strong financial position."
• Tangible asset value is NEGATIVE $30.863B and tangible book value per share is −$57.15, because $40.956B — 45.2% of total assets — is goodwill and intangibles, overwhelmingly from Horizon. priceToBookRatioTTM of 22.92x and returnOnEquityTTM of 89.4% are arithmetic artefacts of a small equity base and are rejected as lenses. The honest return measure is returnOnInvestedCapitalTTM of 13.85%, which is what we use.
• Liquidity is adequate but not lavish. $12.0B of cash, a $4.0B revolving credit facility extended in the March 2026 quarter by one year to March 2029 with no amounts outstanding, and a commercial paper programme increased in 2025 from $2.5B to $4.0B with nothing issued in 2025, 2024 or 2023. Ratings per the 10-K: S&P BBB+, Moody's Baa1, Fitch BBB+ — investment grade, low investment grade. The financial covenant is a minimum interest coverage ratio in the revolver and term loan; the 10-K states Amgen "was in compliance with all applicable covenants." The covenant threshold itself is not quantified in the extracted text and we do not guess at it.
• Working capital is heavy. Inventory $6.225B on 212.5 days of inventory outstanding, receivables at 89.6 days, payables at 100.8 days, for a 201.2-day cash conversion cycle. Current ratio 1.26x.

Two off-balance-sheet items in no vendor field, both from the 10-K (defect class 10):

• "As of December 31, 2025, we have purchase obligations of approximately $6.9 billion primarily related to (i) R&D commitments (including those related to clinical trials) for new and existing products, (ii) capital expenditures and (iii) open purchase orders... Most of these obligations are expected to be paid within one year."
• "As of December 31, 2025, the maximum amount that may be payable in the future for agreements we have entered into with third parties is approximately $7.2 billion" in development, regulatory and commercial milestones, of which only $161M of contingent consideration is recorded on the balance sheet.

And the largest contingent liability of all, which sits in no field on any screen — the IRS transfer-pricing case. From the 10-K, verbatim:

> "The Notices seek to increase our U.S. taxable income for the years 2010–2012 by an amount that would result in additional federal tax of approximately $3.6 billion plus interest. Any additional tax that could be imposed for the years 2010–2012 would be reduced by up to approximately $900 million of repatriation tax previously accrued and paid on our foreign earnings."

>

> "The Notice seeks to increase our U.S. taxable income for the years 2013–2015 by an amount that would result in additional federal tax of approximately $5.1 billion, plus interest, and asserts penalties of approximately $2.0 billion. Any additional tax that could be imposed for the years 2013–2015 would be reduced by up to approximately $2.2 billion of repatriation tax previously accrued and paid."

>

> "The two cases were consolidated in the U.S. Tax Court in 2022. The trial began on November 4, 2024 and concluded on January 17, 2025... The parties filed post-trial reply briefs on October 10, 2025. The Company expects a decision from the U.S. Tax Court no earlier than the second half of 2026."

>

> "We are currently under examination by the IRS for the years 2016–2018 with respect to issues similar to those for the 2010 through 2015 period. We expect that the IRS will begin its audit of 2019–2022 in the first half of 2026."

Arithmetic: $3.6B + $5.1B + $2.0B of penalties = $10.7B gross, less up to $3.1B of repatriation-tax offsets = roughly $7.6B net, PLUS interest accruing since 2010-2015, and BEFORE the 2016-2018 and 2019-2022 years which the company says raise "similar" issues. The 10-K's own closing sentence on the matter is the one to read twice: "the ultimate outcome of any tax matters may result in payments substantially greater than amounts accrued and could have a material adverse impact on our consolidated financial statements."

$10.7B of gross exposure is 5.1% of the market capitalisation, and the decision is expected within months. No vendor field contains it. It is a direct input to the risk score and to the verdict.

5. Cash flow and capital returns

Every one of the 2025 capital-return and financing figures is confirmed verbatim by the 10-K: "Cash flows from operating activities in 2025 totaled $10.0 billion... capital expenditures of $1.9 billion"; "For 2025, we retired $6.0 billion of debt and increased our quarterly cash dividend by 6% to $2.38 per share"; "During 2025 and 2023, we did not repurchase shares under the stock repurchase program"; "Cash used in financing activities during 2025 was primarily due to the repayment and extinguishment of debt of $5.0 billion and $683 million, respectively, and the payment of dividends of $5.1 billion." This is a clean vendor-to-filing reconciliation across the whole capital-return line and it deserves to be recorded as one.

Three things the cash-flow table says that the income statement does not.

First, free cash flow fell 22% in 2025 and the margin fell 910 basis points, from 31.1% to 22.0%. Two drivers: operating cash flow fell $1.532B, which the 10-K attributes to "the timing of working capital items primarily driven by higher collections in the fourth quarter of 2024" — a base effect, not a deterioration — and capital expenditure rose 69.5% to $1.858B for "construction costs for new plants and expansion of manufacturing capacity". And it is going higher: "We currently estimate 2026 investments in capital projects to be approximately $2.6 billion." That is another $742M of cash out, and on 2025's operating cash flow it would take free cash flow to roughly $7.4B.

Second, capital allocation has flipped from buybacks to deleveraging, and Amgen has told you it is deliberate. $6.360B of stock was repurchased in 2022. Since then: zero in 2023, $200M in 2024, zero in 2025, and zero in the March 2026 quarter. Authority remaining is $6.8B — 3.2% of market capitalisation — and it is going unused. Every discretionary dollar has gone to debt: $2.3B, $4.5B, $6.0B retired in 2023, 2024, 2025. That is the correct decision at 2.72x leverage. It also means the share count is not shrinking — diluted shares went 538M (2023) → 541M (2024) → 542M (2025) → 544M (Q1 2026), rising slightly on stock compensation of $494M a year.

Third, the dividend is the entire shareholder return and it is well covered. The 10-K gives the exact sequence: quarterly dividends of "$2.38, $2.25 and $2.13 per share of common stock paid in 2025, 2024 and 2023, respectively, reflecting year-over-year increases of 6% for both 2025 and 2024," and "In December 2025, the Board of Directors declared a cash dividend of $2.52 per share of common stock for the first quarter of 2026, an increase of 6% over the same period in the prior year."

A vendor labelling trap, identical in shape to the one we found on KLAC. profile.lastDividend reads 9.8 and ratios_ttm.dividendPerShareTTM reads 9.8, producing a yield of 2.51%. That $9.80 is the trailing-twelve-month total, not the last declared dividend. Four payments at $2.38, $2.38, $2.52, $2.52 = $9.80 exactly. The forward run-rate is $2.52 per quarter, $10.08 annualised — a 2.58% forward yield. Cover: $10.08 against $8.100B of free cash flow on 540M shares is $15.00 of free cash flow per share, 1.49x cover; against 2026 consensus non-GAAP EPS of $22.36 the payout is 45%.

Total shareholder yield is therefore 2.58% and nothing else — no buyback, no share shrinkage. In a company growing EPS at a consensus 4-5%, the dividend is roughly a third of the total expected return. That is the correct way to think about this equity and it is not the way a 52-week-high price is thinking about it.

6. Valuation — priced in or room?

At $390.02 (market cap $210.497B, ~539.7M shares, rebuilt enterprise value $255.80B):

Basis, stated explicitly, because for a pharmaceutical company this is where dives go wrong (defect class 7). est.epsAvg is a NON-GAAP series and the file nowhere says so. We tested it rather than assumed it: the 2025 row carries epsAvg of $21.318; Amgen's actual 2025 GAAP diluted EPS was $14.23; the sum of the four reported non-GAAP quarters from the earnings calendar is $21.85. The estimate row is within 2.5% of the non-GAAP actual and 50% above the GAAP actual. It is unambiguously non-GAAP. The 2024 row is the same story: epsAvg $19.605 against GAAP $7.56. Every forward multiple in this dive is therefore struck on NON-GAAP EPS, and the trailing multiple is shown on both bases — 17.65x non-GAAP, 27.14x GAAP.

The GAAP-to-non-GAAP bridge, quantified (defect class 3). The 2025 gap is $21.85 − $14.23 = $7.62 per share, or $4.13B after tax on 542M diluted shares — the non-GAAP number is 53.5% above GAAP. That is an enormous adjustment and it needs naming, not waving through:

A clean operating EPS, built by hand. Start from 2025 GAAP net income of $7.711B; add back the $1.2B Otezla impairment at the 14.1% effective tax rate (+$1.031B); remove all non-operating other income of $1.055B at the same rate (−$0.906B), because it is equity marks and debt-repurchase gains, neither of which is a business. That gives $7.836B, or $14.46 per diluted share — a clean GAAP operating EPS, on which the stock trades at 27.0x. Add back the roughly $4.3B pre-tax of acquisition intangible amortisation, after tax, and you land near the reported $21.85 non-GAAP figure, which cross-checks the whole bridge.

Which is the right basis? Genuinely, both — and the gap closes on its own. The 10-Q discloses the future amortisation schedule for finite-lived intangibles, and it is a gift to the reader:

> "As of March 31, 2026, the total estimated future amortization of our finite-lived intangible assets for the remaining nine months ending December 31, 2026, and the years ending December 31, 2027, 2028, 2029, 2030 and 2031, was $2.7 billion, $3.6 billion, $2.8 billion, $2.3 billion, $2.2 billion and $2.1 billion, respectively."

That is a declining schedule. Amortisation falls from roughly $3.6B in 2026 and 2027 to $2.1B by 2031 — a $1.5B pre-tax, roughly $1.3B after-tax, roughly $2.40-per-share mechanical improvement in GAAP EPS by 2031 with no operating improvement whatsoever. The GAAP and non-GAAP lines converge over time by construction. We use non-GAAP for the forward multiples because that is what the estimates are, we show GAAP alongside, and we note that the honest economic number sits between the two and drifts toward the higher one each year.

est.ebitdaAvg and est.ebitAvg — tested, not assumed, and NOT used. Per the data contract these rows are unreliable in a third of all files. Amgen's are not obviously broken — ebitAvg is below ebitdaAvg in every year, no negatives, revenue rises monotonically. But they are wrong in a consistent direction against actuals: 2025 ebitAvg of $11.407B is 6.8% above the vendor's reported 2025 operating income of $10.676B, and ebitdaAvg of $16.591B is 4.7% above operating income plus D&A of $15.843B. They are non-GAAP-basis rows, like epsAvg. All forward valuation runs on epsAvg and revenueAvg only.

Estimate coverage is good where it matters and thin where it does not. 24 analysts on 2026 revenue and 19 on EPS; 22 and 19 on 2027; 25 and 11 on 2028. The 2029 and 2030 EPS rows rest on FOUR analysts each and are excluded from every conclusion in this dive. The 2023 row is discarded outright (Section 8, finding 8).

Peer context. The vendor peer set is unusually sensible — Bristol-Myers Squibb, Gilead, Merck, Novo Nordisk, Novartis, Pfizer, Sanofi and GSK are all genuine large-cap pharmaceutical comparables; Boston Scientific and Danaher are not. No peer multiples are supplied in the file, so no peer-multiple comparison is drawn. What can be said: at $210.5B Amgen is mid-pack — larger than Gilead ($167.9B), Pfizer ($144.8B), Bristol ($134.6B), GSK ($103.2B) and Sanofi ($102.5B); smaller than Merck ($316.3B), Novartis ($291.4B) and Novo Nordisk ($196.6B is closest). Novo Nordisk at $196.6B is the relevant reference point for the MariTide question and it is not a comparison this file lets us make quantitatively.

6a. What today's price assumes (the inversion)

At $390.02 — 16.66x 2027 consensus non-GAAP EPS and 15.97x 2028 — the price embeds the following falsifiable claims, each with a number and a date:

• Revenue grows 2.7% in 2026 to $37.740B, 2.6% in 2027 to $38.736B and 2.7% in 2028 to $39.782B. (Consensus-derived; 24, 22 and 25 analysts.) That is $3.04B of added revenue in three years, against the $3.32B Amgen added in 2025 alone. The near-term check is tonight: consensus revenue of $9,426M for the June quarter. This is the least fragile assumption in the price — the analyst dispersion is narrow (2026 revenue low $37.320B, high $38.240B, a 2.5% band) and the volume engine is real.
• Non-GAAP EPS reaches $23.416 in 2027 and $24.418 in 2028 — growth of 4.7% then 4.3%. (Consensus-derived; 19 and 11 analysts.) The 2028 EPS row rests on ELEVEN analysts against 25 on the revenue row for the same year, and its low-to-high band is $19.679 to $26.861 — a 36% spread. That dispersion is the market telling you it does not know what happens after the Prolia/XGEVA cliff finishes.
• Prolia and XGEVA erode but do not collapse. (Our assumption on the shape; the 17% and 19% volume declines are filing-confirmed for the March 2026 quarter.) This is the most fragile assumption in the price. $6.498B of 2025 revenue is 17.7% of the company. If that base halves over three years — which is what "multiple biosimilars have launched" and "accelerated sales erosion" ordinarily produce — it removes $3.2B of revenue, which is more than the entire $3.04B of revenue growth consensus expects to add over the same period. The consensus revenue line is therefore an assumption that the growth brands out-run the cliff almost exactly. There is no margin in it either way.
• The IRS case does not go badly. (10-K-disclosed exposure of roughly $10.7B gross, $7.6B net of repatriation offsets, plus interest; decision expected in the second half of 2026.) Nothing in the $390.02 price appears to discount this. A full adverse ruling at the net figure is 3.6% of market capitalisation in cash, against a company already at 2.72x leverage — and the 10-K warns payments could be "substantially greater than amounts accrued."
• The market keeps paying roughly 16-17x forward non-GAAP earnings. (Our number.) At 14x 2027E the stock is $328; at 19x it is $445. For a company compounding EPS at 4-5%, the multiple is doing far more work in the outcome than the earnings are.
• MariTide is worth something but not yet much. (Our inference.) At 16.66x 2027E the stock is priced at roughly a market multiple for a low-growth pharmaceutical. We read that as the market assigning modest but non-zero value to six Phase 3 obesity studies — which is a defensible place to be, and is the main reason this is a Hold rather than an Avoid.

6b. The return bridge (why the multiple moves)

Expected return over the next twelve months decomposes as: EPS growth (+4.7%, from 2026E $22.356 to 2027E $23.416) + multiple drift (COMPRESSION, from 16.66x to 16.0x on the respective forward year, −4.0%) + shareholder yield (+2.58%, entirely dividend) ≈ +3.3%.

Our base case assumes mild multiple COMPRESSION and we should be explicit about why. Today the market pays 16.66x on the next calendar year's non-GAAP estimate. If in twelve months it pays 16.0x on the then-next year's estimate, that is a genuine de-rating of 4%, not the mechanical rolldown of the calendar. We assume it for two reasons. First, the loss-of-exclusivity profile gets worse before it gets better: Otezla enters Medicare price setting in 2027, the Colorado ENBREL upper payment limit bites no earlier than 1 January 2027, and Prolia/XGEVA erosion is explicitly guided to accelerate. A multiple appropriate to a company adding revenue is not appropriate to a company defending it. Second, the entry point is a 52-week high with RSI at 71.03 — the multiple has already expanded 20.4% over three months without an equivalent estimate revision.

The dominant term in the return is not earnings growth and it is not the multiple — it is the dividend. At a 4.7% EPS growth rate and a 2.58% yield, 44% of the expected total return is the dividend cheque. That is a materially different equity from the growth compounders elsewhere in this batch, and it should be underwritten as such: if the multiple simply HELD at 16.66x on 2027E the price would be $390 — exactly today's price, plus the 2.58% dividend; if it compressed to 15x the price is $351 (−10.0%); if it expanded to 18x the price is $421 (+8.1%). The entire spread between a good and a bad twelve months lives in three multiple points we do not control.

The bull case at $461 requires multiple EXPANSION and an earnings beat together — 19x on the 2027 consensus high of $24.250. The bear case at $287 is a de-rating to 13x on the 2027 consensus LOW of $22.100 — which is roughly what a market prices when it decides a company's revenue base is shrinking rather than plateauing.

6c. Variant perception (where we differ, what would surprise)

• We do not differ from the street on direction, and we are slightly MORE positive than the street on price — which is itself the finding. The street shows 22 buy, 13 hold and 3 sell across 38 analysts, headline consensus "Buy", and a target of $365.08 — 6.4% BELOW the closing price. Our base of $375 is 3.8% below spot but 2.7% ABOVE the street. When the analyst community carries a majority-buy rating stack and a target below the market price, the rating stack is stale and the target is current. We take the target seriously and the rating stack not at all. On a name where the knowledge base contributes nothing company-specific, having no differentiated view is a verdict input, and it points to Hold.
• We differ on the gross margin, and it is a real correction. Every screen built on this vendor file shows Amgen's 2025 gross margin at 70.8% and a 930 basis point expansion. The 10-K states cost of sales was 32.8% of total revenues, which is a 67.2% gross margin and a 570 basis point expansion. We use 67.2%. Anyone modelling forward margins off the 70.8% figure is starting 360 basis points too high on the single largest line below revenue.
• We differ on what the knowledge base is telling you about obesity, and this is the most useful thing in this dive's attribution work. A sweep on obesity returns 262 claims and GLP-1 returns 315; essentially all of them are about the CATEGORY or about competitors, not about Amgen. Only two entity-tagged claims touch MariTide, both from August 2025, both from a channel with no named speaker, and one of them is bearish: "MariTide's tolerability concerns eclipse compelling weight-loss data; trial design questions remain — more work to be done." The crowd's obesity enthusiasm is not evidence about Amgen. Meanwhile the claim we might have leaned on — that the obesity market is "shifting from a Lilly/Novo duopoly to a biotech game" with Amgen among the winners — comes from the same unnamed channel and is a category call. We report the discards rather than laundering category sentiment into name-level conviction.
• Positive surprise that would force a re-rate: a MariTide Phase 3 readout confirming the Phase 2 signal that patients maintain weight loss on quarterly dosing — quarterly administration in a weekly-injection category would be a genuine differentiator, and the company has said it, so it is half-weighted here until independent data lands. Or a favourable U.S. Tax Court decision, which would remove roughly $10.7B of gross overhang in a single day and is the cleanest binary in the file. Or tonight's print showing Prolia and XGEVA eroding more slowly than the March quarter's −17% and −19%.
• Negative surprise that would break the thesis: an adverse Tax Court decision with a cash payment at or near the $7.6B net figure; a June-quarter print showing product sales growth below the March quarter's +4%; a quantified financial impact from the 2026-07-31 cybersecurity incident, where the company's own language — "the investigation remains ongoing" and "continues to assess whether, and/or the extent to which, patient, confidential business information, intellectual property, research and development, or other information may have been accessed" — leaves the door open; or a MariTide Phase 3 tolerability signal consistent with the one the knowledge base's only bearish claim flagged.

Synthos fair values

All three anchors are multiples of the 2027 consensus non-GAAP EPS distribution (mean $23.416, low $22.100, high $24.250, 19 analysts — the deepest EPS coverage in the estimate block), cross-checked against 2026E of $22.356 and 2028E of $24.418.

• Bear ~$287 — 13x the 2027 consensus LOW of $22.100. Cross-check: 12.8x 2026E; 11.8x 2028E mean; 1.65x the 200-day moving average of $344.72 taken 17% lower. The scenario: Prolia and XGEVA erode faster than the −17%/−19% already printed, Otezla price setting bites in 2027, the Tax Court rules against Amgen for a cash payment near $7.6B, and the market re-rates a shrinking revenue base to a low-teens multiple. −26.4%.
• Base ~$375 — 16x the 2027 consensus MEAN of $23.416. Cross-check: 16.8x 2026E; rebuilt EV of $255.8B on 2027E revenue of $38.736B is 6.6x sales; 15.4x 2028E. Sensitivity, stated openly: 15x gives $351 (−10.0% from spot), 17x gives $398 (+2.1%), and 18x gives $421 (+8.1%). The entire answer lives in a 15-18x band and we claim no more precision than that. The scenario: consensus lands roughly as modelled, the growth brands offset the cliff, MariTide stays an option rather than a franchise, the tax case settles or is delayed, and the multiple drifts down a point on decelerating revenue. −3.8% against spot, plus 2.58% of dividend.
• Bull ~$461 — 19x the 2027 consensus HIGH of $24.250. Cross-check: 20.6x 2026E; 18.9x 2028E mean. This bull case requires BOTH an earnings beat and a three-point multiple expansion, which is what a positive MariTide Phase 3 readout plus a favourable Tax Court decision would plausibly deliver together. +18.1%.

Base is 3.8% BELOW spot; asymmetry is roughly 0.69:1 AGAINST (26.4% down, 18.1% up). That ratio is the verdict. A stock whose base case sits under its price, whose street target sits further under its price, whose downside exceeds its upside, and which reports earnings tonight at RSI 71, is not an entry. It is also not a sell: the business is sound, the dividend is covered, the balance sheet is serviceable and the pipeline optionality is real. That combination is exactly what Hold means.

7. Knowledge base — sixteen hits, zero named speakers, and an obesity lane that is not about Amgen

Raw hits: 16. Entity matches: 13. Text-only matches: 3. Used as name-level colour: 4. Used as name-level conviction: 0. Discarded: 12.

The primary search covered AMGN and Amgen across all 51,928 distilled claims and returned 13 entity matches and 3 text-only matches. Supplementary sweeps on MariTide (0 entity, 2 text — both already in the entity set), olpasiran (0 entity, 1 text — already in the set), Repatha, Prolia, TEZSPIRE, Otezla, ENBREL, KRYSTEXXA, TEPEZZA and Robert Bradway returned ZERO hits each — not one claim in 51,928 mentions any Amgen brand name other than through the MariTide and olpasiran claims already counted. A broad category sweep on obesity (1 entity, 261 text), GLP-1 (7 entity, 308 text), biosimilar (0 entity, 4 text), large pharma (0 entity, 9 text) and drug pricing (0 entity, 98 text) returned several hundred additional matches, all discarded — they are claims about the therapeutic category, about competitors, or about US pricing policy generally, and a claim about a company's market is not a claim about that company's stock.

The most important attribution fact in this dive, stated first because it colours everything else.

> NOT ONE of the sixteen claims in the AMGN lane carries a named speaker. Every single record has speaker: null and is attributed only to a channel slug.

Under the 4-lane attribution policy that is the weakest form of sourcing we accept. A channel-level attribution tells you where a claim was said; it does not tell you who said it, whether they are independent, or what their record is. On a name where sixteen out of sixteen claims fail that test, the correct conviction rating is Low and the correct amount to move the fair value is zero.

Channel concentration, and the management problem.

Eight of thirteen entity claims — 61.5% — come from one channel with no named speaker. That is the same structural failure the LLY dive found, in a different shape: there the lane was 68% the company's own chief executive; here it is 62% a single unnamed channel plus 31% a competitor's chief executive. In neither case is there an independent, named, bearish voice on the equity.

The four claims we actually use, verbatim.

> 2025-08-21 · bearish · conviction 50 · horizon: fact · entities: Amgen · channel: biotech_hangout · NO NAMED SPEAKER

> "MariTide's tolerability concerns eclipse compelling weight-loss data; trial design questions remain—more work to be done."

> 2025-08-26 · bullish · conviction 70 · horizon: thesis · entities: Eli Lilly, Novo Nordisk, Amgen, Metsera, Kailera · channel: biotech_hangout · NO NAMED SPEAKER

> "Obesity shifting from Lilly/Novo duopoly to a biotech game; Amgen, Metsera, Kailera and oral peptides offer differentiated tolerability/dosing upside."

> 2025-08-28 · neutral · conviction 50 · horizon: fact · entities: Amgen · channel: biotech_hangout · NO NAMED SPEAKER

> "Amgen biosimilar (Bekemv) a tailwind, but MariTide obesity data won't clarify until mid-year ADA—stock likely under pressure near-term."

> 2026-01-16 · bullish · conviction 68 · horizon: thesis · entities: Eli Lilly, AbbVie, Merck, Bristol Myers Squibb, Pfizer, Gilead, Amgen · channel: biotech_hangout · NO NAMED SPEAKER

> "Generalists are rotating into pharma/biotech because it looks cheap versus an expensive AI-driven S&P; interest is broadening beyond mega-caps to Merck, Bristol, Pfizer, Gilead, Amgen."

And the one claim quoted separately at HALF WEIGHT, because the channel identifies it as management voice at a competitor:

> 2025-12-05 · bullish · conviction 70 · horizon: thesis · entities: Amgen, Boehringer Ingelheim, AstraZeneca, Eli Lilly, Novo Nordisk · channel: eli_lilly_ceo — COMPETITOR MANAGEMENT, HALF WEIGHT

> "A broad wave of obesity drugs is in development across major players, expanding the treatment toolbox beyond the current two dominant agents."

What the used claims actually say, and what they are worth. The 2025-08-21 and 2025-08-28 claims are the only two in the entire knowledge base that are about Amgen alone and about the thing that matters — and one is bearish on MariTide tolerability while the other calls the stock "likely under pressure near-term" pending obesity data. They are eleven months old and they were right for the period they covered. The 2025-08-26 claim is a category call in which Amgen is one of five names. The 2026-01-16 claim is a sector-rotation call in which Amgen is one of seven names, and it is the freshest claim in the lane at more than six months old — it is also, notably, the only claim that is even indirectly about valuation, and it argues large-cap pharma was cheap in January. The stock is 31.4% higher over twelve months. That claim has largely played out.

The eleven discards and why.

Conclusion, stated as a finding rather than dressed up. The Synthos knowledge base has effectively no current view on Amgen Inc. It has two eleven-month-old, unnamed-speaker claims on MariTide, one of them bearish; a category view that obesity is broadening beyond the incumbent duopoly; and a six-month-old sector-rotation call that has already worked. The single largest theme in the knowledge base that touches this company — obesity, with 262 claims, and GLP-1, with 315 — is almost entirely about the category and about two competitors, and triaging those out was the most useful analytical act in this section. No claim in the lane carries a named speaker. One channel supplies 62% of it. A competitor's chief executive supplies 31%. Conviction: Low. Breadth: 3. Net conviction: neutral-low. This thinness is a direct input to the Hold verdict — a name with no independent conviction and no valuation cushion defaults to Hold or Watch, never to Buy.

8. Data integrity — what we rejected from the vendor file and why

Fourteen findings. Listed rather than silently corrected. Amgen's file is middling: the debt, capital-return and revenue-split figures reconcile to the filings exactly, and the enterprise-value block is right — but the FY2025 gross margin is materially wrong, the product table has a whole product missing, and the annual and quarterly income statements do not tie to each other.

1. inc_a FY2025 cost of revenue understates the 10-K's own disclosed figure by $1.322B — REJECTED, filing used. The vendor reports FY2025 cost of revenue of $10.729B, which is 29.2% of revenue and a 70.8% gross margin. The 10-K states verbatim: "Cost of sales decreased to 32.8% of total revenues for 2025." 32.8% × $36.741B = $12.051B, a gross margin of 67.2%. The FY2024 figure reconciles exactly ($12.858B / $33.424B = 38.47% against the filing's "increased to 38.5%"), so this is the same field computed on different bases in adjacent years — confirmed defect class 4 — manufacturing roughly 360 basis points of gross-margin expansion that did not occur. The vendor reports a 70.8% gross margin; the 10-K implies 67.2%; we use 67.2%. Because the numeric statements are stripped from the extracted text we cannot see where the $1.322B lands, and the vendor's operating income of $10.676B ties arithmetically to the verified pre-tax, tax, net-income and EPS chain — so we flag the gross-profit/operating-expense split as unreliable and run every valuation conclusion off EPS and revenue instead.

2. inc_q Q4-2025 gross margin of 81.9% is an outlier and is REJECTED. Cost of revenue of $1.790B on $9.896B of revenue implies an 81.9% gross margin and a 39.8% operating margin — against 67.2%, 67.8% and 68.2% in the surrounding quarters and 61.5% for FY2024. It is the same $1.3B-scale error as finding 1, landing in the fourth quarter because that quarter is derived as the full year minus nine months. No Q4-2025 margin figure is used in this dive. The Q1-2026 quarter, by contrast, is filing-verified: the 10-Q states cost of sales was "31.8% of total revenues" and the vendor's $2.744B / $8.618B is 31.84% — an exact match.

3. bal_a[0] is one quarter stale and misses a $4.0B note issuance — REJECTED and rebuilt from the 10-Q. The latest annual balance sheet is 2025-12-31. The 10-Q filed 2026-05-01 discloses $4.0B of debt issued in the March 2026 quarter — $1.0B of 4.20% 2031 Notes, $1.75B of 4.85% 2036 Notes, $500M of 5.50% 2046 Notes and $750M of 5.65% 2056 Notes — against €750M ($833M) of 2026 euro Notes repaid and $324M of principal repurchased for $233M at a $90M gain. The 10-Q's fair-value note confirms fixed-rate debt carrying value rising from $52.8B to $55.5B. The vendor reports total debt of $54.604B; the 10-Q shows approximately $57.3B including the $1.8B term loan carried forward; we use $57.3B. (The pre-flight hypothesis for this dive was that the 8-Ks of 2026-05-19 and 2026-05-20 would be a note-offering pricing/closing pair. They are not — see finding 12. The financing is disclosed in the 10-Q, not in an 8-K, and we say so rather than leaving a wrong prior standing.)

4. seg_prod FY2025 is MISSING AN ENTIRE PRODUCT and sums $1.402B short — corrected by derivation. The FY2024 table sums to $33,424M, matching total revenues exactly. The FY2023 table sums to $28,190M, matching exactly. The FY2025 table sums to $35,339M against total revenues of $36,741M — short by $1,402M — and KYPROLIS, which appears in every other year ($1,503M in 2024, $1,403M in 2023), is absent from the FY2025 line-up entirely. The 10-K lists KYPROLIS among Amgen's principal products and the 10-Q discusses KYPROLIS sales for the March 2026 quarter, so the product exists and is still sold. We therefore derive KYPROLIS 2025 revenue as the residual, ≈$1.402B, label it as derived, and note it is consistent with the $1.503B and $1.403B of the two prior years. Also note the vendor mislabels the table type: Amgen states in the 10-K that it "operates in one operating segment: human therapeutics" — there are NO reportable segments, and seg_prod is the MD&A worldwide-product-sales table, not a segment table.

5. seg_prod correctly INCLUDES other revenues — verified, and this one is clean. The line the vendor calls "Product and Service, Other" is Amgen's Other revenues (royalty income and corporate partner revenue). At $1,603M for 2025 against $1,398M for 2024 it is +14.7%, matching the 10-K's "Other revenues increased 15% for 2025". Subtracting it gives product sales of $35,138M against 2024's $32,026M = +9.7%, matching the 10-K's "Total product sales increased 10% in 2025." Both legs of the split independently reconcile to the filing's own stated growth rates. Recorded as a verified-clean check.

6. The FY2025 annual income statement does not tie to the sum of its own quarters — flagged, not silently smoothed. Summing the four 2025 quarters against the annual row: revenue $36,769M vs $36,741M (+$28M), cost of revenue $10,851M vs $10,729M (−$122M), R&D $7,281M vs $6,974M (−$307M), SG&A $7,035M vs $7,050M, other $1,312M vs $1,312M (exact), operating income $10,290M vs $10,676M (−$386M). The two series are not built on the same basis. We use the QUARTERLY series for run-rate and margin work — it is the one that reconciles to the 10-Q's stated percentages — and the annual series for the multi-year history, and we say which is which wherever it matters.

7. est.epsAvg is NON-GAAP and the file nowhere says so — tested, confirmed, and labelled throughout. The 2025 row's epsAvg of $21.318 sits within 2.5% of the sum of the four reported non-GAAP quarters ($21.85) and 50% above the GAAP diluted actual of $14.23. The 2024 row: epsAvg $19.605 against GAAP $7.56. Confirmed non-GAAP. Every forward multiple in this dive is struck on non-GAAP EPS and labelled as such, with the trailing multiple shown on both bases — 17.65x non-GAAP, 27.14x GAAP. This matters more for a pharmaceutical company than for almost any other sector, because the GAAP-to-non-GAAP gap here is 53.5%.

8. The FY2023 estimate row is internally inconsistent with BOTH bases and is DISCARDED in full. It reports epsAvg of 5.83411 and niAvg of $3.130B against a company whose 2023 GAAP diluted EPS was $12.49 on net income of $6.717B, and whose 2023 non-GAAP EPS was higher still. The row is wrong by any measure — roughly 47% of the GAAP actual — and no conclusion in this dive touches it. The 2024 and 2025 rows were verified against actuals and are sound on a non-GAAP basis.

9. est.ebitdaAvg and est.ebitAvg are non-GAAP-basis and NOT USED, per the data contract. Tested rather than assumed: no negative values, ebitAvg below ebitdaAvg in every year, revenue rising monotonically — none of the three published fabrication signatures fires. But against actuals they run consistently high: 2025 ebitAvg of $11.407B is 6.8% above reported operating income of $10.676B, and ebitdaAvg of $16.591B is 4.7% above operating income plus D&A. All forward valuation runs on epsAvg and revenueAvg only. Analyst counts are reported per year and the 2029 and 2030 EPS rows (4 analysts each) are excluded from every conclusion; the 2028 EPS row rests on 11 against 25 on the same year's revenue, and its $19.679–$26.861 range is disclosed rather than averaged away.

10. km_ttm enterprise value is CORRECT and we say so — a rare clean result on this field. The vendor's implied net debt of $45.285B and enterprise value of $255.782B land within $15M of our hand-rebuilt figures ($45.30B and $255.80B) using the 10-Q's $57.3B of debt and $12.0B of cash. The recurring short-term-investments omission does not apply here because Amgen holds none — verified against the 10-Q, whose entire liquidity statement is "Our balance of cash and cash equivalents was $12.0 billion as of March 31, 2026," with no marketable-securities clause. shortTermInvestments: 0 is a true zero, not a null. Verified clean and reported as a finding. (Note that bal_a is nonetheless stale — finding 3 — so the two blocks disagree with each other by a quarter.)

11. Zeros and nulls hiding real data — scanned, and four are genuine defects. cf_a reports incomeTaxesPaid: 0 and interestPaid: 0 in all four years, which is impossible for a company with a $2.7B interest expense and a $1.265B tax charge — these are missing, not zero, and no cash-tax or cash-interest figure appears in this dive. effectOfForexChangesOnCash is 0 in all four years for a company with 28.1% non-US revenue and euro- and sterling-denominated notes. generalAndAdministrativeExpenses and sellingAndMarketingExpenses are $0 in FY2025, FY2023 and FY2020 but populated in FY2024, FY2022 and FY2021 — the SG&A split is unusable and only the combined line is used. capitalLeaseObligations is $0 in all three years; the 10-K has a full leases note and refers to "rental commitments on unoccupied leases," so there are leases, though it also states "As of December 31, 2025, there were no future lease payments for leases that have not yet commenced" — no uncommenced-lease overhang, which is the check that mattered. One apparent zero is CORRECT and worth recording: commonStockRepurchased: 0 for FY2025 matches the 10-K exactly — "During 2025 and 2023, we did not repurchase shares under the stock repurchase program."

12. The 8-K archive contains NO financing document — the pre-flight hypothesis was wrong and here is what the filings actually are. The pair of 8-Ks filed one day apart in May was expected to be a note-offering pricing-and-closing sequence. It is not.

• 8-K filed 2026-05-19 (event 2026-05-19) — Item 5.02, an executive transition. "Peter H. Griffith will retire from the role of the Company's Executive Vice President and Chief Financial Officer, effective August 31, 2026," remaining an EVP until 31 January 2027. Thomas Dittrich, most recently CFO of Galderma Group and previously CFO of Shire Pharmaceuticals and Sulzer AG — and an Amgen finance executive for eight years to 2014 — becomes Executive Vice President and Chief Financial Officer effective 1 September 2026, initially based in Rotkreuz, Switzerland, relocating to Thousand Oaks. Compensation is fully disclosed: CHF 1,070,000 base, 100% target bonus, CHF 4,500,000 of annual long-term incentives, plus CHF 4,700,000 of make-whole RSUs, a CHF 4,000,000 one-time cash bonus and a CHF 5,800,000 two-year retention bonus. A CFO change at a company with $57.3B of debt and an open Tax Court case is a governance fact worth flagging, and it is in no vendor field.
• 8-K filed 2026-05-20 (event 2026-05-19) — Item 5.07, annual meeting results. Twelve directors elected; say-on-pay approved; Ernst & Young ratified; a stockholder proposal to require an independent board chairman was NOT approved — Robert Bradway remains both Chairman and CEO. (This 8-K explains the eight director share awards in the insider file, all dated 2026-05-19 — see Section 10.)
• 8-K filed 2026-07-31 (event 2026-07-29) — Item 1.05, MATERIAL CYBERSECURITY INCIDENT. See finding 13.

Neither May 8-K is a financing. Neither is a bare cover page — both carry full Item text. The $4.0B January-2026 note issue is disclosed only in the 10-Q, and the vendor payload does not reflect it.

13. A MATERIAL cybersecurity incident declared six days before this dive appears in NO vendor field. The 8-K filed 2026-07-31 for an event date of 2026-07-29 states, verbatim:

> "In July 2026, Amgen Inc. identified unauthorized activity involving data stored in cloud environments hosted by third-party cloud service providers... The Company has since learned that some of its data, including proprietary data, patient protected health information, and other information, has been exfiltrated from these cloud environments. To date, the Company has not identified any impact to its products, manufacturing operations, or financial reporting systems... On July 29, 2026, in connection with our evaluation of the volume of the files that appear to have been impacted and the potential that the types of information in such files could be sensitive, the Company determined that this incident is material. The Company believes, as of the date of this Current Report on Form 8-K, that the incident is not reasonably likely to have a material impact on the Company's financial condition or results of operations... The investigation remains ongoing."

Note the precise construction: the INCIDENT is determined material; the FINANCIAL impact is not currently believed likely to be material; the investigation is not finished. The company also states it "continues to evaluate applicable regulatory and legal notification requirements and will make all required notifications based on its findings, including to impacted patients." Patient-notification obligations under health-privacy law are a cost and a litigation surface. No quantification exists in any source available to this dive and none is invented here. It is disclosed as a live, dated, unquantified risk and it is an input to the verdict because it landed six days ago.

14. quote.yearHigh/yearLow disagree with tech.hi52/lo52 — we use tech and say so. The quote block reports a 52-week high of $398.00 and low of $269.77; the computed technical block reports $393.10 and $271.18. The discrepancies are 1.2% and 0.5%. We use the tech figures throughout, because they are the 252-day closing extremes computed from the same six-year series that produces the moving averages, drawdown and relative-return figures. (tech.max_dd_from_peak of −0.78% is identical to pct_from_hi — that is not a defect, it correctly reflects a stock sitting at a multi-year peak.)

Not defects, correctly reported and independently confirmed: total debt at 2025-12-31 ($52.8B of notes plus $1.8B term loan = $54.6B, matching the vendor's $54.604B to the decimal); the entire capital-return series (dividends of $5.124B, zero buyback, $6.0B of debt retired, all verbatim in the 10-K); the revenue split into product sales and other revenues, both matching the filing's stated growth rates; the geographic split (2024, 2023 and 2022 sum exactly; 2025 is $10M over on $36.741B); researchAndDevelopementToRevenueTTM of 20.18% reconciling to $7.514B on $37.238B; interestCoverageRatioTTM of 4.357x; effectiveTaxRateTTM of 14.16% against the 10-Q's 12.7% for the March quarter; enterpriseValueTTM; and commonStockRepurchased: 0 for 2025.

Currency check (defect class 5) — CLEAN. inc_a[0].reportedCurrency is USD and profile.currency is USD. Amgen reports in the currency of its quote. No multiple in this dive requires a translation. Stated because the check is mandatory even when it passes.

Share-count check (defect class 6) — CLEAN. Market capitalisation of $210,496,914,160 divided by the $390.02 price implies 539,707,231 shares. The 10-K cover page states 539,067,675 shares outstanding as of 2026-02-10 — agreement to 0.12%, entirely explained by four months of equity-plan issuance. Cross-check from the income statement: Q1 2026 net income of $1,819M divided by basic EPS of $3.37 gives 539.8M basic shares; diluted is 544M. Three independent counts within 0.9% of each other. Single share class, no splits, no dual-class structure.

Non-equity tripwire — checked and passed. AMGN is common stock, NASDAQ Global Select, CUSIP 031162100, ISIN US0311621009, IPO 1983-06-17. The $390.02 price is not par-like. The 52-week band of $271.18–$393.10 is a 45% range — far too wide for a fixed-income-like instrument. The dividend is variable and has been raised 6% in each of the last three years, not fixed. Volume was 3.337M shares, roughly $1.3B of turnover. The one field that superficially resembles a warning sign is beta at 0.413 — but that is a genuine characteristic of defensive large-cap pharmaceuticals, not evidence of a non-equity instrument, and every other test points the other way. This is common equity.

9. Technicals

• Price $390.02. −0.78% from the 52-week high of $393.10, +43.8% above the 52-week low of $271.18. Position within the annual range: 97.5th percentile — effectively at the top.
• Moving averages, both above and both rising: +9.02% above a 50-day average of $357.74; +13.14% above a 200-day average of $344.72. The trend is intact and unambiguous in both timeframes. There is no technical objection to this chart — the objection is that there is nothing left in it.
• RSI 71.03 — OVERBOUGHT. The conventional threshold is 70. MACD +8.37 — positive and expanding. Maximum drawdown from peak over the trailing year: −0.78%. The stock has essentially not had a drawdown.
• Relative performance: 3-month +20.43% vs SPY +7.59% and QQQ +7.67% — a 12.8-point outperformance in a quarter; 6-month +13.15% vs SPY +11.09% and QQQ +15.71% — underperforming the Nasdaq; 12-month +31.37% vs SPY +24.26% and QQQ +30.80% — a 7-point edge on the S&P and essentially level with the Nasdaq. Read those three rows together and the entire twelve-month outperformance was earned in the last three months.
• Beta 0.413 — the defining characteristic of this equity as a portfolio object. It is a low-volatility, dividend-paying defensive.
• Sentiment: 22 buy, 13 hold, 3 sell across 38 analysts, headline consensus "Buy" — attached to a consensus target of $365.08, which is 6.4% BELOW spot, a median of $360 (−7.7%), a high of $427 (+9.5%) and a low of $303 (−22.3%). The high target is only 9.5% above the current price. There is no analyst in this file with a target implying more than a 10% gain.

Today's move and what it does to the entry

AMGN closed 2026-08-04 at $390.02, up 2.94% or $11.15 from a previous close of $378.87. It opened at $380.27 — a 0.37% gap — traded a $376.02 to $390.95 range, and closed within 0.24% of the day's high on 3.337M shares, 1.24x the 2.689M average. Closing at the high of the day on above-average volume is a strong close, and it took the stock to within 0.8% of its 52-week closing high.

No company-specific news in this file explains it. The last dated company event before today was the 2026-07-31 cybersecurity 8-K, which is negative. The market context for 2026-08-04 was a violent semiconductor and memory rally — MRVL +12.8%, INTC +10.9%, LRCX +7.9%, MU +7.6%, KLAC +7.0%, AMD +7.0% — which is not a healthcare event. The most likely explanation for a 2.94% move in a 0.413-beta pharmaceutical on the afternoon of its own earnings print is positioning into that print. We label it as such and do not pretend to know.

The honest read, and it is the opposite of the KLAC setup in this batch. There, a 6.95% gap came from 35% below the 52-week high with RSI at 36.5 — a bounce inside a drawdown, where the gap was not the objection. Here a 2.94% move takes an already-extended stock to a 52-week high with RSI at 71.03, 9% above its 50-day average, after a 20.4% three-month run, on the day of an unseen earnings print, with the street's own target 6.4% below the price. Every one of those facts pushes the same way.

So the entry objection is not one thing, it is four things stacked:

1. You are paying the 52-week high. The stock is at the 97.5th percentile of its annual range with a 0.78% maximum drawdown from peak.

2. You are paying it overbought. RSI 71.03.

3. You are paying it before the number. Q2 2026 lands tonight and consensus asks for a 6.6% year-on-year EPS decline.

4. You are paying above both our number and the crowd's. Our base is $375; the street's is $365.08.

What would change it, concretely, and these are the falsifiers:

• A price near $351 — 15x the 2027 consensus mean, roughly the 50-day average of $357.74 with a little overshoot — which would put our base $375 at a +6.8% premium and restore a positive expected return before the dividend. That is 10% below today's close and well inside the range this stock traverses in a normal quarter.
• Or tonight's print beating $9,426M of revenue with Prolia and XGEVA eroding more slowly than the March quarter's −17% and −19% volume declines, which would attack the most fragile assumption in 6a directly. At that point paying up is defensible, because the uncertainty being paid for has been resolved.
• Or a favourable U.S. Tax Court decision, which removes roughly $10.7B of gross overhang in a single day.

Buying today means paying a 52-week-high price for a negative base-case return, hours before a print that consensus expects to show earnings going backwards. That is not a trade with an edge in it.

10. Insiders — eight transactions, all the same thing, and none of them a purchase

Every insider transaction in this file was filed on 2026-05-21 for transactions dated 2026-05-19 — the date of the annual meeting — and every one is a director award at a stated price of $0.

Total: 5,320 shares, awarded, not bought.

The reading, stated carefully. Every transaction is coded A-Award at a price of zero — these are the annual non-employee director equity grants made at the annual meeting, confirmed by the 8-K of 2026-05-20 recording that meeting on 2026-05-19. The uniform 665-share size across all eight directors is the signature of a fixed-value annual retainer grant, not a discretionary decision by any individual.

What the file contains is therefore: zero open-market purchases, zero open-market sales, and — notably — zero transactions of any kind by any EXECUTIVE officer. No chief executive, chief financial officer or divisional president appears in the last eight transactions. That is an information vacuum, not a signal, and it is reported as neutral.

Two contextual notes worth more than the table. First, the directors' holdings are modest in dollar terms — the largest, Robert Eckert's 26,251 shares, is $10.2M at the closing price, and the smallest, Mary Klotman's 2,211 shares, is $862,000. These are board-level stakes, not founder stakes. Second, and more important, the 8-K of 2026-05-19 discloses that the chief financial officer is retiring effective 31 August 2026 — a change of the officer who signs the financial statements, at a company with $57.3B of debt and an unresolved Tax Court case, three months after the last balance sheet in this file. The insider table does not capture that; the 8-K does.

11. Verdict, kill-criteria and flip conditions

Hold.

This is an entry rejection, not a thesis rejection, and the distinction matters.

What is genuinely good, and none of it is in dispute: $36.741B of 2025 revenue, up 9.9%, on 13% volume growth; $9.958B of operating cash flow and $8.100B of free cash flow; a product portfolio of fifteen-plus brands with nothing above 12% of revenue; six growth brands compounding at 13-52%; a 2.58% forward dividend raised 6% a year and covered 1.49x by free cash flow; $12.8B of debt retired across 2023-2025; investment-grade ratings of BBB+/Baa1/BBB+; a $4.0B undrawn revolver to March 2029; a 13.85% return on invested capital; an intangible-amortisation schedule that mechanically improves GAAP earnings by roughly $2.40 a share by 2031; and six global Phase 3 studies of MariTide in the largest therapeutic category in the industry, with a Phase 2 second-year signal on monthly and quarterly dosing.

What we are declining to pay for: a 52-week-high price at the 97.5th percentile of the annual range with RSI at 71.03; a base fair value of $375 that is 3.8% BELOW spot and a street target of $365.08 that is 6.4% below; consensus revenue growth of 2.6-2.7% a year and EPS growth of 4-5%; 17.7% of revenue in active biosimilar erosion with the company itself guiding to "accelerated sales erosion" and the March quarter already showing −17% and −19% volume; $45.3B of net debt at 2.72x EBITDA against negative $30.9B of tangible asset value; roughly $10.7B of gross IRS exposure with a Tax Court decision expected within months; a material cybersecurity incident declared six days ago with an open investigation; and an earnings print tonight that consensus expects to show a 6.6% year-on-year EPS decline.

The distinction that matters. We are not saying Amgen is a bad company or that MariTide will fail. We are saying that at $390.02 our own arithmetic and the street's both land below the price, the downside case is larger than the upside case (26.4% versus 18.1%), and the number that could resolve the uncertainty lands after this dive is struck. In the Synthos frame, a name where our fair value sits below the market price, where the knowledge base contributes no named-speaker conviction, and where the next catalyst is hours away and unseen, is a Hold by construction, not by hesitation. Manufacturing a differentiated view here would be dishonest, and buying into a print we cannot see would be worse.

Pre-registered KILL criteria — what would take this to Avoid:

• An adverse U.S. Tax Court decision with a cash liability approaching the $7.6B net exposure, or any indication the 2016-2018 and 2019-2022 examinations will pursue the same theory at similar scale. At 2.72x leverage this is a balance-sheet event, not an earnings event.
• Prolia plus XGEVA revenue falling below roughly $4.0B annualised — a 38% decline from the $6.498B 2025 base. The March quarter's −17% and −19% volume declines put that inside two years on a straight-line extrapolation.
• Total product sales growth turning NEGATIVE in any quarter. It was +4% in March 2026 against +10% for full-year 2025; one more step down of that size and the cliff is winning.
• A quantified, material financial impact from the cybersecurity incident, or a disclosure that intellectual property or clinical data specifically was exfiltrated — the 2026-07-31 8-K explicitly leaves both open.
• A MariTide Phase 3 tolerability failure, which would validate the knowledge base's only bearish name-level claim and remove the entire growth option from the story.
• A dividend increase below 6%, or any suspension of the annual raise. At a 45% payout on 2026E non-GAAP EPS there is room, so a break in the pattern would be a signal about management's view of the cliff, not about affordability.

Pre-registered FLIP conditions — what would take this to Buy — Tactical:

• Price below approximately $351 (15x 2027E consensus, roughly the 50-day average) with no deterioration in the fundamental picture. At $351 the base case is +6.8% plus a 2.87% dividend yield, and the risk/reward turns positive. This is the single most likely path to a Buy and it is only 10% below today's close.
• A favourable or materially narrowed U.S. Tax Court outcome, removing the largest unquantified liability in the file. This is the cleanest binary available and it is expected in the second half of 2026.
• A MariTide Phase 3 readout confirming the Phase 2 maintenance signal, particularly on quarterly dosing — the one thing in this company that could re-rate the multiple rather than the earnings.
• Tonight's print beating on revenue with a Prolia/XGEVA erosion rate BELOW the March quarter's, which would directly attack the most fragile assumption in the price and make paying 17x defensible.
• Genuine name-level knowledge-base conviction appearing — from a NAMED, independent speaker. At present there is not a single named speaker in the entire sixteen-claim lane, and that alone is worth roughly one tier.

Where AMGN fits in the Synthos Framework Portfolio. The large-cap pharmaceutical / defensive-healthcare sleeve, at 0% today with a 1.5-2% target on a fill near $351 or a resolved Tax Court decision. On the portfolio-construction question: Amgen's beta of 0.413 and 2.58% dividend make it a genuine defensive ballast rather than a growth position, and it should be underwritten that way — as a bond-like holding with a free option on obesity attached, not as a biotechnology growth name. An investor who already owns it has no reason to sell into tonight's print; an investor who does not owns nothing and has given up 3.8% of theoretical value by waiting. That asymmetry is exactly what Hold describes. Logged as a tracked Synthos call (Hold) as of 2026-08-04 at $390.02.

Single biggest risk: the loss-of-exclusivity cliff is already running and consensus assumes it stops. Prolia and XGEVA were $6.498B, 17.7% of 2025 revenue; their RANKL patents expired in February 2025 in the United States and November 2025 in parts of Europe; multiple biosimilars have launched; and in the first full quarter of competition Prolia volume fell 17% with price down 10%, and XGEVA volume fell 19%. The company's own guidance, repeated in both the 10-K and the 10-Q, is "accelerated sales erosion." Consensus wants total revenue up $3.04B over three years. If the Prolia/XGEVA base halves over that period it removes $3.2B — more than the entire expected gain. The consensus line is therefore a bet that the growth brands out-run the cliff almost exactly, with no margin in either direction, and the price offers no cushion for being wrong: at 16.66x forward earnings there is nothing below the stock but the dividend.

Most fragile assumption in the price: that the multiple holds at 16-17x while revenue growth runs at 2.6%. Every other assumption in 6a is a volume or price assumption and both are observable quarter by quarter. This one is a market-behaviour assumption, and the market has just paid a 20.4% three-month premium for it without an equivalent estimate revision. Consensus non-GAAP EPS for 2026 rose from a 2025 actual of $21.85 to $22.356 — a 2.3% increase — while the stock rose 20.4% in three months. That entire move was multiple, not earnings. If the multiple gives back three points to 13.7x — which is where a market prices a business it believes is shrinking — the price is $321, 18% below today, with the earnings entirely unchanged.

Provenance & disclosures

• Traceability: 16 tagged knowledge-base claims touch Amgen (raw hits 16; 13 entity matches, 3 text-only; 4 used as name-level colour, 0 used as name-level conviction, 12 discarded; breadth 3, net conviction neutral-low). NOT ONE claim in the lane carries a named speaker — every record has speaker: null and is attributed only to a channel slug, the weakest sourcing the 4-lane attribution policy admits. Channels: biotech_hangout (8 entity claims, 61.5% of the lane — the concentration is a finding in itself); eli_lilly_ceo (4 claims, 30.8% — a competing pharmaceutical company's chief executive, treated as management/promoter voice, pulled OUT of the conviction pool and quoted separately at half weight); huberman_lab (1 claim, on leptin physiology, discarded as not about the equity). The freshest claim is 2026-01-16, over six months stale. Searches run: the primary sweep on AMGN and Amgen across 51,928 distilled claims (16 hits); brand sweeps on MariTide, Repatha, Prolia, TEZSPIRE, Otezla, ENBREL, KRYSTEXXA, TEPEZZA, olpasiran and Robert Bradway (zero incremental hits — no claim in the knowledge base names any Amgen brand outside the entity set already counted); and category sweeps on obesity (262 matches), GLP-1 (315), drug pricing (98), large pharma (9) and biosimilar (4), all discarded because they are claims about the therapeutic category, about competitors, or about US pricing policy — a claim about a company's market is not a claim about that company's stock, and triaging several hundred obesity/GLP-1 matches out of an Amgen lane was the most useful attribution work in this dive. Discards also include one homograph-class collision ("future Amgens" used as a common noun) and one 2019 resveratrol claim which, ironically, is the only claim in the entire sweep with a named speaker. All quotes are verbatim from the stored claim text.
• Data as-of: fundamentals — income statement, balance sheet and cash flow for the year ended 2025-12-31, from the 10-K filed 2026-02-13; most recent filing-verified financials 2026-03-31, from the 10-Q filed 2026-05-01; quarterly income data through 2026-03-31 · estimates 2026-08-04 · prices 2026-08-04, quote timestamp 1785873601 = 2026-08-04T20:00:01Z ($390.02, +2.94%; 50-DMA $357.74; 200-DMA $344.72; RSI 71.03; MACD +8.37) · knowledge-base claims 2026-08-04. Fiscal/calendar note: Amgen's fiscal year is the calendar year; "2025" means the twelve months to 2025-12-31. All figures come from the Synthos vendor data file for AMGN or from the SEC filings in the AMGN archive; no figure comes from memory, recall or external retrieval, and no clinical-trial detail is supplied beyond what the filings state.
• EARNINGS TIMING — the single most important provenance note. earn_cal carries an entry dated 2026-08-04 — the date of this dive — with epsActual: null and revenueActual: null, against estimates of $5.62 non-GAAP EPS and $9,426,191,000 of revenue. The second-quarter 2026 print had not landed when this dive was written and almost certainly lands after the 16:00 Eastern close from which the $390.02 price is taken. NO second-quarter 2026 figure appears anywhere in this document. The verdict was set with that timing as an explicit input: paying a 52-week-high, overbought price hours before an unseen print that consensus expects to show a 6.6% year-on-year EPS decline is the entry question, and the answer is Hold.
• Filing archive contents: 10-K filed 2026-02-13 (fiscal 2025); 10-Q filed 2025-11-05 (September 2025 quarter); 10-Q filed 2026-05-01 (March 2026 quarter); 8-K filed 2026-05-19 (Item 5.02 — CFO retirement and successor appointment); 8-K filed 2026-05-20 (Item 5.07 — annual meeting results); 8-K filed 2026-07-31 (Item 1.05 — material cybersecurity incident). None of the three 8-Ks is a bare cover page — all three carry full Item text, and no Exhibit 99.1 is required to read them. The pre-flight hypothesis that the 2026-05-19 and 2026-05-20 pair would be a note-offering pricing-and-closing sequence was WRONG and is stated as wrong: there is no financing 8-K in this archive. The $4.0B of notes issued in the March 2026 quarter is disclosed only in the 10-Q. Known extraction limit, stated rather than estimated around: numeric tables are stripped from the extracted filing text throughout, so every financial statement figure in this dive comes from the vendor payload, and filing corroboration comes from MD&A prose, percentages and narrative disclosures.
• Where the filings contradicted or corrected the vendor (detailed in Section 8): FY2025 cost of sales rejected — the vendor's $10.729B (29.2% of revenue, a 70.8% gross margin) against the 10-K's explicit "Cost of sales decreased to 32.8% of total revenues for 2025", implying $12.051B and a 67.2% gross margin, a $1.322B gap, with FY2024 reconciling exactly at 38.5% — we use 67.2%; the Q4-2025 gross margin of 81.9% rejected as the same error landing in a derived quarter, against a Q1-2026 figure of 31.84% that matches the 10-Q's stated 31.8% exactly; total debt rebuilt from the vendor's $54.604B at 2025-12-31 to approximately $57.3B at 2026-03-31, per the 10-Q's disclosure of $4.0B of notes issued, €750M ($833M) repaid and $324M repurchased, and its fair-value note showing fixed-rate carrying value rising from $52.8B to $55.5B; seg_prod FY2025 identified as missing KYPROLIS entirely, summing $1.402B short of total revenues while FY2024 and FY2023 sum exactly, with KYPROLIS 2025 revenue derived as the ≈$1.402B residual and labelled as derived; seg_prod relabelled as the worldwide product sales table, not a segment table, because the 10-K states Amgen "operates in one operating segment: human therapeutics"; the annual and quarterly income statements shown not to tie (revenue +$28M, cost of revenue −$122M, R&D −$307M, operating income −$386M), with the quarterly series preferred for run-rate work; est.epsAvg confirmed NON-GAAP by testing the 2025 row against both actual bases; the 2023 estimate row discarded outright as inconsistent with GAAP and non-GAAP alike; est.ebitdaAvg/ebitAvg not used after verification showed them 5-7% above actuals; quote.yearHigh/yearLow ($398.00/$269.77) rejected in favour of tech.hi52/lo52 ($393.10/$271.18); lastDividend of $9.80 identified as the TTM total rather than the last declared dividend, with the 10-K giving a forward run-rate of $10.08 (2.58% yield); and incomeTaxesPaid/interestPaid of zero identified as missing, not zero. Where vendor and filing AGREED — worth recording: total debt at 2025-12-31 ($52.8B of notes plus $1.8B of term loan = $54.6B against the vendor's $54.604B); the entire 2025 capital-return series (dividends $5.124B, zero buyback, $6.0B of debt retired, capex $1.9B, operating cash flow $10.0B); the product-sales and other-revenues split, both matching the filing's stated +10% and +15% growth rates; enterpriseValueTTM and the implied net debt, within $15M of our hand rebuild; and shortTermInvestments: 0, which is a TRUE zero — Amgen holds no marketable securities, the 10-Q's liquidity statement is cash-only, and the usual short-term-investment omission does not apply to this name.
• Basis note: analyst estimates are NON-GAAP — verified, not assumed, by testing the 2025 epsAvg of $21.318 against the GAAP actual of $14.23 and the reported non-GAAP actual of $21.85. The gap is enormous for this company: 2025 non-GAAP EPS is 53.5% above GAAP, or $4.13B after tax, driven overwhelmingly by acquisition-related intangible amortisation from Horizon (roughly $4.3B pre-tax in 2025) plus $1.2B of Otezla impairment. All forward multiples use consensus non-GAAP EPS; the trailing multiple is shown on both bases — 17.65x non-GAAP, 27.14x GAAP — and a hand-built "clean GAAP operating EPS" of $14.46 (27.0x) is given in Section 6. The 10-Q's disclosed amortisation run-off ($2.7B for the remaining nine months of 2026, then $3.6B, $2.8B, $2.3B, $2.2B and $2.1B for 2027-2031) means the two bases converge mechanically over time.
• Estimate coverage: 24 analysts on 2026 revenue and 19 on 2026 EPS; 22 and 19 on 2027; 25 and 11 on 2028 — note the 25-versus-11 split on the same year, and a 2028 EPS range of $19.679 to $26.861, a 36% spread. The 2029 and 2030 EPS rows rest on FOUR analysts each and are excluded from every conclusion. The 2023 row is discarded outright.
• Peer note: the vendor peer set contains eight genuine large-cap pharmaceutical comparables — Merck ($316.3B), Novartis ($291.4B), Novo Nordisk ($196.6B), Gilead ($167.9B), Pfizer ($144.8B), Bristol-Myers Squibb ($134.6B), GSK ($103.2B), Sanofi ($102.5B) — alongside two that are not (Boston Scientific, Danaher). No peer multiples are supplied in the file, so no peer-multiple comparison is drawn.
• Fair-value caveat: the $287 / $375 / $461 anchors are multiples of the 2027 consensus non-GAAP EPS distribution — 13x the low of $22.100, 16x the mean of $23.416, and 19x the high of $24.250. Stated arithmetic, not a discounted cash flow. The base is deliberately sensitivity-disclosed: 15x gives $351, 17x gives $398, 18x gives $421 — the whole answer lives in a 15-18x band and no greater precision is claimed. The base case sits 3.8% BELOW the market price and 2.7% above the street's $365.08, and the resulting asymmetry (26.4% down against 18.1% up) is the arithmetic basis of the Hold.
• Disclosures with no vendor field, recorded here because they are material and invisible on any screen: roughly $10.7B of gross IRS transfer-pricing exposure ($3.6B for 2010-2012; $5.1B plus $2.0B of penalties for 2013-2015; reduced by up to $3.1B of previously paid repatriation tax; plus interest), with a U.S. Tax Court decision expected no earlier than the second half of 2026 and examinations of 2016-2018 and 2019-2022 to follow; $6.9B of purchase obligations and up to $7.2B of contingent milestone payments off the balance sheet; and the Item 1.05 8-K of 2026-07-31 declaring a MATERIAL cybersecurity incident with exfiltration of proprietary data and patient protected health information, an ongoing investigation, and pending patient-notification obligations.
• Accessibility note: no information in this dive is conveyed by colour. All emphasis is carried by bold text, table structure and explicit labelling.
• Not investment advice. Independent research, educational and informational only, never personalised. No recommendation to buy, sell or hold any security is made to any person. This document was written before the release of Amgen's second-quarter 2026 results and contains no information about them.
• Version: 2026-08-04-full.