NVIDIA
The company whose chips, networking and software most AI data centres are built on.
The scores
Growth3/4 ↑
Revenue up 106% and Data Center 117% with buyers broadening, but a quarter of next year leans on labs NVIDIA helps finance.
Business quality3/4 →
About 79% of the three makers' AI chip revenue on twenty years of software, but Broadcom's custom chips grow twice as fast.
Financial position3/4 ↓
Net cash and $69.9B of half-year free cash flow, but $366B of commitments and $108.5B of guarantees sit outside the balance sheet.
Execution record4/4 →
Eight straight beats, Q1 FY2027 revenue 4.6% above its guide, and a new platform each year since Hopper; the H20 write-off was an export ban.
Per share capture4/4 →
Share count falling, $39.8B of buybacks in six months, $235B more authorised, and deals paid in cash.
Potential2/4 →
A $1T market still rising, but FY2028's 70% rests on others' spending plans and NVIDIA-financed labs; little is signed beyond $15.6B of advances.
Risk profile2/4 ↓
OpenAI sits in revenue, holdings and a $105.0B guarantee at once; three customers are 44%; manufacturing runs through Taiwan; China is shut.
Valuation3/4 ↑
At 28.8 times trailing operating profit and 13.9 times FY2028's expected, it pays for visible profit; Q2 FY2027's margin may be the peak.
Evidence quality3/4 →
Revenue, margins, receivables and guarantees are filed and the market measured by Dell'Oro; the FY2028 demand and financing share rest on management alone.
The Verdict
We rate NVIDIA attractive. It sells the computers most AI is built on, its data-centre sales more than doubled in a year, and most AI software is written for its platform, so the business is solid, though custom chips from rivals are growing faster. The price is fair: at about 15 times next year's expected earnings per share, it is about right for profits already in view, with nothing for the boom lasting beyond that. The view hinges on how much of next year's growth NVIDIA is paying for itself, by lending to and guaranteeing its own buyers, and on cash arriving on time.
The long caseClouds and AI labs are building data centres faster than chips can be supplied, and most of them run on NVIDIA's computers, so its sales keep climbing.
The evidence under this - 6 claims
The largest buyers write their own software stacks and run custom chips at scale (C17, C18), so the switching cost binds smaller buyers more than the ones paying most.
Google, Amazon and Microsoft's in-house chips are not sold through Broadcom alone and appear in none of the three figures (C18), so custom share is understated.
Consensus FY2028 already assumes the 70% guide (C10, C30); if a quarter of it is financed demand that does not recur (C09), the true multiple is higher.
- The Wall Street Journal reported on August 27, 2026 that NVIDIA paused some revenue-sharing credit deals with AI clouds (C42), so the scheme is still changing.
- Management says every customer is investment grade or backed by one, and the SB Energy guarantee ends if OpenAI earns a satisfactory credit rating (C44).
The terms go to investment-grade buyers, and $15.6B of customer advances in deferred revenue (C46) shows other buyers paying ahead.
The company
Training and running an AI model takes enormous numbers of small calculations done at once. NVIDIA designs the chips that do this, called GPUs, and sells them packed into whole racks with its own processors, networking and software.
The buyers are cloud companies such as Amazon and Microsoft, AI labs such as OpenAI, newer AI clouds and big businesses. Data centres were 92.5% of its sales in Q2 FY2027. The rest is chips for gaming, design workstations and cars.
The evidence under this - 4 claims
OpenAI says its own Jalapeño chip serves more work per unit of power than existing hardware (C19).
How it makes money
NVIDIA designs the chips and pays TSMC to make them, then sells finished systems at a high price. In Q2 FY2027 it kept 75 cents of every sales dollar after the cost of making the product, and 66 cents after all its running costs.
A new, faster platform every year keeps customers buying. Memory is getting dearer, so management expects the 75 cents to dip to 71 or 72 before price rises restore it.
The evidence under this - 4 claims
- Dell'Oro says some 2026 spending may have been pulled forward ahead of price increases (C14), which would flatter the year before the guided one.
- FY2023's guide-to-downturn shows supply commitments can turn into $2.17B of provisions within a year (C37).
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 FY2027) | How it earns it |
|---|---|---|
| Data Center: Hyperscale | 50.6% | Rack-scale GPU systems, networking and software sold to the largest cloud providers for their own and rented AI work. |
| Data Center: AI Clouds, Industrial and Enterprise | 41.9% | The same systems sold to AI-focused clouds, AI labs, companies and governments, directly or through server makers. |
| Edge Computing | 7.5% | GPUs for gaming PCs and workstations and other computing outside the data centre, sold mainly through board and PC makers. |
Who runs it
Jensen Huang co-founded NVIDIA in 1993 and still runs it; Colette Kress has run finance since 2013. Insiders own about 4%, nearly all Huang's. His bonus rides on sales, his shares on profit and the stock, none on collecting cash.
The evidence under this - 3 claims
- The FY2026 stretch goals were cut automatically after the H20 export controls, revenue from $190.0B to $160.0B and non-GAAP operating income from $120.0B to $96.0B, so the maximum was reached against lowered bars.
- No metric pays on cash conversion or return on capital, so the receivables build in C06 costs the plan nothing.
Why it wins, and who could beat it
Why customers come back
Since 2006 NVIDIA has given developers free software, called CUDA, for programming its chips, and more than 7.5 million use it. Most AI software is written for it. A rival chip has to be as fast and also persuade those developers to rewrite and retest their work.
That shows up in share. NVIDIA's data-centre sales were about 13 times those of AMD, its nearest rival selling the same kind of chip, and an outside research firm says spending stayed concentrated on NVIDIA.
The evidence under this - 5 claims
The largest buyers write their own software stacks and run custom chips at scale (C17, C18), so the switching cost binds smaller buyers more than the ones paying most.
Google, Amazon and Microsoft's in-house chips are not sold through Broadcom alone and appear in none of the three figures (C18), so custom share is understated.
The threat
The biggest buyers can design their own chips. Broadcom builds custom chips for the large clouds, and its AI chip sales grew 221% in a year against NVIDIA's 117%. Google now sells its own chips to outside customers, and OpenAI, NVIDIA's biggest financed customer, will start using its own chip in small numbers at the end of 2026.
The evidence under this - 4 claims
OpenAI is also NVIDIA's largest financed customer, with commitments of about 12 GW and a $105.0B guaranteed site (C44), so Jalapeño adds to its NVIDIA purchases for now.
Google, Amazon and Microsoft's in-house chips are not sold through Broadcom alone and appear in none of the three figures (C18), so custom share is understated.
The rivals: AVGO, GOOG, AMD, AMZN, MSFT, and 2 more that are private, in-house or listed abroad.
Who else does this, rival by rival
| Ticker | Rival | How far along | Priced against it |
|---|---|---|---|
| AVGO | Broadcom's custom accelerators for the large cloudsPriced against it: Same product | On sale | Same product |
| GOOG | Google's TPU systemsPriced against it: One segment only | On sale | One segment only |
| Private | OpenAI's Jalapeño chip, built with BroadcomPriced against it: Private | Announced | Private |
| AMD | AMD Instinct GPUsPriced against it: Same product | On sale | Same product |
| AMZN · MSFT | Amazon's Trainium and Microsoft's Maia chipsPriced against it: One segment only | On sale | One segment only |
| Private | Huawei's Ascend chipsPriced against it: Private | On sale | Private |
How big it could get
How big this could get
Dell'Oro, a research firm, measured worldwide spending on data centres up 57% in 2025 and up 92% in Q2 2026, and expects more than $1T this year, much of it on well over 10 million high-end AI chips.
Part of that growth is memory getting dearer rather than more computers being built. Nobody independent publishes how much computing each buyer installs, so the market can be bounded but not counted.
The evidence under this - 2 claims
Part of the dollar growth is memory and storage prices lifting server prices, not more computing (C15).
Who has signed
Few buyers have signed for fixed amounts. Customers have paid $15.6B in advance. Meta agreed a multiyear partnership for millions of chips without a number attached. The biggest promise runs the other way: NVIDIA guarantees up to $105B on an Ohio data-centre campus leased to an OpenAI company, which will use NVIDIA's computers.
The evidence under this - 4 claims
Dell'Oro warns that this level of investment raises the potential for overcapacity (C14).
Where the growth is coming from, business by business
Where the growth is coming from
| Business | Share of revenue | Growth: last quarter | Last half-year | Last year | Share in 3 years at its half-year rate |
|---|---|---|---|---|---|
| Data Center | 93% | +116.6% | +104.8% | +89.5% | 98% |
| Rest of the company | 7% | +27.5% | +28.0% | +34.9% | 2% |
| Whole company | 100% | +105.9% | +95.8% | +83.4% | 100% |
On the filed split through 2026-07-26, if each business keeps its last half-year's growth, the company grows about 99% next year and 102% in year 3, against 83.4% over the last twelve months. A scenario, not a forecast: no business holds its rate.
How we read it: Data Center is 92.5% of revenue and the blend is essentially that business; the score believes the Data Center rate, and caps it for the financed share, not for the mix.
The results
The numbers
Sales grew 83% over the twelve months to Jul 2026, faster than the 65% in the year to Jan 2026.
Operating margin is 65%, up from 54% in the year to Jan 2024.
Running costs take 9¢ of each sales dollar, down from 28¢ in the year to Jan 2022.
Each dollar put into the business earns 89% a year after tax, up from 32% in the year to Jan 2022.
It turned 42% of sales into free cash, $127B over the last twelve months.
It holds $56.6B of cash against $33.4B of debt, $23.2B more cash than it owes.
The share count fell 3.1% in 3.5 years, and pay in shares costs 2.4% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-04. Fiscal years end in January.
| Jan 2022 | Jan 2023 | Jan 2024 | Jan 2025 | Jan 2026 | 12 mo to Jul 2026 | |
|---|---|---|---|---|---|---|
| Size and growth | ||||||
| Sales | $26.9B | $27.0B | $60.9B | $130B | $216B | $303B |
| Sales growth | +61% | +0.2% | +126% | +114% | +65% | +83% |
| Profit | ||||||
| Gross margin | 65% | 57% | 73% | 75% | 71% | 75% |
| Operating margin | 37% | 16% | 54% | 62% | 60% | 65% |
| Net margin | 36% | 16% | 49% | 56% | 56% | 64% |
| What growth costs, per sales dollar | ||||||
| Research and development | 20% | 27% | 14% | 9.9% | 8.6% | 7.8% |
| Selling, general and admin | 8.0% | 9.0% | 4.4% | 2.7% | 2.1% | 1.7% |
| Pay in shares | 7.4% | 10% | 5.8% | 3.6% | 3.0% | 2.4% |
| Capital spending | 3.6% | 6.8% | 1.8% | 2.5% | 2.8% | 2.4% |
| Cash | ||||||
| Cash from operations | $9.11B | $5.64B | $28.1B | $64.1B | $103B | $134B |
| Free cash flow | $8.13B | $3.81B | $27.0B | $60.9B | $96.7B | $127B |
| Balance sheet | ||||||
| Cash and short-term investments | $21.2B | $13.3B | $26.0B | $43.2B | $49.7B | $56.6B |
| Debt | $10.9B | $11.0B | $9.71B | $8.46B | $8.47B | $33.4B |
| Net cash | $10.3B | $2.34B | $16.3B | $34.7B | $41.2B | $23.2B |
| Returns and shares | ||||||
| Return on capital | 32% | 12% | 68% | 101% | 87% | 89% |
| Shares (diluted) | - | 25.1B | 24.9B | 24.8B | 24.5B | 24.3B |
What growth costs
Growth keeps making NVIDIA more profitable: research spending rises, but sales rise far faster, so each new sale costs less to win. A flat year reverses that, as the last downturn showed.
The evidence under this - 2 claims
- In FY2023, when revenue stalled, R&D rose to 27.2 cents per sales dollar and operating margin fell to 15.7% from 37.3% (C37), so the leverage runs both ways.
- Operating expenses rose 10% to 11% in Q2 FY2027 alone, on computing infrastructure and pay (S06).
The last report (26 Aug)
Quarter ended July 2026, filed figures.
On the surface Q2 FY2027 was another record. Sales rose 106% to $96.2B, data-centre sales 117%, and NVIDIA kept 66 cents of each sales dollar as operating profit. It expects $108.0B next quarter.
Underneath, two things changed. $7.8B of profit was gains on shares NVIDIA owns in other companies, not chip sales. And cash from operations fell to $24.1B from $50.3B the quarter before, because customers were given longer to pay: up to a year for the strongest buyers.
What it did to the thesisIt proved demand is still rising and that NVIDIA can raise prices into it. It left open how good the profit is, because more of it is arriving as money owed and as paper gains, and the next report has to show the cash catching up.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $2.22 against $2.09 expected; $2.46 as filed, the gap being what the company adds back.
The evidence under this - 5 claims
The terms go to investment-grade buyers, and $15.6B of customer advances in deferred revenue (C46) shows other buyers paying ahead.
What management did with the money
Spare cash mostly buys back shares, $39.8B in six months; about $11.9B in cash is going on Hugging Face, home of open AI models.
Insiders in 2026Q3: 3 purchases and 34 sales, from their Form 4 filings.
The evidence under this - 2 claims
The price, and where we disagree
In short. Fair. The price pays for profits NVIDIA is already earning and the growth already guided for next year, and assumes almost nothing after that.
Over the past year sales per share rose 85.2% while the price buyers pay for each dollar of sales fell 32.5%, so the shares rose only 24.9%.
The test is close. NVIDIA reports Q3 FY2027 on 17 November 2026, and that report is the first to show whether customers are paying on time and whether the financing of AI labs is growing.
The evidence under this - 5 claims
- The Wall Street Journal reported on August 27, 2026 that NVIDIA paused some revenue-sharing credit deals with AI clouds (C42), so the scheme is still changing.
- Management says every customer is investment grade or backed by one, and the SB Energy guarantee ends if OpenAI earns a satisfactory credit rating (C44).
How we read the multiple
We measure the price against operating profit, the money the business makes before tax and before gains on shares it holds, because those gains have swelled reported profit.
- Buyers pay about 29 dollars for each dollar of last year's operating profit, debt included.
- At last quarter's pace that falls to about 22 dollars.
- On the operating profit analysts expect in FY2028, it is about 14 dollars.
- Forecasters themselves have profit flat or falling after FY2030, so the price assumes the boom fades.
Against its own history. Far below what buyers paid over the past six years, but those years had much smaller profits, so part of the gap is the market doubting today's profits will last.
The evidence under this - 5 claims
Q2 FY2027 sits at a 66.2% operating margin, the highest the company has reported, so a year at that rate assumes the peak holds.
Consensus FY2028 already assumes the 70% guide (C10, C30); if a quarter of it is financed demand that does not recur (C09), the true multiple is higher.
Priced against its rivals
There is no group to price it against. No filed cohort to price against; against its own six-year history the multiple sits well below the median. The price is judged against its own history instead.
Private or foreign rivals that cannot be priced: Huawei Technologies Co. Ltd., Alibaba Group and Baidu, Inc., Cerebras, SambaNova and other AI chip start-ups.
Where we disagree with the market
NVIDIA is the indispensable supplier of the AI boom, its results keep beating forecasts, and the shares are cheap for the growth.
The shares are cheap only if next year's growth is real demand; we think most of it is, but NVIDIA is financing part of it.
The market argues about whether the boom lasts, but the price already assumes it fades after FY2028. The question it skips is how much of FY2028 NVIDIA is paying for: management says about a quarter of next year's business comes from AI labs it helps finance.
Against us: an outside research firm warns of too many data centres being built and some spending brought forward to beat price rises.
What analysts say, and the evidence
The evidence under this - 6 claims
Consensus FY2028 already assumes the 70% guide (C10, C30); if a quarter of it is financed demand that does not recur (C09), the true multiple is higher.
- The Wall Street Journal reported on August 27, 2026 that NVIDIA paused some revenue-sharing credit deals with AI clouds (C42), so the scheme is still changing.
- Management says every customer is investment grade or backed by one, and the SB Energy guarantee ends if OpenAI earns a satisfactory credit rating (C44).
The terms go to investment-grade buyers, and $15.6B of customer advances in deferred revenue (C46) shows other buyers paying ahead.
Dell'Oro warns of overcapacity and pull-forward (C14), and a quarter of next year depends on labs NVIDIA finances (C09).
What could go wrong, and how we would know
The risks
- A few buyers Three customers were 16%, 15% and 13% of half-year sales, and OpenAI is at once a customer, an investment and the tenant behind a $105B guarantee.
- Customers building their own Broadcom's custom chips are growing faster than NVIDIA's, Google now sells its chips outside, and OpenAI starts its own in late 2026.
- A downturn hits hard In FY2023 sales went flat and NVIDIA wrote off $2.17B of stock and supply promises; today it has promised $279B to suppliers.
- China is closed The outlook assumes no data-centre sales in China, which export rules have shut, and Huawei fills that market.
- One place makes everything Its chips are made mainly by TSMC in Taiwan, so a disruption there would stop supply.
- Regulators are looking The Justice Department is reported to be examining NVIDIA's Groq licensing deal.
The evidence under this - 7 claims
What we still do not know
- How much of next year's sales comes from buyers NVIDIA lends to, invests in or guarantees; management gives only a rough quarter.
- Which of its revenue-sharing deals with AI clouds were paused in August, and whether that changes next year's demand.
- How many custom chips the big clouds are installing instead of NVIDIA's; no free source counts units.
- Whether independent tests confirm management's claims for its newest chips.
The evidence under this - 4 claims
- The Wall Street Journal reported on August 27, 2026 that NVIDIA paused some revenue-sharing credit deals with AI clouds (C42), so the scheme is still changing.
- Management says every customer is investment grade or backed by one, and the SB Energy guarantee ends if OpenAI earns a satisfactory credit rating (C44).
Google, Amazon and Microsoft's in-house chips are not sold through Broadcom alone and appear in none of the three figures (C18), so custom share is understated.
OpenAI says its own Jalapeño chip serves more work per unit of power than existing hardware (C19).
What would change our view
- Attractive→NeutralThe price runs ahead of next year's profit
It needs:
- ○Buyers pay about 41 dollars for each dollar of last year's operating profit, debt included, about 20 times what analysts expect for FY2028.about ≈$336.30 today, 44% above the last close
- Attractive→NeutralData-centre demand turns down
It needs any of:
- ○Data-centre sales fall from one quarter to the next, after rising 18.3% in Q2 FY2027.checked at the next results, due by 17 Nov
- ○NVIDIA keeps less than 70 cents of each sales dollar after making the product, from 75 cents.checked at the next results, due by 17 Nov
- Attractive→NeutralGrowth that NVIDIA pays for
It needs any of:
- ○Customers take more than 75 days on average to pay, up from 60.checked at the next results, due by 17 Nov
- ○NVIDIA's guarantees for customers' data centres rise above $150B, from $108.5B.checked at the next results, due by 17 Nov
We would also be wrong if
- Sales grow less than 40% in FY2028, against the roughly 70% guided.
What is coming
- 21 October 2026 GTC Berlin keynote by Jensen Huang: Whether the Vera Rubin ramp and the yearly platform cadence hold as described; a slip would test the switching-cost argument that rests on staying a generation ahead.
Every dated event ahead
- 21 October 2026 GTC Berlin keynote by Jensen Huang: Whether the Vera Rubin ramp and the yearly platform cadence hold as described; a slip would test the switching-cost argument that rests on staying a generation ahead.
- 17 November 2026 Q3 FY2027 results: Revenue against the $108.0B guide, gross margin against 74%, receivable days against 60, Rubin near a fifth of Data Center, and the 70% FY2028 outlook kept or cut.
- February 2027 Q4 FY2027 results and the FY2027 10-K: Whether gross margin bottoms at 71% to 72% as guided, Hyperscale reaccelerates on Rubin supply, and the FY2028 outlook survives a full year of memory costs.
- June 2027 Hugging Face acquisition closes: Whether the largest cash deal since Mellanox closes on its terms while regulators are already examining the Groq licence.
Since we wrote this
- 28 OctNewsThe Justice Department is asking Nvidia how buying Hugging Face could affect rivals. View heldDOJ asks Nvidia about its $12.9B Hugging Face purchase.
- 9 OctNewsOpenAI's revenue run rate is $50B, not the $68B that circulated; AI stocks fell. View heldOpenAI told investors its run rate is about $50B, below the $68B figure that circulated.
- 7 OctNewsSpaceX reportedly seeks $40B of debt for Nvidia chips; only press reports exist. View heldFT reports SpaceX seeks $40B of debt to buy Nvidia chips; reporting only, no primary source.
- 6 OctNewsGroq's stockholders sue Nvidia over the $20B deal; buyback coverage repeats a known step. View heldGroq stockholders sue Nvidia over the $20B Groq deal; coverage of the larger buyback authorization.
- 5 OctThesis reviewThe long case now sits under the verdictAfter the five-name pilot, the owner asked for the long case on every thesis.
- 4 OctThesis writtenAttractive, medium convictionWritten six weeks before Q3 FY2027 results, after the 70% FY2028 guide, the SB Energy guarantee and a slower quarter for cash, with the shares at $233.95.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the proxy and filings.
Also checked 6 times with nothing new to act on.
Sources 28
Sources - 28
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-02-25Item 1 (platform, segments, named competitors, CUDA developers, R&D since inception, workforce), suppliers, FY2026 gross margin and the H20 chargebacksC01C03C21C45C47C52C53C56C57
- S02primarycompany release2026-08-26revenue by platform, gross margin, operating income by quarter, other income, EPS on both bases, the Q3 FY2027 outlookbacksC02C03C04C05C11C22C29C33C34C35C38
- S03primarycompany release2026-02-25Q3 and Q4 FY2026 operating income, FY2026 Data Center revenue, the Q1 FY2027 outlook, the Meta partnership, capital returnedbacksC01C04C34C38C48
- S04primarycompany filing2026-08-26customer and receivable concentration, payment terms, commitments, guarantees, equity investments, deferred revenue, share count, buybacks, debt, geographybacksC06C07C08C23C24C25C33C44C45C46C49C56
- S05primarycompany release2026-08-26Hyperscale and ACIE revenue, days sales outstanding, cash flow, free cash flow, commitments table, guarantees, cash and debtbacksC02C05C06C08C24
- S06primarytranscript2026-08-26Q2 FY2027 earnings call transcript (held 2026-08-26)data/cache/transcripts/NVDA-2027Q2.jsonthe FY2028 outlook, gross margin path, AI lab financing and its share of next year, hyperscaler capital spending, revenue per gigawatt; analysts from Morgan Stanley, Bernstein, BofA, UBS, Goldman Sachs and others
We expect to grow revenue by approximately 70% in fiscal 2028. This is a supply-constrained outlook.
backsC09C10C12C33C36C41C55C58 - S07secondaryvendor data2026-10-04Research pack (FMP market tier and filed statements), frozen at content/NVDA/research-pack.jsoncontent/NVDA/research-pack.jsonprice, market value, enterprise value, returns and their decomposition, six-year valuation history, consensus estimates and history, price-target drift, earnings history, insider statisticsbacksC26C27C28C29C30C31C32C38
- S08primaryindustry data2026-03-17worldwide data-centre capital spending up 57% in 2025, more than $1T forecast for 2026, more than 10M high-end accelerators, the overcapacity warning
We forecast full-year 2026 data center capex to surpass $1 trillion. The trillion-dollar threshold marks a major milestone, supported by well over 10 M high-end accelerators as the primary capex driver
backsC13C14 - S09primaryindustry data2026-06-10the 2026 outlook raised to more than $1T, the top four US clouds up 78%, the pull-forward caution
While near-term demand remains healthy, some spending may have been pulled forward ahead of expected price increases later this year
backsC13C14 - S10primaryindustry data2026-09-16Q2 2026 spending up 92%, where it went, memory prices lifting server prices, neoclouds and AI model builders growing fastest
Spending remained concentrated in NVIDIA Blackwell Ultra and hyperscaler custom accelerators
backsC15 - S11primarycompany filing2026-08-05AMD's Data Center segment revenue and operating income for Q2 2026backsC16C22
- S12primarycompany filing2026-02-04the Instinct GPU line and its buyers, the OpenAI agreement for 6 GW of AMD GPUsbacksC16
- S13primarycompany release2026-09-02Broadcom's AI semiconductor revenue, mostly custom accelerators and networking for hyperscalers
Q3 AI semiconductor revenue of $16.7 billion grew 221% year-over-year, and 54% quarter-over-quarter
backsC17C22 - S14primarycompany filing2026-07-23Alphabet selling TPU systems to outside customers, recognised mainly in 2027
agreements with a limited number of customers to sell TPU systems
backsC18 - S15secondarynews2026-08-26OpenAI's account of its Broadcom-built Jalapeño chip and Richard Ho's deployment timing, quoted from OpenAI's report and talk
Jalapeño can serve more AI work per unit of power, while also returning responses more quickly.
backsC19 - S16primarycompany release2026-09-28the buyback increase and the period it is expected to run
authorized a $150 billion increase to the share repurchase program, raising the remaining total program to $235 billion
backsC23 - S17primarycompany filing2026-09-03price, retention pool, closing period and the commitment to keep the platform open to other chipsbacksC25
- S18primarycompany filing2026-10-04open-market sales and purchases in the year, the Rule 10b5-1 flag on eachbacksC26
- S19primaryregulator2026-09-15short interest in NVDA and its share of floatbacksC32
- S20secondaryvendor data2026-10-04closing prices around each repricingbacksC34C35C36
- S21secondarynews2025-01-28the objection stated in January 2025, from the article summary in FMP's news feed
concerns over a low-cost artificial intelligence model from Chinese startup DeepSeek stripped 17% off its shares, or $593 billion from its market value
backsC34 - S22secondarynews2026-03-30the objection stated in March 2026, from the article summary in FMP's news feed
Nvidia, the world's most valuable company, finds itself trading at its cheapest price-to-earnings multiple since before ChatGPT kicked off the AI boom
backsC35 - S23primarycompany filing2023-02-24the last downturn: Data Center and Gaming revenue, inventory provisions, supply commitmentsbacksC37C56
- S24secondarynews2026-08-27Reuters, citing the Wall Street Journal: Nvidia pauses revenue-sharing deals with AI cloud companiesthe reported pause, from the article summary in FMP's news feed
Nvidia has paused some deals in a new financing initiative that offered credit support to AI cloud companies in exchange for a share of revenue
backsC42 - S25secondarynews2026-08-14the guarantee as first planned, from the article summary in FMP's news feed
Nvidia has downsized plans for a $250 billion guarantee for an OpenAI data center
backsC42 - S26secondarynews2026-09-09the reported antitrust review of the Groq licence, headline onlybacksC43
- S27primarycompany filing2026-08-17the guarantee cap, the capacity and the OpenAI affiliate as tenantbacksC42C44
- S28primarycompany filing2026-05-12beneficial ownership as of March 23, 2026, the pay plan's metrics and goals, the FY2026 payouts and the automatic goal cut after the H20 ban, related-person employmentbacksC50C51C52