What we said in September — and what happened
The discipline held; three of our facts did not
On 10 September we published Broadcom at $362.04 under the title "Too Concentrated at This Price". We called it two companies at one price: a toll bridge of software and old chips worth about a hundred dollars a share, and an AI promise resting on six customers. We put a blended value of about $430 on the whole and said we would own it below about $280. On 5 November the shares closed at $291.53 and touched $289.77 during the day, 3.5% above our level. They closed at $341.60 on 17 November. We bought nothing, by our own rule, and the rule has not yet cost anything. Here is the account, including three facts we got wrong.
| What we said on 10 Sep | What happened by 17 Nov | Grade |
|---|---|---|
| 'Own below ~$280' | Lowest close $291.53 (5 Nov); lowest trade $289.77. Not reached. | ◆ Not triggered |
| 'The financing vehicle is the line to read first'; no aggregate cap | A second backstop of up to ~$43bn was disclosed on 27 October. And the 10-Q we cited also disclosed a commitment to accept up to $42bn of the customer's convertible notes, which we left out. | ✓ Right, and understated |
| Anthropic: 'a private lab that loses money' | Its S-1/A shows a GAAP loss of $104bn for nine months, mostly fair-value changes, but positive adjusted net income (+$5.6bn) and operating cash flow (+$6.6bn). And it is about to list. | ✗ Wrong on cash |
| Top five customers '~40%+' of revenue | The 10-Q filed the same day said the top five end customers were about 55% of Q3 revenue. We quoted a year-old figure. | ✗ Wrong |
| Apple's '$30B+ July commitment' | The 8-K of 6 July describes multi-year agreements through 2031 and gives no dollar figure. We cannot source the $30bn and withdraw it. | ✗ Unsourced |
| Software 'growing 29%' | Q3 was +29%, after +1% and +9%; the nine months were +13%. Q4 is guided to hold at ~$8.7bn. A step, not yet a trend. | ◆ Partly |
| Gross debt '~$58B after the post-quarter paydown' | Principal was $61.08bn at 2 August; no further paydown disclosed beyond $752m of notes due in September. Net debt of ~$35bn was right. | ◆ Minor error |
| Dividend 'policy 50% of FCF'; December raise the largest ever | We could not find the policy in this year's filings, and the board has not applied it mechanically (dividends were 57% of prior-year FCF in FY25, ~46% in FY26). The decision is on 9 December. | ◆ Too early |
| FY27 AI ~$115bn, FY28 ~$230bn | Repeated by the chief executive on 12 October, as reported. The finance chief said the AI outlook 'is not expected to be updated quarterly'. | ◆ Too early |
| The chairman selling | A second sale of ~$250m on 23 September, at $354.71–361.79, under a trading plan adopted in December 2025. | ✓ Right |
The errors, plainly. Two of them came from the same mistake: we read the 10-Q for the financing vehicle and did not read the rest of it, which held both the 55% customer figure and the $42bn note commitment. The Apple figure came from reports we did not check against the filing. None of the three changes our conclusion: the concentration was worse than we said, the exposure larger, and the one customer we wrote off as a cash drain has turned out to generate cash. They cut in different directions, and an honest account has to show all of them.
★ Down, nearly to our zone, and back
Ten weeks of sector weather, two pieces of Broadcom news
| Date | What happened | AVGO that day |
|---|---|---|
| 14 Sep | Chip shares fall after technology executives call for slower AI development; the Q4 guide reads as slightly light (finviz, third-party). | −4.8% |
| 16 Oct | Fortune reports Tan's remarks at an investor forum: AI investors are 'almost like they are in a casino', and capital 'might be the biggest constraint'. | −4.5% |
| 27 Oct | 8-K: a second residual-value backstop, up to ~$43bn. Reuters reports $63bn of bank and Blackstone financing for Anthropic's Broadcom racks (via TIKR). | +4.0% |
| 30 Oct | Alphabet's results disappoint; chips slide. | −5.5% |
| 5 Nov | The semiconductor index falls 12.25%, its second-worst day on record, on fears about AI spending and regulation after the midterms (24/7 Wall St, third-party). | −10.4% |
| 12 Nov | Nvidia's chief executive resigns with immediate effect (Nvidia 8-K); Nvidia gives no reason. Broadcom has no news. | +5.5% |
| 13 Nov | Redburn downgrades to Neutral with a $398 target, questioning the 'quality and durability' of earnings tied to Anthropic and OpenAI (via TIKR). | −4.8% |
What moved the price, and what did not. Read the table again and count the Broadcom news: the second backstop, and an analyst's downgrade. Everything else was weather. The semiconductor index lost an eighth of its value in a day on 5 November; Broadcom lost a tenth. Nvidia's chief executive resigned and Broadcom rose more than Nvidia's own shares fell. No guidance changed. No customer was lost or announced. The fourth-quarter outlook given in September stands until 9 December.
That matters for how an owner should read the low. At $291.53 the market was not telling us anything new about Broadcom; it was selling chip shares. That is exactly the kind of day a price discipline exists for, and it is also why missing it by four per cent is not a reason to move the line. A level set for good reasons is not improved by chasing the price that nearly reached it.
★★ The financing loop, measured
Two backstops, a note commitment and an equity cheque — all for one customer
In September we called the financing vehicle "the line to read first". Here is the full reading, at the maximums Broadcom has disclosed. First, a residual-value backstop on the first leasing vehicle: if the lessee defaults and the racks are worth less than 85% of the lease balance, Broadcom makes up the difference, up to about $29bn. Second, disclosed in an 8-K on 27 October, a similar backstop on a new vehicle, at 72.5% of the lease balance, up to about $43bn. Third, in the 10-Q of September, a commitment to accept up to $42bn of the customer's convertible notes. Fourth, according to Anthropic's S-1/A, Broadcom will buy about $1.0bn of its shares in a private placement alongside the listing.
The 8-K does not say whether the two backstops add up, and none of these sums is an expected loss. The backstops pay only if the lessee stops paying and the equipment has lost value; the notes are issued only to pay rent. But the maximums together come to about $115bn, about three times Broadcom's net debt, all tied to one customer. Reuters, as reported by TIKR, says banks and Blackstone have lined up $63bn to fund the racks behind the second vehicle, with deliveries from 2027.
| Commitment | Source | Maximum |
|---|---|---|
| Backstop, first vehicle (85% of lease balance) | 10-Q, Q3 FY26 | ~$29bn |
| Backstop, second vehicle (72.5%) | 8-K, 27 Oct 2026 | ~$43bn |
| Convertible notes to accept (9% paid in kind; 13% if revenue conditions are missed) | 10-Q; Anthropic S-1/A | $42bn |
| Anthropic shares in the listing (~13.5m at the $74 mid-point) | Anthropic S-1/A | ~$1bn |
| OpenAI financing (reported, unconfirmed) | WSJ via TIKR, October | >$50bn arranged |
What the rating agencies see. Fitch affirmed Broadcom at A− with a Positive outlook on 27 October, according to Investing.com, counting more than $70bn of lease financing and the $42bn note facility; it said the outlook would return to Stable if net debt adjusted for the backstops passed $150bn. That is the right frame. On a strict reading, Broadcom has become part chip-maker and part lender to its largest future customer, and the size of that lending is now comparable to a year of its revenue. An owner does not need to fear it, but should price it.
The OpenAI question. The Wall Street Journal reported in October that Broadcom is arranging more than $50bn of financing for OpenAI's custom chip, with talks at an early stage. Broadcom has not commented. If a third structure of this size appears, the frame Fitch set out becomes the one to watch.
★ The customer, in its own words
Anthropic's S-1/A turns the largest unknown into a set of accounts
In September the biggest single risk in our report was a customer we could not see: a private laboratory, expected to become Broadcom's largest buyer of custom processors in 2027, which we described as losing money. On 26 October Anthropic filed to list its shares, and on 17 November it filed its third amended prospectus. The accounts are now public, and they change the picture in both directions.
| From Anthropic's S-1/A (nine months of 2026) | Value | What it means for Broadcom |
|---|---|---|
| Revenue | $33.2bn | A real business, not a research budget. |
| Net loss, GAAP | −$104.0bn | Mostly $94.3bn of fair-value changes on its own financial instruments, not cash. |
| Adjusted net income (company's measure) | +$5.6bn | Our September phrase 'loses money' was wrong on a cash basis. |
| Operating cash flow | +$6.6bn | Cash coming in, before the new racks arrive. |
| Rent owed to Broadcom and its financing entities, five years | ~$161.2bn | About $32bn a year, against ~$44bn of annualised revenue. |
The good news is that the counterparty is stronger than we described: a company with tens of billions of revenue, generating cash, and about to raise a large sum in public markets. The listing gives it access to capital that a private lab does not have. The sobering news is the size of the obligation. Rent of about $161bn over five years, around $32bn a year, is a large share of what Anthropic earns today, and it rises as the racks are delivered. It depends on Anthropic's revenue continuing to grow very fast. The prospectus also says Broadcom assigned part of the equipment to a new financing entity in October, which did not raise the rent.
So the question we asked in September has changed shape. It is no longer whether a loss-making laboratory can pay. It is whether a fast-growing, cash-generating company can keep growing fast enough to carry a very large fixed commitment, in a market where its competitors, among them OpenAI and Google, are spending at the same pace. That is a question about the whole AI industry, and Broadcom's maximum exposure to it now has a number.
Google remains the foundation: its supply agreement runs as late as 2031, according to an 8-K cited by TIKR. Bloomberg Intelligence, as reported, expects Google's TPU sales to rise from about $12bn this year to more than $135bn in 2028, and Google is reportedly now selling TPUs to other companies, including Anthropic. That raises the volume Broadcom builds while making Google more of a merchant of its own design. Meta and ByteDance: nothing new from primary sources. OpenAI: the reported financing above, and first-generation volume reportedly due in 2027. Apple: custom-chip agreements through 2031, with no dollar figure. We found no seventh custom-processor customer.
The ramp, and the bridge
AI revenue doubling on schedule; software stepping up, margins stepping down
Broadcom has not reported since our September report; its fourth-quarter results arrive on 9 December. What we have is the guide: revenue of about $34.8bn, of which AI semiconductors about $21.7bn, other chips about $4.3bn and software about $8.7bn, at a gross margin of about 73% and an operating margin of about 66% on the company's adjusted measure. If it lands, AI revenue for the fiscal year will be about $57.6bn, against about $20bn the year before. The contracted backlog reported in the 10-Q was $179.2bn, about a quarter of it due within twelve months, and Broadcom had $126.8bn of commitments to buy from its own suppliers, mostly in 2027 and 2028.
The bridge. We wrote in September that software was growing 29%. The quarter was +29%, but it followed +1% and +9%; for nine months the rise was 13%. The fourth-quarter guide of about $8.7bn would be 25% above a year earlier, so the third quarter's level may be the new floor, but two quarters do not make a trend, and VMware's price increases, which drove the growth, are also what drove its customers to court (Part XII). The margin. Adjusted gross margin was 75.0% in the third quarter and is guided to about 73% in the fourth, down from 78% earlier in the year, as AI hardware becomes a larger share of sales. Broadcom's profits will keep rising in dollars while falling as a percentage, and an owner should not mistake one for the other.
| Q3 FY2026 | Value | vs a year earlier |
|---|---|---|
| Revenue | $29.59bn | +86% |
| AI semiconductors | $16.7bn | +221% (+54% on Q2) |
| Infrastructure software | $8.75bn | +29% |
| Adjusted operating margin | 67.9% | — |
| Free cash flow | $13.67bn | +95% |
| Q4 guide | $34.8bn revenue; AI $21.7bn | Results 9 December |
The moat, rechecked
Wide where it was old, still contested where it is new
| The claim (September) | Evidence now | Verdict |
|---|---|---|
| Networking: the Ethernet toll gate | No market-share data published since; but Meta is expanding Nvidia's Spectrum-X Ethernet (Nvidia, 17 Nov). | ◆ Contested |
| Custom chips: the implementation craft | Six customers, backlog $179bn; MediaTek's next Google part reportedly slipped to January 2027; Nvidia now offers NVLink Fusion for MediaTek's custom chips. | ✓ Holding |
| Software: customers who cannot leave | Ten-year VMware deal with NTT DOCOMO and a FedEx renewal (as reported); Tesco's suit set for trial in 2027–28. | ✓ Holding, resented |
| Scale and supply | $126.8bn of purchase commitments secure supply through 2028; Broadcom now also finances demand. | ✓ Wider, riskier |
We keep the moat at 8. Nothing in ten weeks narrowed it. The sharpest new fact is that Broadcom's moat now has a financial component: part of the reason the largest AI laboratories choose its chips is that Broadcom helps pay for the racks. That is a real advantage while capital is plentiful. It is the kind of advantage that turns into a liability when capital is scarce, which is what Tan warned of in October, saying capital "might be the biggest constraint".
- That fiscal 2027 lands near the guide (about $115bn of AI revenue) and fiscal 2028 grows from there, to roughly $22 of adjusted earnings a share, without the doubling the company has promised.
- That the commitments behind Anthropic stay commitments: the lessee pays its rent, the racks hold their value, and the backstops are never called.
- That the software bridge holds its new level and the gross margin settles in the low 70s rather than sliding further as hardware grows.
The competition
Nvidia still scaling, Marvell buying optics, Google selling its own chips
| Rival | Where it meets Broadcom | The latest | Read |
|---|---|---|---|
| Nvidia | AI systems and networking. | Q3 revenue $115.8bn (+103%), Q4 guided at $130bn; Meta expanding Spectrum-X; NVLink Fusion extended to MediaTek's custom chips; Anthropic buying more Nvidia capacity through clouds. Chief executive resigned 11 Nov. | ◆ Bigger, and in the custom market |
| Marvell | Custom chips and optical interconnect. | Agreed to buy Lumentum in an all-stock deal valued at about $115bn (Light Reading); results on 1 December. | ◆ Integrating |
| MediaTek | Google's second source for TPUs. | Next TPU reportedly slipped to a January 2027 tape-out (unverified press). | ✓ Slower than feared |
| The customers' own teams | Google, Meta and others design more in-house. | Google reportedly selling TPUs to third parties. | ◆ The long-term risk |
Across our board. Nvidia's quarter on 17 November answered one fear for the whole sector: demand for AI computing is still rising, and fast. Its supply commitments rose from $279bn to $390bn in a quarter. For Broadcom that is good weather. But two details in the same release cut the other way: Meta, a Broadcom customer, is expanding its use of Nvidia's Ethernet networking, and Nvidia's NVLink Fusion now reaches into the custom-chip market through MediaTek. Nvidia is no longer only the rival Broadcom's customers buy custom chips to avoid. It is beginning to sell into the custom market too.
The resignation of Nvidia's chief executive is a governance shock at the industry's largest company, and Broadcom's shares rose on it. We would not read anything into that. Nvidia said its operations and roadmap continue without interruption, and no customer has been reported to move orders.
Management & capital allocation
A candid chief executive, a selling chairman, and a company that now lends
Candour. We said in September that Tan's guidance record is the cleanest of any large chief executive we have scored. The October remarks add to it. A seller of AI chips telling investors that AI investing resembles a casino, and that capital may be the constraint, is saying something against the short-term interest of the business. That is the kind of candour owners should value, and it is also a warning to take seriously: the person who knows the order book best is telling you the financing is the weak link.
Capital allocation. We lower the score from 7 to 6. Broadcom bought back $8.45bn of shares in the first nine months, nothing in the third quarter, and repaid $5.64bn of notes in that quarter; that is the old discipline. But the commitments in Part III are a new kind of capital allocation, and not one Tan's record prepares us for. A company that buys businesses at fair prices and runs them for cash is easy to judge. A company that lends its balance sheet to its largest customer so that the customer can buy its products is harder, because the profit is booked now and the risk arrives later. It may be the right decision; it is certainly a different one.
Insiders. No officer or director bought shares in the open market since September; the finance chief and chief legal officer had shares withheld for tax. The chairman's sales are under a plan set in advance and are small against the remaining holding, but $500m in four months is a pattern, not an event. At the April annual meeting 66.4% of votes cast supported executive pay.
The numbers
A cash machine growing into its price, and six faults in our feed
Three ways to read the price. Trailing GAAP earnings of about $7.94 a share put the shares at 43 times, burdened by the amortisation of past acquisitions. Trailing adjusted earnings of $9.76 put them at 35 times. Free cash flow of $39.4bn over twelve months, about $8.06 a share, puts them at about 42 times; on our estimate of about $48bn for the fiscal year now ending, about 34 times. Forward, the picture changes: consensus adjusted earnings of $19.32 for fiscal 2027 and $30.71 for fiscal 2028 put the shares at 17.7 and 11.1 times. Since September those estimates have barely moved, by +0.2% and +2.5% in our feed. The whole fall in the price is a fall in the multiple.
Stock pay. The adjusted figures exclude stock-based compensation, which we counted at about $7.6bn a year in September and which is rising in dollars. We subtract it in Part XI. At about $1.50–2 a share it is a real cost to owners, though a shrinking share of a fast-growing profit.
| What the screen shows | What is wrong, and the number to use |
|---|---|
| Enterprise value and market cap snapshot | Dated <b>2 November 2025</b>, a year old (EV $1.757tn at $362.55). Three different market caps elsewhere. Use about <b>$1.63tn</b> (4.77bn shares × $341.60) and an EV of about $1.66tn. |
| Segments: 'latest' year | Shows <b>fiscal 2018</b> as latest (a date-sorting error), and fiscal 2024 segments double-counted to $74.1bn. FY2024 revenue was <b>$51.6bn</b>; FY2025 $63.9bn. |
| Non-operating income, Q3 FY26: −$98m | Sign flipped: other income was <b>+$98m</b>. The same fault appears in our feed for Salesforce. |
| Dividends before the 2024 split: $4.10–5.25 | Not adjusted for the 10-for-1 split: $0.41–0.525. The dividend summary is empty; the trailing dividend is <b>$2.60</b>. |
| Analyst targets: low $350, no recent actions | Stops at 27 October; misses Redburn's downgrade to Neutral and <b>$398</b> target (13 Nov). |
| Owner earnings Q3: $14.58bn, above operating cash flow | An artefact of the method (negative 'growth capex'). Free cash flow was <b>$13.67bn</b>; do not cite '$2.98 a share' of owner earnings, as we did in September. |
The dividend
Covered three times, decided on 9 December — tested
Broadcom pays $0.65 a quarter, $2.60 a year, a yield of 0.8%. It raised the dividend by 10.2% last December, and the board decides the next rate with the results on 9 December. In September we said the arithmetic of a policy of paying out half of the prior year's free cash flow pointed to the largest raise in the company's history. We should be more careful. Management has described such a target in the past, but we could not find it in this year's filings, and the board has not applied it mechanically: dividends were about 57% of prior-year free cash flow in fiscal 2025 and about 46% in fiscal 2026. Tan said in September, as reported, that higher dividends and buybacks were the likely uses of cash and that prepaying cheap debt "makes little sense".
If the board did pay half of this year's estimated free cash flow of about $48bn, the dividend would roughly double, to about $1.26 a quarter. That is a hypothesis, not a forecast. A smaller raise would not be a bad sign in itself: a company that has just committed up to $115bn of support to one customer may sensibly keep more cash at home.
| Test | Evidence | Result |
|---|---|---|
| 1. Cover on free cash flow | $12.08bn of dividends against $39.40bn of free cash flow over twelve months: about 31%. | Pass |
| 2. The trend of that cover | Falling as cash flow grows faster than the dividend: ~57% of prior-year FCF in FY25, ~46% in FY26. | Pass |
| 3. From operations or from paper? | From operations; debt was being repaid ($10.5bn repaid, $4.5bn issued in nine months). | Pass |
| 4. Balance-sheet room | Net debt $35.4bn, about half a year of free cash flow; but up to ~$115bn of contingent support behind one customer. | ◆ Adequate, with a footnote |
| 5. What would force a cut | Free cash flow falling by about two-thirds, or a large call on the backstops in a bad year. | Unlikely |
| 6. The growth rate and its direction | +10.2% last December; the next decision on 9 December. | ◆ To be seen |
★ Valuation
The price pays for 2027; we still will not pay for 2028
Our method. For a business whose earnings are expected to more than double in two years, ten-year projections are fiction. We value Broadcom on its fiscal-2028 earnings, after subtracting stock pay, at a multiple that suits the kind of earnings they are; discount two years at 10%; and add the dividends received on the way (about $5.50 a share). In September we used a different frame, the bridge plus the promise, and reached about $430 for the whole. The two methods agree on the shape: most of the value lies in the AI line, and most of the uncertainty with it.
| Case | FY2028 earnings a share, after stock pay | Multiple | Value today |
|---|---|---|---|
| The promise stalls | ~$13.30 (AI ~$80bn) | 18× | ~$205 |
| 2027 lands, 2028 doesn't double ◆ | ~$20.10 (AI ~$130bn) | 20× | ~$340 |
| The guide lands | ~$28.60 (AI ~$230bn) | 18× | ~$430 |
The cases. "The promise stalls" assumes fiscal 2027 falls short, to about $80bn of AI revenue, as Google diversifies and the laboratories' financing tightens, and that 2028 is no better. "2027 lands, 2028 doesn't double" takes the fiscal-2027 guide and adds modest growth, about $22 of adjusted earnings a share in 2028. "The guide lands" takes the consensus of $30.71; we give it a lower multiple than the middle case, because earnings at that level would be more heavily AI hardware and more exposed to the next cycle.
The middle case moves to about $290 at 17 times and $390 at 23 times; to about $290 at $19 of earnings and $385 at $25; to $350 at an 8% discount rate and $325 at 12%. What is priced in? At $341.60, and 20 times, the market asks for about $22 of adjusted earnings a share in fiscal 2028: roughly the fiscal-2027 guide plus a modest step. The price already pays for next year's doubling and for some growth after it. It does not pay for the second doubling, and that is the right place for the market to stop.
Margin of safety. The price sits at our central value, which means there is none. Our zone of $280 is about 18% below the central case, about 14.5 times fiscal-2027 consensus, and close to the value of the middle case if the multiple fell to 17. We considered moving it. The case for moving it up is that the customer behind the financing turned out to be stronger than we said. The case for moving it down is that the financing is larger than we said. They roughly cancel, and a level that was nearly reached in a sector rout should not be moved because it was nearly reached.
Risks, lawsuits & controversies
Verified 18 November 2026 against the Q3 FY2026 10-Q, the 27 October 8-K and attributed reports
Legal. Tesco's claim in the UK High Court over VMware licensing is listed for trial between November 2027 and February 2028, according to The Register; the suit alleges breaches of contract over VMware licences. Fidelity sued in November 2025 over access to VMware software. In the EU, the cloud trade body CISPE complained in March 2026; the Commission was reported to have sent information requests in July, which Broadcom failed to have suspended in the General Court, but we found no formal proceeding. Japan's competition authority was reported to have closed its VMware inquiry in July without finding a violation. Broadcom's 10-Q says pending matters are not expected to be material.
Ratings. Fitch affirmed A− with a Positive outlook on 27 October (as reported); S&P rates Broadcom A−, and was reported in June to have called the first financing vehicle "credit negative". We did not confirm either at the agencies.
| Test | Today | What would change our mind |
|---|---|---|
| Q4 results, 9 Dec | AI $21.7bn guided | A miss on AI revenue, or any softening of the fiscal-2027 outlook, would move us towards the stall case. |
| The dividend, 9 Dec | $0.65 a quarter | A large raise says the board sees the cash as secure; a small one says it is keeping cash for the commitments. Neither changes our level. |
| The 10-K, December | Top five ~55% | Whether the two backstops add up, their fair value, and the customer concentration for the full year. |
| Marvell, 1 Dec | Custom ramp | A strong Google contribution at Marvell would confirm that Broadcom's largest customer is spreading its work. |
| Anthropic's listing | S-1/A filed | A successful listing strengthens the lessee; a struggling one would make the backstops the first thing to read. |
| The price | $341.60 | Below $280 with the fiscal-2027 outlook intact: own it. |
In September I told you Broadcom was a magnificent business at a price that asked you to trust a year that had not happened, and that I would own it below two hundred and eighty dollars. On 5 November it closed at two hundred and ninety-one. I did not buy, and I would like to explain why missing it by four per cent does not trouble me, and then to correct three things I told you.
The corrections first. I said the largest of Broadcom's future customers was a laboratory that loses money. Its prospectus, filed since, shows a company with thirty-three billion dollars of revenue in nine months and positive cash flow from operations. I said the top five customers were about forty per cent of revenue; the quarterly filing I was reading said fifty-five. And I quoted a thirty-billion-dollar Apple commitment that I cannot find in any filing, so I withdraw it. One of those errors flattered the risk and two hid it. I should have read the whole filing, not the page I was looking for.
Now what has changed. Broadcom has put up to about a hundred and fifteen billion dollars of support behind that one customer, if every commitment were added together and drawn in full: two guarantees on the value of the racks it leases, an obligation to accept forty-two billion dollars of its notes, and a billion-dollar purchase of its shares. None of that is a loss, and most of it may never be called. The customer owes about a hundred and sixty billion dollars of rent over five years, and its accounts suggest it can pay as long as it keeps growing very fast. Broadcom has become, in part, a lender to its own best customer. That can be a fine business. It is not the business I scored for twenty years.
Nothing else in the business has changed for the worse. The guidance stands, the AI line is doubling, Nvidia's results last night said demand for computing is still rising, and the fall in the shares since June is entirely a fall in the multiple. At three hundred and forty-two dollars you are paying about eighteen times next year's expected earnings, a fair price for a business of this quality if next year arrives. You are not yet paying for the year after, and I think that is right.
So I hold my level. Below two hundred and eighty, about fourteen and a half times next year's earnings, I would own Broadcom and accept the concentration as the price of the growth. Above it, I would wait for the fourth-quarter results and the dividend on 9 December, and for the annual report, which should tell us whether the two guarantees are one exposure or two. If Broadcom finances a second laboratory on the same terms, I will want a lower price, not a higher one.
— The Buffett Lens. A level set for good reasons is not improved by chasing the price that nearly reached it.
Own Below $280A wonderful business whose price already pays for next year's doubling, now carrying a lender's exposure to its largest customer. ★ At $341.60 the price is at our central value of ~$340, within a range of ~$205 to ~$430. Own below $280, about 14.5× fiscal-2027 consensus, unchanged from September. Results and the dividend on 9 December.
Articles that take this apart
Companies in FocusBroadcom or Nvidia: two ways to sell the same boomBoth are extraordinary businesses riding the same wave. They are not, however, the same investment — and the thing that separates them is a number you can count on one hand.
Companies in FocusIs Nvidia's growth sustainable? One line in the accounts is trying to tell youNvidia earned in six months what it took the whole of last year to earn. That is not the interesting part. The interesting part grew faster than revenue did.