Broadcom and AMD sell silicon into the same AI buildout, and the market prices them as two versions of one trade. Their filings disagree. Read side by side, the two sets of accounts describe opposite machines: one that buys earnings and converts revenue into cash at software rates, and one that grows earnings organically and converts at hardware rates while paying for its seat at the frontier in paper. This article puts the numbers next to each other and lets the interpretation follow the arithmetic.
Every figure comes from a primary filing, the Broadcom fiscal 2025 10-K and its third-quarter fiscal 2026 10-Q, and the AMD fiscal 2025 10-K/A and its second-quarter 2026 10-Q, plus the earnings releases, unless a figure is labeled an estimate where it stands. Market ratios use the Friday, October 30, 2026 closes: Broadcom at $326.08 and AMD at $504.80. The fiscal calendars differ. Broadcom's latest quarter is Q3 fiscal 2026, ended August 2, 2026. AMD's latest quarter is Q2 2026, ended June 27, 2026, and AMD reports its third quarter after today's close, so the AMD side of every table is struck pre-print.
What do the two companies look like side by side?
| Metric | Broadcom | AMD |
|---|---|---|
| Revenue, latest fiscal year ($M) | 63,887 | 34,639 |
| Revenue, latest quarter ($M) | 29,591 | 11,536 |
| Revenue, trailing twelve months ($M) | 89,104 | 41,305 |
| Latest quarter growth, year over year | +86% | +50% |
| Market cap at October 30 close ($B) | 1,557 | 824 |
| Enterprise value ($B) | 1,592 | 814 |
| EV to sales, trailing | 17.9x | 19.7x |
| Trailing P/E, GAAP | 41.6x | 129.8x |
| Trailing P/E, non-GAAP | 33.6x | 86.0x |
| Forward P/E, consensus estimate | 18.7x | 45.0x |
| Revenue per employee, latest fiscal year | $1.94M | $1.12M |
| Free cash flow, trailing ($M) | 39,403 | 7,736 continuing |
| Free cash flow yield on market cap | 2.5% | 0.9% |
Start with the inversion in the middle of the table. On trailing sales the market prices AMD richer than Broadcom, 19.7 times against 17.9 times, even though Broadcom's revenue base is more than twice the size and its latest quarter grew faster. On trailing GAAP earnings the order flips violently: 41.6 times for Broadcom against 129.8 times for AMD. The gap between those two rows is the whole comparison in miniature. AMD's price is a claim on conversion that has not happened yet. Broadcom's price is a claim on conversion already in the bank, and the argument about Broadcom is never whether the cash exists. It is what the cash cost to assemble.
One basis note before the tables multiply. Data vendors print Broadcom trailing P/E figures anywhere from about 40 times to about 68 times depending on which EPS series they divide by. The 41.6 times here is computed from filed trailing GAAP diluted EPS of $7.84, so the denominator is explicit and the reader can re-perform it. The same discipline is applied to every ratio below: where a number could not be verified from a filing, it is omitted rather than estimated.
Where does the cash conversion gap come from?
The single widest river between the two companies is not growth. It is the share of each revenue dollar that survives to free cash flow. Broadcom turned $89.1B of trailing revenue into $39.4B of free cash flow, a 44.2% margin. AMD turned $41.3B into $7.7B on continuing operations, an 18.7% margin. A gap of about 25 points, on the same definition, in the same industry, in the same cycle.
| Cash metric ($M) | Broadcom | AMD |
|---|---|---|
| Operating cash flow, trailing | 40,653 | about 10,080 |
| Free cash flow, trailing | 39,403 | 7,736 continuing |
| Free cash flow margin, trailing | 44.2% | 18.7% |
| Free cash flow, latest quarter | 13,665 | 1,560 |
| Free cash flow to adjusted EBITDA, fiscal 2025 | 62.6% | 64.8% |
| Free cash flow to GAAP net income, trailing | 103% | 120% |
| Capex as a share of revenue, fiscal 2025 | 1.0% | 2.8% |
The lower rows locate the gap precisely, and it is not where most readers would look. Once profit exists, the two machines convert it to cash at nearly identical rates: free cash flow runs at 62.6% of adjusted EBITDA for Broadcom and 64.8% for AMD in fiscal 2025, and both companies convert GAAP net income into free cash flow at better than 100% over the trailing year. Neither company has a cash collection problem. The gap is upstream, in how much profit a revenue dollar produces in the first place, which is a margin story and gets its own section below.
What a 25-point conversion gap compounds into is the strategic difference between the two balance sheets. Broadcom's trailing free cash flow of $39.4B nearly equals AMD's entire trailing revenue of $41.3B. A company converting at 44% funds its commitments, its buybacks, and its next acquisition from operations. A company converting at 19% funds a buildout from its balance sheet and, at the frontier, from instruments tied to its own share price. The latest quarter shows the direction of travel on both sides: Broadcom printed $13,665M of free cash flow in one quarter, 46% of its revenue, while AMD's free cash flow fell 39% sequentially to $1,560M as capex more than doubled to $808M and working capital absorbed the Helios build ahead of its revenue.
A basis note on AMD free cash flow. Two AMD free cash flow bases circulate. Figures computed from total operating cash flow, including the divested ZT Systems manufacturing business in discontinued operations, give fiscal 2025 free cash flow of $6,735M and trailing free cash flow of $8,403M. On the continuing operations basis used in AMD's own selected data, the figures are $5,519M and $7,736M. This article uses the continuing basis throughout and labels it where it stands.
How do two opposite accounting shapes produce these numbers?
Broadcom is an acquisition machine that reports like one. It buys earnings streams, consolidates them, and then asks the reader to look past the amortization of the purchase price when judging the result. AMD is an organic grower whose reported base is distorted less by what it bought than by what hit it: China export charges, a tax reserve release, and investment marks running through GAAP profit.
| Fiscal 2025 ($M unless noted) | Broadcom | AMD |
|---|---|---|
| Gross margin, GAAP | 67.8% | 49.5% |
| Operating margin, GAAP, latest quarter | 54.1% | 17% |
| Operating margin, non-GAAP, latest quarter | 67.9% | 27% |
| Stock-based compensation | 7,568 | 1,638 |
| Amortization of acquisition intangibles | about 8,100 | 2,254 |
| Net income, GAAP | 23,126 | 4,335 |
| Net income, non-GAAP | 33,728 | about 6,822 |
| Non-GAAP uplift over GAAP | +46% | +57% |
| Goodwill and net intangibles | about 124,121 | 41,831 |
| Goodwill and intangibles as a share of total assets | 66% | 54% |
| GAAP effective tax rate | -1.7% | -2.5% |
The Broadcom bridge deserves a direct read. In fiscal 2025 the company excluded about $8.1B of acquisition amortization and $7,568M of stock compensation to get from $23,126M of GAAP net income to $33,728M non-GAAP, an uplift of 46%. The stock comp is a real cost that shows up in the share count, and the next section prices it. The amortization is the line that defines the model: at about $2,006M a quarter, it is the recurring purchase price of earnings that were bought rather than built, and excluding it every year excludes the model's core input cost. Two-thirds of the balance sheet carries the same signature. Goodwill of $97,801M plus net intangibles of about $26,320M equals about $124B, or 66% of total assets of $188,150M. The fiscal 2024 accounts show what the machine costs in a deal year: net income of $5,895M, an 11.4% margin, as VMware closing costs, restructuring, and amortization ran through the P&L. By the third quarter of fiscal 2026 the same machine printed a 54.1% GAAP operating margin, with the infrastructure software segment at an 83.7% operating margin.
AMD's distortions are of a different kind. Its fiscal 2025 GAAP gross margin of 49.5% carries about $800M of MI308 inventory and related charges booked in the second quarter of 2025, when the GAAP gross margin printed 40%, and a net China charge of about $440M for the year after licensed MI308 shipments resumed in the fourth quarter. The same base year holds an $853M release of reserves for uncertain tax positions, a GAAP benefit excluded from non-GAAP. In the latest quarter, GAAP net income of $2,297M includes $483M of gains on long-term investments, about a fifth of the quarter's GAAP profit arriving as marks rather than operations. Year-over-year comparisons for AMD through mid-2026 are comparisons against this base, and any reader scoring its margin recovery without adjusting for it is grading the charges rolling off, not only the business improving.
One line in the table belongs to both. Each company printed a negative GAAP effective tax rate in fiscal 2025, Broadcom at -1.7% on a $397M provision credit and AMD at -2.5%, while their own normalized non-GAAP rates sit at about 16% and 13% respectively. GAAP EPS at both companies is currently flattered by a tax line that is not a run rate, and both companies say so themselves in their non-GAAP guides.
Who dilutes, and how?
Both companies dilute. They do it in opposite tenses. Broadcom's dilution is in the past tense, already issued, printed in a share count that rose 13.6% in three years. AMD's is in the future conditional, a claim on 320M shares that vests only if the AI deals succeed, and whose cost will be recognized through the income statement before a single share is issued.
| Stock comp and share count ($M unless noted) | Broadcom | AMD |
|---|---|---|
| Stock-based compensation, fiscal 2023 | 2,171 | 1,384 |
| Stock-based compensation, fiscal 2024 | 5,741 | 1,407 |
| Stock-based compensation, fiscal 2025 | 7,568 | 1,638 |
| SBC as a share of revenue, fiscal 2025 | 11.8% | 4.7% |
| SBC as a share of free cash flow, fiscal 2025 | 28.1% | 29.7% |
| Buybacks, fiscal 2025 | 2,450 | 1,316 |
| Buybacks, latest nine or six months | 8,450 | 221 |
| Share count change, fiscal 2023 to 2025 | +13.6% diluted | +0.9% basic |
On revenue, Broadcom's stock comp intensity is two and a half times AMD's, 11.8% against 4.7%. On free cash flow the gap nearly vanishes, 28.1% against 29.7%, because AMD's cash base is so much smaller. Which denominator matters depends on which company is being examined, and the share count settles the argument for Broadcom. Its diluted count rose 581M shares from fiscal 2023 to fiscal 2025 as VMware deal stock and equity awards landed. In the first nine months of fiscal 2026 the company spent $8,450M on buybacks, and the shares outstanding still rose, from 4,741M to 4,774M. At current run rates the Broadcom buyback is an SBC offset program, not a count reduction program, and in fiscal 2023 and 2024 the offset did not even run: buybacks were zero against $2,171M and $5,741M of stock comp.
AMD's count, by contrast, barely moves: basic shares up 15M, or 0.9%, across the same three years. The dilution sits off the count, in two warrants disclosed in the second-quarter 2026 10-Q. OpenAI holds a warrant for up to 160M shares at a $0.01 exercise price, vesting in tranches on deployment scale from 1 GW to 6 GW, share price targets, and technical and commercial milestones. Meta holds a second warrant for up to 160M shares at $0.01, with stock price thresholds escalating to $600 and an exercise window through February 2031. Combined, 320M shares equal 19.6% of the 1,632M basic count if every tranche vests. Both warrants are carried as liabilities and marked to market through earnings. No tranche had vested at June 27, 2026. The forensic point is the sequence: the better the deals work, the higher the share price climbs, the larger the liability mark and the eventual dilution. Success dilutes AMD mechanically, and the charge arrives before the shares do. AMD also spent $607M in fiscal 2025 and $355M in the first half of 2026 repurchasing shares for employee tax withholding, outside the program figures in the table, while its program buybacks ran at just $221M in the first half of 2026 with none at all in the second quarter.
Whose backlog can an outsider audit?
Forward visibility is where the two disclosure regimes diverge most sharply, and it is the section of this comparison with the least interpretation in it. One company files its forward book in dollars under contract. The other states its forward book in gigawatts under agreements whose operative word is "up to."
| Forward book | Broadcom | AMD |
|---|---|---|
| Unit of account | Dollars | Gigawatts |
| Order backlog | about $73B of AI orders, 18-month view | not disclosed |
| Remaining performance obligations, filed | $179.2B | not disclosed |
| Purchase commitments ($M) | 126,820 | 30,276 |
| Forward agreements | signed contracts, incl. non-terminable | about 14 GW, up to language |
| AI revenue guide, next two fiscal years | about $115B and about $230B | no dollar guide issued |
| AI accelerator revenue line | reported every quarter | Instinct revenue not reported |
Broadcom's remaining performance obligations, the transaction price allocated to unsatisfied obligations under firmly committed multi-year contracts, rose from $33.3B at November 2, 2025 to $179.2B at August 2, 2026. The jump is driven by one long-term custom AI accelerator contract signed in the second quarter of fiscal 2026, and the customer is not named in the filing. About 25% of the total is expected to be recognized within twelve months, and the measure excludes uncommitted software and subscription contracts. Alongside it sit the $73B AI order backlog management cited on the fiscal Q1 2026 call and $126,820M of unconditional purchase commitments, essentially all due inside fiscal 2027 and 2028 on the prior quarter's schedule. The purchase book is the mirror image of the order book: both sides of the AI buildout now run through Broadcom's signature, and both can be checked against delivery every quarter.
AMD's forward book is three capacity agreements. OpenAI covers up to 6 GW, with the first 1 GW of MI450 systems scheduled to begin deployment in the second half of 2026. Meta covers about 6 GW on the same first-gigawatt schedule, built on a custom MI450-based GPU. Anthropic covers up to 2 GW, first gigawatt in the first half of 2027, paired with up to $5B of AMD equity investment in Anthropic. Total, about 14 GW. No filing or release states a dollar backlog, and none states Instinct revenue for any period, so the conversion of gigawatts into dollars cannot be verified from the outside. Analyst estimates put AMD's 2026 AI GPU revenue at about $8B, with upside cases of $10B to $12B, and they are labeled estimates here because that is all they are. The status evidence is real but qualitative: initial Helios systems were installed and being tested at Meta as of October 27, 2026, and Microsoft and Oracle are named deployment customers. What an outsider can check next quarter follows directly. For Broadcom: fourth-quarter AI revenue against the $21.7B guide, and RPO progression in the 10-K. For AMD: the Data Center segment line only, which blends EPYC CPUs, whose sales grew about 70% year over year in the second quarter, with the Instinct GPUs the segment's growth story actually depends on.
How are both sellers financing their own demand?
In The Compute Price Reset, the credit analysis ran from the buyers: anchor tenants with prime balance sheets signing term deals, and a spot tail paying ceiling rates. The filings of the two chipmakers show the same machine running in reverse. The sellers are now financing the demand side of their own market, and the difference between them is scale.
| Customer financing exposure ($M unless noted) | Broadcom | AMD |
|---|---|---|
| Lease backstop or guarantee maximum | about $29B backstop | 4,100 guarantee exposure |
| Customer note capacity | up to $42B of convertible notes | n/a |
| Equity committed into customers | XPV platform with Apollo and Blackstone | up to $5B into Anthropic |
| Contract assets | 11,901, up 33% in nine months | not separately disclosed |
| Customer concentration | top five about 50% of 9M fiscal 2026 revenue | warrant-linked anchor deals |
Broadcom's exposure is structural. Under a lease backstop arranged in June 2026 its maximum potential liability is about $29B, undiscounted, if all covered AI racks deploy. A customer may issue Broadcom convertible promissory notes up to $42B of aggregate principal, usable solely for that customer's lease obligations, with none issued at August 2, 2026. The XPV platform, co-built with Apollo and Blackstone, closed its first $35B tranche in June 2026 and targets more than 20 GW for Anthropic and OpenAI by the end of 2028. None of this is hidden; all of it is disclosed, contingent, and concentrated in the same customers who produce the revenue. Broadcom's top five end customers were about 40% of fiscal 2025 net revenue and about 50% for the first nine months of fiscal 2026, and distributors carried 56% of nine-month revenue. Part of the demand behind the $179.2B backlog is therefore financed with Broadcom contingent support, and the forensic question is not whether the exposure exists. It is whether the revenue and the support are being counted as independent evidence of the same demand.
AMD's version is smaller by roughly an order of magnitude and different in kind. Its maximum gross exposure from guaranteeing commercial partners' data center lease obligations is $4,100M, on terms up to 15 years, carried as a credit derivative at a fair value the filing calls not material. It has committed up to $5B of equity into Anthropic through fiscal 2028 and invested $1,000M in Meta's October 2026 equity offering. Where Broadcom finances customers with contingent balance sheet, AMD finances them with its own equity and equity-linked paper: the warrants of the previous section, plus cash invested directly into the buyers. The direction is the same. The seller's balance sheet is now part of the demand stack, at both companies, and any grading of either backlog has to net the support out before it counts the orders as proof.
What does working capital say about strategy?
| Working capital | Broadcom, Nov 2025 | Broadcom, Aug 2026 | AMD, Dec 2025 | AMD, Jun 2026 |
|---|---|---|---|---|
| Accounts receivable ($M) | 7,145 | 13,707 | 6,315 | 7,281 |
| Inventories ($M) | 2,270 | 4,523 | 7,920 | 8,468 |
| Accounts payable ($M) | 1,560 | 4,000 | 2,929 | 5,359 |
| Receivable days (DSO) | 36.1 | 42.2 | 55.9 | 57.4 |
| Inventory days | 35.8 | 45.1 | 153.6 | 144.5 |
| Payables days | 24.6 | 39.9 | 56.8 | 91.4 |
| Cash conversion cycle (days) | 47.3 | 47.4 | 152.7 | 110.5 |
AMD carries about 145 days of inventory against Broadcom's about 45, and that single line is a strategy statement. The $8,468M of inventory at June 27, 2026 is a bet, placed ahead of revenue, that Helios racks and MI450 systems convert in the second half of 2026. The downside case sits in the company's own recent history: when China licenses stopped in 2025, the same posture produced about $800M of MI308 charges in one quarter. Note also how AMD's cycle improved. The cash conversion cycle tightened from about 153 days to about 111 days in six months, and the improvement was funded by stretching payables from 56.8 to 91.4 days, an 83% rise in the payables balance, not by moving inventory or collecting faster. Suppliers are financing the build.
Broadcom runs a 47 day cash conversion cycle, and it did not move at all across the window, 47.3 days to 47.4, which makes the levels underneath it more interesting. Receivables rose 92% in nine months, from $7,145M to $13,707M, inventories rose 99%, and DSO drifted from 36.1 to 42.2 days while contract assets climbed 33% to $11,901M. The acquisition machine historically ran asset-light, with capex at 1.0% of revenue. At AI rack scale, growth has started absorbing its balance sheet the way it absorbs everyone else's. The direction, not the level, is the flag: Broadcom's working capital is beginning to behave like a hardware company's, because at the rack level it now is one.
Whose guidance can be graded?
| Broadcom AI revenue, guide vs delivered ($B) | Guide | Delivered | Score |
|---|---|---|---|
| Q4 fiscal 2025 | about 6.4 | about 6.5 | Beat |
| Q1 fiscal 2026 | 8.2 | 8.4 | Beat |
| Q2 fiscal 2026 | 10.7 | 10.8 | Beat |
| Q3 fiscal 2026 | about 16.0 | 16.7 | Beat |
| Q4 fiscal 2026 | 21.7 | pending | n/a |
Broadcom guides its AI line in dollars, every quarter, as a point estimate, and it has now beaten that guide for eight consecutive scored quarters, a streak that runs back to the fiscal 2024 AI target of about $10B, set in December 2023 and delivered at about $12.2B. The streak matters forensically because of what sits on top of it: the fiscal 2027 AI guide of about $115B and the fiscal 2028 guide of about $230B are extrapolations issued by a management team whose short-range guides keep landing low. A reader can disagree with the extrapolation. A reader cannot say the guide-maker has no record.
Fairness requires the parallel fact for AMD: its total revenue guides have also been beaten, four quarters running, with deliveries at or above the top of the guided ranges from Q3 2025 through Q1 2026 and above the midpoint in Q2 2026, and its Q3 2026 guide of about $13.0B is scored after today's close. The asymmetry is narrower and more specific than a credibility gap. AMD guides the total and has never issued a dollar guide for Data Center or for Instinct. The segment that carries its entire growth story, and the product line on which 14 GW of commitments and 320M warrant shares depend, has no guidance record to score, because there is nothing to score it against. Auditability is a property of the disclosure regime, not of the quarter.
What do the numbers say?
Stripped of narrative, the contrasts are these. Broadcom converts trailing revenue to free cash flow at 44.2% and AMD at 18.7% on continuing operations. Broadcom's latest-quarter GAAP operating margin is 54.1% and AMD's is 17%. Broadcom carries $35.4B of net debt against $179.2B of filed obligations to deliver and $126.8B of commitments to buy. AMD carries $9.9B of net cash, and stays net cash pro forma for the $4.75B of notes it issued in August 2026, against $30.3B of purchase commitments, $14.0B of leases that commence later, $4.1B of guarantee exposure, and up to $5B of customer equity commitments. On fiscal 2025 returns, Broadcom earned a 29.7% return on equity and a 16.2% return on invested capital on the formulas stated in our research files, against 7.2% and 5.8% for AMD, with AMD at 12.3% on a non-GAAP operating basis. AMD spends the larger share of revenue on research and development, 23.4% against 17.2%, on a revenue base about half the size.
Neither shape is a verdict by itself. The acquisition machine's numbers are audited, contracted, and cash-verified, and they rest on bought earnings, a balance sheet that is two-thirds acquisition residue, and a customer financing stack whose support and revenue point at the same names. The organic grower's numbers are cleaner in origin and harder to verify where it counts, with its key revenue line undisclosed, its forward book denominated in gigawatts, and a fifth of its share count waiting in warrants. The risk in Broadcom's accounts is concentration and circularity, visible in the filings. The risk in AMD's accounts is that the most important line is not in the filings at all.
The DalalBytes five-pillar library scores Broadcom at 76 of 100, Positive, and AMD at 68 of 100, Selective Positive, both struck at the October 30, 2026 closes used throughout this article and linked here for the full pillar work. Those scores were built from the same filings mined above; this article is the machinery underneath them, published so the reader can re-perform the arithmetic and disagree with the weightings rather than with the numbers. AMD's side of every table predates its third-quarter print, which lands after today's close, and the AMD package refresh follows it.
Frequently asked questions
Which company converts revenue into cash more efficiently?
Broadcom by a wide margin. It converted 44.2% of trailing twelve-month revenue into free cash flow against 18.7% for AMD on continuing operations. Broadcom also generated $39.4B of trailing free cash flow on $89.1B of revenue while AMD generated $7.7B on $41.3B.
Why does AMD not disclose a dollar backlog?
AMD states its forward book in gigawatts rather than dollars. Its OpenAI and Meta and Anthropic agreements cover about 14 GW of capacity under up to language tied to warrants and milestones. Broadcom by contrast files remaining performance obligations of $179.2B under signed contracts. Only one of the two books can be audited in dollars.
What is the biggest accounting difference between the two?
Broadcom runs an acquisition machine and AMD runs an organic one. Broadcom carries goodwill and intangibles of about $124B at 66% of total assets and its non-GAAP profit runs 46% above GAAP. AMD carries its dilution as warrants on 320M shares held as liabilities and marked through earnings.
Disclosures
Not financial advice. This article is published for educational purposes only. Nothing here is a recommendation to buy, sell, or hold any security, and nothing here is financial advice. Investing involves risk, including the possible loss of principal.
Positions. The author holds a long position in AMD, discussed in this article. The author holds no position in Broadcom.
Accuracy. Company figures are from primary filings: Broadcom's fiscal 2025 Form 10-K, its third-quarter fiscal 2026 Form 10-Q, and its earnings releases through the third quarter of fiscal 2026; and AMD's fiscal 2025 Form 10-K/A, its second-quarter 2026 Form 10-Q, and its earnings releases through the second quarter of 2026. Forward P/E and forward revenue growth are analyst consensus estimates from a StockAnalysis snapshot of November 2, 2026, and are labeled estimates where they stand. Analyst estimates of AMD's 2026 AI GPU revenue are third-party estimates, not reported figures, and are labeled as such. AMD figures predate its third-quarter 2026 report of November 3, 2026. The interpretations, forensic flags, and contrasts are the author's analysis of the filed numbers presented. Errors can occur and facts change; verify independently before acting on anything you read.
No advisory relationship. Reading this site does not create an advisor-client relationship of any kind. See the full disclosures page.