Q2 fiscal 2027 is the quarter the company’s own part in its buyers’ financing comes into the filings in full. Data Center revenue was of total revenue of : Hyperscale and ACIE , read for their AI part at and . A company moved from ACIE to Hyperscale, so the sub-markets are on a new definition from this quarter.
Behind the buyers the company stands with its own balance sheet. By the call it had invested nearly in frontier labs, a cumulative amount; at the quarter end it held equity investments of with committed, had committed to purchase capacity from AI clouds that procure its systems, and had extended payment terms for certain investment-grade customers. After the quarter end, in August 2026, it entered into credit support capped at for leases to an OpenAI affiliate, effective as the leases commence from fiscal 2029. The frontier lab channel reads mixed, as in the prior quarters, since no payment in the quarter has the company’s money as its stated source; the hyperscale channel reads mixed as before, and a capacity commitment channel opens with no spend until fiscal 2028. The CFO expects labs backed this way to contribute a share of next year’s business, a forward statement; the ledger sizes them at this quarter.
On the cost side, supply commitments of , mostly memory, and memory prices the CFO ties to the AI build-out make the cost of the AI systems sold, , partly a price effect. Research cloud capacity reads at and the engineering productivity the CFO says AI tools have already delivered at , both the ledger’s own sizes; the AI tools bill itself has no seat count or usage volume and is left unsized. Equity gains of are non-operating.
Sized channels against the income statement, Q3 2026
11 of 13 channels sized
Each blue mark is one channel's dollars for the quarter; a bar is the range of an estimate. Grey marks are the company's reported lines. The distance between them is the point: how large the AI channel is next to the line it sits in.
Revenuereported line
$96.22bn
Cost of revenuereported line
$24.08bn
Total operating expensesreported line
$8.41bn
Research and developmentreported line
$7.05bn
Sales, general and administrativereported line
$1.35bn
Data center AI infrastructure bought by the hyperscale clouds and the largest consumer internet companiesrevenue in · our inference
Data center AI infrastructure bought by AI clouds, AI model makers, sovereigns, industrial and enterprise customers (ACIE)revenue in · our inference
Cost of the AI systems sold: wafers, memory, packaging and system assemblyspend · our inference
Compute bought by and for the frontier AI labs, directly and through cloudsrevenue in · our inference
Sovereign AI: AI infrastructure bought by or for national governments, mostly through regional cloudsrevenue in · our inference
Cloud capacity bought from cloud providers for the company’s own research and modelsspend · our inference
Gains and losses on equity stakes in AI model makers, AI clouds and other ecosystem companiesrevenue in · our inference
Physical AI: compute for autonomous vehicles and robots, in the data center and at the edgerevenue in · our inference
AI workstations and deskside systems (RTX PRO workstations, DGX Spark, DGX Station)revenue in · our inference
Amortization of the Groq inference technology licence and the hired teamspend · our inference
Engineering cost avoided through the use of AI toolscost displaced · our inference
$1mn$10mn$100mn$1bn$10bn$100bn
AI channel, dollars for the quarter low to high of an estimate reported lineLog scale: each gridline is ten times the one before.
New money and old money, Q3 2026
2 new9 expanded2 relabelled
Each channel is tagged once for whether its money existed before language models, from the company's annual report and call at the start of the period. A bar splits one flow's sized dollars by that tag. The incremental total is the part that would not be there without the models: a new channel counts in full, an expanded one only for what AI added, a relabelled one at zero.
Flow, sized total
Split by novelty
Incremental total
Paid for AI$19.84bn to $26.26bn sized
new not sizedexpanded $19.84bn to $26.26bn
Incremental total $100mn to $26.26bnpoint $125mn$23.10bn in 2 channels has no traced baseline
Cost displaced by AI$8.5mn to $317mn sized
expanded $8.5mn to $317mn
Incremental total $8.5mn to $317mnpoint $85mn
Revenue arriving through AI$77.34bn to $87.98bn sized
new not sizedexpanded $77.08bn to $86.37bnrelabelled $259mn to $1.61bn
Incremental total $0 to $86.37bnpoint $0$83.72bn in 2 channels has no traced baseline
The sized total counts every channel the company credits to AI, including relabelled money that existed before language models and the ledger's own estimates for it. The incremental total counts relabelled channels at zero. A flow is split by layer where its dollars sit at more than one: end use, compute sold to builders, and hardware. The same dollar can be a buyer's spend, a cloud's revenue and a chipmaker's revenue, so the layers are never added together.
Paid for AI5 channels · $19.84bn to $26.26bn sized · $100mn to $26.26bn incremental · 2 not sized
research
Cloud capacity bought from cloud providers for the company’s own research and models
2.6% to 4.7% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: quantified· motive: exploratory· before LLMs: expanded
Cloud service agreement commitments were , with payable in the rest of fiscal 2027; they and the data center leases power research, including the open models (claim c28). Compute infrastructure rose within research and development, which grew to (claim c25). The schedule’s remaining fiscal 2027 entry fell from in the Q1 10-Q to , so about was paid in the quarter before the of commitments added in it; the size, , builds on that difference. The company’s own compute build is not registered as a channel: the 10-Q gives its capital expenditure commitments as data center equipment for engineering and manufacturing and its leases not yet commenced as being for engineering, product design and testing of chips and systems (claims c53, c54), and the release names the open models only beside that work (claim c28), so no source attributes the build to AI on its own. The cloud service agreements are attributed to research on the open models (claim c55).
Evidence: 5 quotes, 9 figures, 2 confounds, 2 from before coverage
Multi-year cloud service agreements under which the company rents accelerated capacity, largely its own systems, from cloud providers to develop its chips, systems and software and to train its open models (Nemotron, Cosmos, GR00T) and autonomous vehicle software. Its cost sits in the compute and infrastructure cost the filings give as a driver of research and development, beside the company’s own data center equipment and leases, which the filings attribute to engineering and manufacturing rather than to AI and which are not registered as a channel. The same cloud providers are customers, so this is capacity bought back from buyers of the hardware; the filing says capacity may be reduced, terminated or sold to others by the providers. The quarter’s expense is not given: payments stand in for it, from the scheduled payments and, where successive schedules allow, their change between filings. The layer is end-use: the capacity is the company’s own use, for its research and its open models; the same dollars are compute revenue at the cloud providers.
Why this motive
Research framing: capacity for chips, systems and the company’s open models (claim c28).
Before LLMs: expanded
The activity existed at the anchor: the company had of multi-year cloud service agreements, primarily for research and development, beside its DGX Cloud offering hosted at cloud providers. No quarter of the expense was given, so no baseline is traced. The size is the whole of an activity that existed before.
“Other non-inventory purchase obligations were $4.6 billion, which includes $3.5 billion of multi-year cloud service agreements, primarily to support our research and development efforts.”
“We have partnered with CSPs to host such software and services in their data centers, and we entered and may continue to enter into multi-year cloud service agreements to support these offerings and our research and development activities.”
Cloud service agreement commitments, total · as-of 2026-07-26
Cloud service agreement payments in the remainder of fiscal 2027 (two quarters) · 2026-07-27..2027-01-31
Cloud service agreement payments scheduled for the remainder of fiscal 2027 (three quarters) · 2026-04-27..2027-01-31
Multi-year cloud service agreement commitments · as-of 2026-04-26
Cloud service agreement payments for fiscal 2027 that left the schedule during the quarter (a floor on payments in the quarter) · 2026-CQ3
Cloud service agreement commitments added during the quarter, net (change in the total plus the payments that left the schedule) · 2026-CQ3
Compute infrastructure within research and development, year-over-year growth · 2026-CQ3
Research and development · 2026-CQ3
Research and development, year-over-year growth · 2026-CQ3
What else could explain it
line composition: Compute infrastructure also includes the company’s own data centers and leases.
other: The quarter’s expense is not disclosed; payments stand in for it.
Quotes
“The increases in research and development expenses for the second quarter and first half of fiscal year 2027 were primarily driven by a 127% and 120% increase in compute infrastructure, respectively, and a 30% increase in each fiscal year 2027 period in compensation and benefits, including stock-based compensation, reflecting employee growth and compensation increases.”
“Our cloud service agreements and data center lease commitments together provide the physical and cloud infrastructure that powers our research and development — from the engineering, product design, and testing of our compute chips, networking products, and systems, to the development of our open models, such as NVIDIA Nemotron™, NVIDIA Cosmos™, and GR00T, and our autonomous vehicle software.”
“Capital expenditures – Our capital expenditures primarily include obligations for data center equipment and infrastructure used for engineering and manufacturing operations.”
“Data center leases not commenced – These leases will be primarily used for engineering, product design, and testing of our compute chips, networking products, and systems.”
“Cloud service agreements – These commitments provide the cloud infrastructure to support our research and development of our open models, such as NVIDIA Nemotron, Cosmos, and GR00T, and our autonomous vehicle software.”
Capacity commitments to AI clouds that purchase the company’s systems (AI cloud agreements)
Not sized
The commitments begin in fiscal 2028: the schedule carries no payment for the remainder of fiscal 2027, so the quarter has no spend on this channel. It is left unsized rather than sized at zero, and the commitment is recorded as metrics.
described, no sizedisclosure: quantified· motive: channel-defensive· before LLMs: expanded· layer: compute
Under AI cloud agreements introduced this quarter, the clouds procure the company’s systems and it commits to purchase cloud services from them, in total, typically lasting years, falling as third parties use the capacity (claims c18, c19); the CFO calls it a take-or-pay commitment and minimum revenue floor that lets lenders finance the cloud (claim c4). The schedule shows no payments in the rest of fiscal 2027 and in fiscal 2028. It is the company committing to take back, in advance, part of what its buyers purchase.
Evidence: 5 quotes, 3 figures, 1 confound, 2 from before coverage
Under agreements introduced in fiscal Q2 2027, AI clouds purchase the company’s data center systems and the company commits to purchase cloud services from them, which they may stop providing and offer to third parties at better rates; the company uses what it takes for research and may share in the clouds’ revenue above a floor. It is a take-or-pay backstop on part of a buyer’s capacity: a cost to the company that exists so the buyer can finance and purchase the hardware. Payments begin in fiscal 2028. The layer is compute: the commitment backs capacity the clouds offer to builders, and the 10-Q says it falls as third parties use that capacity or as the company uses it for its own research, the other use.
Why this motive
A commitment taken so that buyers can finance the company’s systems, in exchange for a share of their revenue (claims c4, c18): spending to keep the platform where independent clouds are being built. The closest tell in the table; no tell fits exactly.
Before LLMs: expanded
Buying cloud capacity from providers that host the company’s systems predates the agreements: of multi-year cloud service agreements at the anchor. Commitments tied to a buyer’s purchase of the systems, with a revenue share, are not in the anchor. The size is the change AI made, not the whole line.
“Other non-inventory purchase obligations were $4.6 billion, which includes $3.5 billion of multi-year cloud service agreements, primarily to support our research and development efforts.”
“We have partnered with CSPs to host such software and services in their data centers, and we entered and may continue to enter into multi-year cloud service agreements to support these offerings and our research and development activities.”
AI cloud agreements: committed purchases of capacity from AI clouds, total · as-of 2026-07-26
AI cloud agreements: payments in fiscal 2028 · FY2028
Typical duration of the AI cloud agreement commitments, years · as-of 2026-07-26
What else could explain it
other: Capacity taken may be used for the company’s own research, which would otherwise sit in the research cloud channel.
Quotes
“Rather than allocating their entire capacity to a single long-term offtake guarantee that lenders typically require to finance a data center independently, we have introduced a revenue-sharing structure. NVIDIA provides a take or pay commitment on a portion of the facility's capacity, a minimum revenue guarantee that gives lenders the confidence to underwrite the project, and in exchange, we share in a portion of the neocloud's revenue earned above that floor.”
“Under these agreements, AI clouds procure our data center infrastructure products and we commit to cloud service agreements, which the AI clouds can unilaterally stop providing to us and sell to third-party customers at more advantageous rates.”
“Our commitments, which are typically six years in duration, totaled $36 billion as of July 26, 2026, and decrease as capacity is used by third-party customers or by us for our research and development efforts.”
“In this model, we get paid twice, once on the hardware sale and again through the share of rental revenue, a highly reoccurring stream layered on top of a one-time equipment purchase.”
AI coding agents and model usage in the company’s own engineering
Not sized
The CFO names a further increase in AI tool use among the reasons operating expenses grow and the CEO encourages employees to rent cloud AI services, with no amount, seat count or usage volume (claims c26, c27). The former ballpark multiplied a headcount by a judgment adoption share and a judgment price; a count of employees with no measured usage volume may not size the bill (the methodology rules for counts and for volume times price).
described, no sizedisclosure: direction only· motive: exploratory· before LLMs: new
The CFO names a further increase in AI tool use among the reasons operating expenses grow (claim c26); the CEO says the company rents intelligence and encourages employees to use cloud AI services as much as they can (claim c27). No amount, seat count or usage volume is given, so the channel is left unsized; the company has roughly employees.
Evidence: 2 quotes, 2 figures, 1 confound, 1 from before coverage
What the company pays for the coding agents its engineers use (management names Claude Code, OpenAI Codex and Cursor) and for other AI tools staff use. It sits inside research and development; the CFO names rising use of AI tools as a reason operating expenses grow, without an amount. Part of the use may run on capacity the company rents or owns.
Why this motive
Use of AI tools increasing further, said to enhance engineering productivity, with no measure (claim c26).
Before LLMs: new
The anchor has no outside AI tool bill. Research and development was in fiscal 2024, with employees in research and development.
“As of the end of fiscal year 2024, we had approximately 29,600 employees in 36 countries, 22,200 were engaged in research and development and 7,400 were engaged in sales, marketing, operations, and administrative positions.”
Employees engaged in research and development · as-of 2026-01-25
Employees, roughly · as-of 2026-08-26
What else could explain it
line composition: The bill sits inside research and development.
Quotes
“For the full year, we now expect OpEx to grow in the low 50s, driven by a broadening of our product portfolio and further increase in the usage of AI tools, which has already and will continue to enhance engineering productivity.”
“The reason for that is because you should rent intelligence, strong intelligence, smart intelligence wherever you can, which is the reason why we rent it, and I encourage my employees to use the cloud service as much as they can.”
Q1 2026direction only · described, no size · exploratory
Q2 2026direction only · described, no size · exploratory
Q3 2026direction only · described, no size · exploratory
research
Amortization of the Groq inference technology licence and the hired team
0.1% to 0.16% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: quantified· motive: offensive· before LLMs: expanded
The first rack-scale system on the Groq technology is in full production (claim c29), and the deferred consideration was paid in the first half, in financing activities (claim c30). Amortization of intangible assets was ; the ledger’s estimate for the Groq asset is . The filing does not say which line holds the Groq amortization: cost of revenue also includes acquisition-related intangible amortization (claim c52), so part of it may also sit inside the cost-of-revenue channel, a small overlap that is not removed.
Evidence: 3 quotes, 2 figures, 2 confounds, 2 from before coverage
The income-statement cost of the December 2025 non-exclusive licence of Groq’s language processing unit technology for low-latency inference, with the hiring of Groq employees: amortization of the developed technology intangible over its useful life. The consideration and the goodwill are outside the income statement and are recorded as metrics. Groq’s technology is an AI inference product now sold inside Data Center revenue (Groq LPX), so the channel is an AI acquisition in the ledger’s sense. The filings do not say which line holds the amortization; cost of revenue includes acquisition-related amortization, so a small part may also sit in the cost-of-revenue channel.
Why this motive
The licensed technology now ships as a product (claim c29).
Before LLMs: expanded
Buying technology and companies, and amortizing developed technology and licensed rights, predates the licence: the anchor describes both. No Groq cost was in the income statement before the licence closed in December 2025. The size is the change AI made, not the whole line.
“We have acquired and invested and may continue to do so in businesses that offer products, services and technologies that we believe will help expand or enhance our strategic objectives.”
“Intangible assets primarily represent acquired intangible assets including developed technology and customer relationships, as well as rights acquired under technology licenses, patents, and acquired IP.”
Amortization expense of intangible assets · 2026-CQ3
Payment related to Groq in financing activities, first half of fiscal 2027 · 2026-01-26..2026-07-26
What else could explain it
acquisition: Compensation of the hired Groq employees is not separated.
line composition: Total amortization covers all intangible assets.
Quotes
“Since the announcement of our Groq partnership last year, we've been working to unite NVIDIA's high throughput and Groq's high interactivity architectures. At Hot Chips earlier this week, we announced that Groq 3 LPX, our first rack-scale LPU system, is in full production and already setting records, demonstrating nearly 4x the number of tokens per second against the next best alternative on our Artificial Analysis benchmark.”
“Cash used in financing activities was flat in the first half of fiscal year 2027 compared to the first half of fiscal year 2026, mainly due to higher share repurchases, dividends, and a payment related to Groq, Inc. in the first half of fiscal year 2027, offset by higher cash proceeds from debt issuance.”
“Cost of revenue also includes acquisition-related intangible amortization expense, IP-related costs, and stock-based compensation related to personnel associated with manufacturing operations.”
Cost of the AI systems sold: wafers, memory, packaging and system assembly
17.9% to 22.5% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: described· motive: offensive· before LLMs: expanded· layer: hardware
Cost of revenue was , of revenue. Supply and capacity commitments rose to , primarily for memory, whose scarcity the CFO attributes in large part to the AI build-out itself (claims c32, c38). The size, , is the ledger’s: the cost of the AI systems sold, paid on to suppliers.
Evidence: 4 quotes, 5 figures, 2 confounds, 2 from before coverage
The part of cost of revenue that buys the AI systems the company sells: wafer fabrication, high-bandwidth memory, advanced packaging, boards and rack assembly paid to foundries, memory makers and contract manufacturers. It is the next layer of the same money: buyers’ capital spending becomes the company’s revenue, and this part of it is paid on to suppliers. The filing gives cost of revenue by component type only in words; the size is cost of revenue times the Data Center share and an assumed AI share. Supply and capacity commitments and prepaid supply are recorded as metrics.
Why this motive
The cost of delivering separately priced AI systems; the CFO ties memory scarcity to the AI build-out (claim c32).
Before LLMs: expanded
Cost of revenue was in fiscal 2024 against revenue of , with gross margin of in Q1 of that year; supply and capacity obligations were and long-term prepaid supply at its end. No quarter of the AI part can be traced. The size is the whole of an activity that existed before.
“Our overall gross margin increased to 72.7% in fiscal year 2024 from 56.9% in fiscal year 2023. The year over year increase was primarily due to strong Data Center revenue growth of 217% and lower net inventory provisions as a percentage of revenue.”
Supply and capacity commitments, total · as-of 2026-07-26
Provisions for inventory and excess purchase obligations · 2026-CQ3
What else could explain it
other: Memory prices have risen beyond expectations (claim c31); part of the cost increase is price, not volume.
line composition: Cost of revenue includes edge products and inventory provisions.
Quotes
“As you are already aware, we are experiencing extreme pricing conditions in memory. The magnitude of the price increase has exceeded our prior expectations and are headed even higher into next year.”
“Memory scarcity today is being driven in large part by the AI build-out itself, and unlike a component that simply raises our cost with no offset benefit.”
“We have significantly increased our supply and capacity commitments from $119 billion last quarter to $279 billion as of July 26, 2026 to meet future demand.”
Cost displaced by AI1 channel · $8.5mn to $317mn sized · $8.5mn to $317mn incremental
engineering · cheap to verify
Engineering cost avoided through the use of AI tools
0.01% to 0.33% of the quarter’s revenue
Incremental total: counts in full.
our inferencedisclosure: described· motive: exploratory· before LLMs: expanded
The CFO says AI tools have already enhanced and will continue to enhance engineering productivity (claim c26); research and development rose to . The size, , is the ledger’s counterfactual.
Evidence: 2 quotes, 3 figures, 1 confound, 1 from before coverage
Research and development cost the company would otherwise spend for the same output, which management says AI tools enhance. No saving, rate or headcount effect is given and research and development keeps rising; the saving is the ledger’s counterfactual.
Why this motive
AI tools are said to have already enhanced engineering productivity, with no measure (claim c26).
Before LLMs: expanded
Engineering is the anchor’s largest cost: research and development was in fiscal 2024 with employees in research and development. Putting LLM tools into that work is the change. The size is the change AI made, not the whole line.
“As of the end of fiscal year 2024, we had approximately 29,600 employees in 36 countries, 22,200 were engaged in research and development and 7,400 were engaged in sales, marketing, operations, and administrative positions.”
Research and development, year-over-year growth · 2026-CQ3
Compute infrastructure within research and development, year-over-year growth · 2026-CQ3
What else could explain it
operating leverage: Research and development keeps rising; any saving is a counterfactual.
Quotes
“The increases in research and development expenses for the second quarter and first half of fiscal year 2027 were primarily driven by a 127% and 120% increase in compute infrastructure, respectively, and a 30% increase in each fiscal year 2027 period in compensation and benefits, including stock-based compensation, reflecting employee growth and compensation increases.”
“For the full year, we now expect OpEx to grow in the low 50s, driven by a broadening of our product portfolio and further increase in the usage of AI tools, which has already and will continue to enhance engineering productivity.”
Revenue arriving through AI7 channels · $77.34bn to $87.98bn sized · $0 to $86.37bn incremental
product revenue
Data center AI infrastructure bought by the hyperscale clouds and the largest consumer internet companies
46.5% to 49.1% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: quantified· motive: offensive· before LLMs: expanded· layer: hardware
Hyperscale revenue was , up , within Data Center revenue of (claim c1). The sub-market was redefined this quarter: a company moved from ACIE to Hyperscale, which recasts Q1 fiscal 2027 Hyperscale to (claim c2). The size, , is the sub-market times an assumed AI share. Funding is mixed, read for the buyers as in the prior quarters: the ledger’s capital-spending channels for them read mixed, operating cash flow with debt, at MSFT and ORCL and in the draft quarters at AMZN, GOOGL and META. The company’s own part is its payment terms: receivables rose to days on extended payment terms for large multi-quarter purchases by certain investment-grade customers, which the 10-Q calls financing arrangements (claims c16, c17). The money is the hyperscalers’ capital spending, part of it traced at MSFT and ORCL.
Evidence: 10 quotes, 8 figures, 3 confounds, 5 from before coverage
Accelerated systems, GPUs, CPUs and networking sold for AI training and inference to the public clouds and the largest consumer internet companies, the sub-market the company calls Hyperscale from fiscal 2027 and, before that, its hyperscaler customer category. Most of it reaches these buyers through direct customers that build the systems (contract manufacturers, system makers), so the buyer is an indirect customer in the concentration note. The sub-market also holds non-AI work the company accelerates for the same buyers (data processing, classic search and recommendation workloads moved to GPUs), so its revenue is a ceiling on the AI part; the size is the sub-market times an assumed AI share. This money is the capital spending of the buyers: part of it is the accelerator purchases traced at MSFT and ORCL as capital spending.
Why this motive
Separately priced AI systems; the CFO ties the hyperscalers’ growing revenue and backlog to new GPU capacity (claims c1, c40).
Before LLMs: expanded
At the anchor cloud service providers and consumer internet companies were named customer categories, and an indirect customer was estimated at of fiscal 2024 revenue. Data Center revenue as a whole was in Q1 of fiscal 2024 and for that year; no hyperscaler level was given, so no quarter of this sub-market can be traced as a baseline. The size is the whole of an activity that existed before.
“Let me give you some color across our three major customer categories: cloud service providers or CSPs, consumer internet companies, and enterprises.”
“Multiple CSPs announced the availability of H100 on their platforms, including private previews at Microsoft Azure, Google Cloud, and Oracle Cloud Infrastructure, upcoming offerings at AWS and general availability at emerging GPU specialized cloud providers like CoreWeave and Lambda.”
“One indirect customer which primarily purchases our products through system integrators and distributors, including through Customer A, is estimated to have represented approximately 19% of total revenue for fiscal year 2024, attributable to the Compute & Networking segment.”
“Our end customers include the world’s leading public cloud and consumer internet companies, thousands of enterprises and startups, and public sector entities.”
Hyperscale revenue, Q1 fiscal 2027, recast for a reclassified company · 2026-CQ2
Q1 fiscal 2027 revenue moved from ACIE to Hyperscale by the reclassification · 2026-CQ2
Data Center revenue · 2026-CQ3
Largest direct customer, share of total revenue · 2026-CQ3
Days sales outstanding · as-of 2026-07-26
What else could explain it
relabel: A company was moved from ACIE to Hyperscale this quarter and prior periods recast (claim c2); Q1 fiscal 2027 Hyperscale rises by on the new definition.
line composition: Hyperscale also holds data processing and classic machine learning; the AI share is assumed.
other: Hyperscalers also rent capacity from neoclouds, which is counted in ACIE (claim c3).
Quotes
“Q2 data center revenue increased 18% quarter-over-quarter to $89 billion, with strong contributions from both sub-segments, hyperscale and ACIE, which includes our neocloud, industrial, and enterprise customers. Hyperscale revenue of $49 billion grew 13% sequentially, driven by sustained strength in Blackwell.”
“During the second quarter we reclassified a company from AI Clouds, Industrial, & Enterprise (ACIE) to Hyperscale due to a change in their business model and recast the prior period revenue associated with this company.”
“ACIE revenue of $40 billion increased 25% sequentially and 138% year-over-year. Growth was driven by neocloud capacity additions to meet the rising demand from enterprises, AI startups, and sovereigns, as well as hyperscalers purchasing capacity to supplement their own build-outs.”
“We estimate that one AI research and deployment company contributed a meaningful amount of our revenue by purchasing cloud services from our customers in the second quarter and first half of fiscal year 2027.”
“Accounts receivable was $63.1 billion with 60 days sales outstanding (DSO), up from 45 days sequentially, due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers.”
“Financing arrangements with certain investment-grade customers, including extended payment terms under large, multi-quarter agreements, will continue to affect the timing of our operating cash flows.”
“In Q2, we shipped less than 1% of our total data center revenue in Hopper 200 products to customers based in China in accordance with the U.S. government licenses.”
“With cloud industry backlog now greater than $2 trillion, CapEx by the top five hyperscalers is expected to reach nearly $800 billion in 2026 and $1.3 trillion in 2027.”
Data center AI infrastructure bought by AI clouds, AI model makers, sovereigns, industrial and enterprise customers (ACIE)
33.6% to 40.6% of the quarter’s revenue
Incremental total: counts at zero at the point and in full at the high end, with no traced baseline.
our inferencedisclosure: quantified· motive: offensive· before LLMs: expanded· layer: hardware
ACIE revenue was , up , driven by neoclouds serving enterprises, AI startups and sovereigns and by hyperscalers renting extra capacity (claim c3). The size, , is the sub-market times an assumed AI share. Funding is mixed, and the company’s part grew: under new AI cloud agreements the clouds procure its systems and it commits to purchase their capacity, in total, a take-or-pay floor that lets lenders finance the cloud, with a revenue share above it (claims c4, c18); land, power and shell backstops for AI clouds reach . The CFO says independent capital still underwrites each deal (claim c5), and cites AI venture funding of more than in the half, most of it spent on compute (claim c23). Model makers buying directly fall under the rule for labs: the company’s part is its stakes in them, nearly by the call (claim c7), beside those outside rounds. The capacity commitments carry no payment until fiscal 2028 and the CFO says independent capital underwrites each deal, so no payment in the quarter has the company’s money as its stated source.
Evidence: 19 quotes, 8 figures, 3 confounds, 3 from before coverage
Data center revenue from every customer outside the hyperscale sub-market: AI clouds (neoclouds), AI model makers purchasing directly, sovereign and regional clouds, enterprises and industrial companies building their own AI factories, and supercomputing centers. The company reports it as AI Clouds, Industrial, and Enterprise from fiscal 2027; in the fiscal 2026 fourth quarter it is Data Center revenue less the hyperscaler share. It also holds non-AI accelerated work (computational lithography, scientific computing, data processing), so the sub-market is a ceiling and the size is the sub-market times an assumed AI share. Funding is mixed and partly the company’s own: lease guarantees for AI clouds, capacity it commits to take back from them, and equity stakes in the buyers.
Why this motive
Separately priced AI systems for neoclouds, enterprises, AI startups and sovereigns (claim c3).
Before LLMs: expanded
At the anchor enterprises, startups, public sector entities and GPU-specialized clouds such as CoreWeave and Lambda were already customers. Data Center revenue as a whole was in Q1 of fiscal 2024 and in fiscal 2022, before ChatGPT; no level for this group of customers was given. The size is the whole of an activity that existed before.
“Let me give you some color across our three major customer categories: cloud service providers or CSPs, consumer internet companies, and enterprises.”
“Multiple CSPs announced the availability of H100 on their platforms, including private previews at Microsoft Azure, Google Cloud, and Oracle Cloud Infrastructure, upcoming offerings at AWS and general availability at emerging GPU specialized cloud providers like CoreWeave and Lambda.”
“Our end customers include the world’s leading public cloud and consumer internet companies, thousands of enterprises and startups, and public sector entities.”
ACIE revenue, Q1 fiscal 2027, recast for a reclassified company · 2026-CQ2
Maximum exposure under land, power and shell guarantees for AI clouds · as-of 2026-07-26
AI cloud agreements: committed purchases of capacity from AI clouds, total · as-of 2026-07-26
Global venture funding in AI, first half of 2026 (a floor), as cited by the CFO · 2026-H1
Invested in the frontier AI labs, cumulative to the call date (nearly) · as-of 2026-08-26
What else could explain it
relabel: The reclassification of a company to Hyperscale lowers ACIE in recast periods (claim c2).
line composition: ACIE includes industrial work such as computational lithography (claim c14); the AI share is assumed.
other: Part of the buyers’ financing is the company’s own: capacity commitments, lease backstops and equity stakes.
Quotes
“During the second quarter we reclassified a company from AI Clouds, Industrial, & Enterprise (ACIE) to Hyperscale due to a change in their business model and recast the prior period revenue associated with this company.”
“ACIE revenue of $40 billion increased 25% sequentially and 138% year-over-year. Growth was driven by neocloud capacity additions to meet the rising demand from enterprises, AI startups, and sovereigns, as well as hyperscalers purchasing capacity to supplement their own build-outs.”
“Rather than allocating their entire capacity to a single long-term offtake guarantee that lenders typically require to finance a data center independently, we have introduced a revenue-sharing structure. NVIDIA provides a take or pay commitment on a portion of the facility's capacity, a minimum revenue guarantee that gives lenders the confidence to underwrite the project, and in exchange, we share in a portion of the neocloud's revenue earned above that floor.”
“Samsung Electronics is using NVIDIA cuLitho to achieve up to 20x greater performance in computational lithography, while Bristol Myers Squibb is investing in Vera Rubin AI factory, a fast follow to the Roche and Lilly build-outs as drug R&D timelines compress from years to months.”
“Under these agreements, AI clouds procure our data center infrastructure products and we commit to cloud service agreements, which the AI clouds can unilaterally stop providing to us and sell to third-party customers at more advantageous rates.”
“We have made, and may continue to make, investments and commitments in our ecosystem to enhance our growth opportunities, cultivate our ecosystem, and strengthen our competitive position. These include equity investments of $99 billion and equity investment commitments of $25 billion as of July 26, 2026.”
“We believe AI clouds and AI model makers have significant demand for training and inference compute and currently lack the ability to secure long-term infrastructure contracts and investment-grade financing capacity to secure the AI infrastructure necessary to grow.”
“Global VC funding in AI, roughly 70% of which is spent on compute, exceeded $400 billion in the first half of 2026, surpassing the $265 billion raised in all of 2025.”
“In enterprise, on a trailing 12-month basis, on-prem revenue in the automotive vertical reached $8 billion, while financial services, manufacturing, and healthcare combined contributed $7 billion in revenue.”
“In Q2, we shipped less than 1% of our total data center revenue in Hopper 200 products to customers based in China in accordance with the U.S. government licenses.”
“In August 2026, we entered into memorandums of understanding with several large capital providers to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital over time to support the deployment of AI infrastructure.”
“Hyperscalers will remain a major growth driver, but non-hyperscaler growth, our ACIE segment spanning sovereign regional neoclouds, enterprise edge, and air gap data centers will represent roughly half of our data center business.”
“Using NVIDIA DSX reference designs, our neocloud partners are bringing capacity online faster and at lower token cost. They are expected to exit the year with 8 GW in total installed capacity, up from approximately 3 GW at the end of 2025.”
“In this model, we get paid twice, once on the hardware sale and again through the share of rental revenue, a highly reoccurring stream layered on top of a one-time equipment purchase.”
“There is sovereign AI, there are regional AIs, there are neoclouds, there are AI startups at enterprises where we are seeing, which represents about half of our business, and that is growing 100% a year.”
“ACIE revenue increased 138% from a year ago and 25% sequentially driven by end-demand from AI natives, enterprises, and sovereign customers, as well as hyperscalers utilizing AI clouds.”
“•Announced strategic partnerships to establish independent compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third-party capital for the buildout of AI infrastructure over time, subject to definitive agreements.”
Sovereign AI: AI infrastructure bought by or for national governments, mostly through regional clouds
7.6% to 24.6% of the quarter’s revenue; overlaps another channel, not added into totals
our inferencedisclosure: direction only· motive: offensive· before LLMs: expanded· layer: hardware
Sovereign AI, mostly through regional neoclouds, grew on the quarter and more than tripled on the year (claim c15). The size, , rolls the prior estimate forward at that rate. It sits inside ACIE.
Evidence: 4 quotes, 1 figure, 1 confound, 2 from before coverage
Data center revenue the company labels sovereign AI: AI infrastructure for countries, bought by governments, national AI companies and regional clouds that host national capacity. Management gives a fiscal-year level and growth rates, not a quarter. It sits inside the ACIE channel and is not additive with it. The parts of the counterparty are public budgets, state-backed companies and regional clouds financed by investors and lenders.
Why this motive
A named AI customer group growing on the quarter (claim c15).
Before LLMs: expanded
Public sector entities and publicly funded supercomputers were customers at the anchor (the University of Bristol’s new supercomputer on the anchor call). The anchor gives no sovereign AI level; Data Center revenue was in Q1 of fiscal 2024. The size is the whole of an activity that existed before.
“Our end customers include the world’s leading public cloud and consumer internet companies, thousands of enterprises and startups, and public sector entities.”
“At this week's International Supercomputing Conference in Germany, the University of Bristol announced a new supercomputer based on the NVIDIA Grace CPU Superchip, which is 6x more energy efficient than their previous supercomputer.”
Sovereign AI revenue, sequential growth · 2026-CQ3
What else could explain it
other: No level is given; the estimate rolls forward from Q4 fiscal 2026 and compounds that range.
Quotes
“ACIE revenue of $40 billion increased 25% sequentially and 138% year-over-year. Growth was driven by neocloud capacity additions to meet the rising demand from enterprises, AI startups, and sovereigns, as well as hyperscalers purchasing capacity to supplement their own build-outs.”
“There is sovereign AI, there are regional AIs, there are neoclouds, there are AI startups at enterprises where we are seeing, which represents about half of our business, and that is growing 100% a year.”
“ACIE revenue increased 138% from a year ago and 25% sequentially driven by end-demand from AI natives, enterprises, and sovereign customers, as well as hyperscalers utilizing AI clouds.”
Q2 2026direction only · our inference · offensive · $5.40bn to $17.55bn
Q3 2026direction only · our inference · offensive · $7.29bn to $23.69bn
customer cohort
Compute bought by and for the frontier AI labs, directly and through clouds
7.4% to 23.1% of the quarter’s revenue; overlaps another channel, not added into totals
our inferencedisclosure: described· motive: offensive· before LLMs: new· layer: hardware
The CFO says the frontier labs are growing faster than their balance sheets and credit can support (claim c6) and that the company has invested nearly in them, a cumulative amount to the call date (claim c7). After the quarter end, in August 2026, it entered into credit support capped at for leases to an OpenAI affiliate at SB Energy’s campus; each piece takes effect as its lease commences, the first expected in fiscal 2029, so it is a forward commitment, not financing of this quarter’s purchases (claim c13). The CFO says the company will provide credit enhancement for another lab, and it backs financing platforms meant to raise over of third-party capital (claims c9, c41). She expects the labs it backs with its balance sheet to contribute a share of next year’s business, a forward statement (claim c8), and acknowledges this may be called circular financing (claim c10); she also says independent capital still underwrites every deal (claim c5) and cites AI venture funding of more than in the half (claim c23). The extended payment terms of the quarter are for certain investment-grade customers (claims c16, c17), which the filing does not tie to the labs. Funding reads mixed under the rule for labs: the company’s part is its equity stakes in the labs (claim c7), and the outside rounds are the AI venture funding the CFO cites (claim c23). No payment in the quarter has the company’s money as its stated source: the credit support and the financing platforms are forward commitments, the extended payment terms are not tied to the labs, and the capacity commitments to AI clouds carry no payment until fiscal 2028. Equity investments held were , with more committed (claim c20). The size, , is the ledger’s ballpark.
Evidence: 21 quotes, 7 figures, 2 confounds, 1 from before coverage
The part of Data Center revenue whose end buyer is a frontier model maker (OpenAI, Anthropic, Meta Superintelligence Labs, xAI and others the company names), whether the lab buys systems itself or rents them from a cloud that buys them. The concentration note estimates that one AI research and deployment company contributed a meaningful amount of revenue by purchasing cloud services from the company’s customers. The company is also an investor in and credit supporter of these labs, so under the rule for labs their payments read mixed: the company’s stakes beside the labs’ outside funding rounds, unless a payment’s stated source is the company’s own money. It sits inside the hyperscale and ACIE channels and is not additive with them.
Why this motive
Compute bought for training and inference by labs whose growth is limited by compute (claim c6).
Before LLMs: new
The payer exists because of LLMs, so the money is new by the ledger’s first test. At the anchor generative AI pioneers reached the accelerators through the cloud providers and were not a disclosed customer group; Data Center revenue was in Q1 of fiscal 2024.
“In addition to enterprise AI adoption, these CSPs are serving strong demand for H100 from generative AI pioneers.”
Invested in the frontier AI labs, cumulative to the call date (nearly) · as-of 2026-08-26
Cap on credit support for the OpenAI site leases at SB Energy’s campus, entered in August 2026, after the quarter end; effective as each lease commences, the first expected in fiscal 2029 · as-of 2026-08-26
Equity investments held · as-of 2026-07-26
Equity investment commitments · as-of 2026-07-26
Third-party capital the financing platforms are to raise (a floor) · as-of 2026-08-26
Global venture funding in AI, first half of 2026 (a floor), as cited by the CFO · 2026-H1
Data Center revenue · 2026-CQ3
What else could explain it
other: The lab share of Data Center revenue is the ledger’s judgment, bounded above by the CFO’s expectation for next year.
relabel: Capacity clouds purchase and rent to labs is counted here and in the hyperscale or ACIE channel; not additive.
Quotes
“Independent capital still underwrites every deal on its own merits. We're not making loans.”
“The frontier AI labs have extraordinary demand for training and inference compute, but they are growing faster than what their balance sheets and credit profiles can support.”
“First, we have invested nearly $50 billion in the frontier AI labs. This was a meaningful commitment, but it represented a small fraction of our expected free cash flow over the same period.”
“For context, we expect demand from the AI labs for which we expect to leverage our balance sheet to contribute toward roughly a quarter of our business next year.”
“OpenAI has committed to substantial deployments of NVIDIA AI infrastructure through 2030. OpenAI's existing and planned commitments represent approximately 12 GW of NVIDIA Compute. For another frontier AI lab, we will provide selective credit enhancement for nearly 2 GW of compute.”
“We estimate that one AI research and deployment company contributed a meaningful amount of our revenue by purchasing cloud services from our customers in the second quarter and first half of fiscal year 2027.”
“SB Energy Corp. guarantees – In August 2026, we entered into guarantees, capped at a total of $105 billion, to provide credit support on a land, power, and shell buildout with affiliates of SB Energy Corp. (SB Energy) on behalf of a customer, an affiliate of OpenAI Group PBC (OpenAI), related to leases for approximately 4.25 gigawatts of IT load in the aggregate at SB Energy’s PORTS Technology Campus in Pike County, Ohio.”
“Accounts receivable was $63.1 billion with 60 days sales outstanding (DSO), up from 45 days sequentially, due to extended payment terms on large, multi-quarter agreements with certain investment-grade customers.”
“Financing arrangements with certain investment-grade customers, including extended payment terms under large, multi-quarter agreements, will continue to affect the timing of our operating cash flows.”
“We have made, and may continue to make, investments and commitments in our ecosystem to enhance our growth opportunities, cultivate our ecosystem, and strengthen our competitive position. These include equity investments of $99 billion and equity investment commitments of $25 billion as of July 26, 2026.”
“Equity investments – We committed to make certain equity investments in AI model makers, infrastructure financiers, and other private companies, subject to certain contingencies.”
“We believe AI clouds and AI model makers have significant demand for training and inference compute and currently lack the ability to secure long-term infrastructure contracts and investment-grade financing capacity to secure the AI infrastructure necessary to grow.”
“Global VC funding in AI, roughly 70% of which is spent on compute, exceeded $400 billion in the first half of 2026, surpassing the $265 billion raised in all of 2025.”
“In August 2026, we entered into memorandums of understanding with several large capital providers to establish independent financing platforms designed to mobilize more than $500 billion of third-party capital over time to support the deployment of AI infrastructure.”
“Further, to support the frontier labs infrastructure build-outs, we recently announced partnerships with six of the world's leading infrastructure capital providers, Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR, to establish financing platforms that will raise over $500 billion of third-party capital.”
“Now, having said that, taking a step backwards, investing in these two companies, or there are several AI labs that we've invested in, investing in these companies are a once-in-a-generation opportunity. I think the only regret that I have is that I didn't invest more and sooner. Two of the companies will likely go public soon, and others will follow.”
“Each generation of NVIDIA infrastructure deployed at PORTS-Pike could represent approximately 1.5 million NVIDIA GPUs, or approximately $150 billion to $200 billion in NVIDIA revenue.”
“•Announced strategic partnerships to establish independent compute financing platforms with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to mobilize over $500 billion of third-party capital for the buildout of AI infrastructure over time, subject to definitive agreements.”
Physical AI: compute for autonomous vehicles and robots, in the data center and at the edge
1.8% to 5.3% of the quarter’s revenue; overlaps another channel, not added into totals
our inferencedisclosure: withdrawn· motive: offensive· before LLMs: relabelled· layer: hardware
Management gave a physical AI level in each prior covered quarter, north of for fiscal 2026 and more than over the trailing twelve months to Q1 fiscal 2027, and gives none this quarter, so the metric reads withdrawn. On-premises data center revenue from automakers reached over the trailing year, a different measure, and Amazon is adopting the physical AI stack for warehouse robots (claims c35, c36). The size, , rolls forward the estimate built from the last stated level, the trailing-twelve-month floor. It overlaps the Data Center channels.
Evidence: 2 quotes, 3 figures, 2 confounds, 2 from before coverage
Revenue the company groups as physical AI: data center systems that automakers and robotics companies train and simulate on, the in-vehicle DRIVE platform, and Jetson and Thor robotics modules. It cuts across the Data Center channels and the automotive and edge lines, so it is not additive with the Data Center channels. Management gives fiscal-year and trailing levels as floors.
Why this motive
Carried from the prior quarters: a product line with stated levels.
Before LLMs: relabelled
The same products were sold at the anchor under their own names: the DRIVE automated-driving platform and Jetson robotics in the Compute & Networking segment, with automotive revenue of in Q1 of fiscal 2024. Physical AI is a grouping of existing lines; no movement the company attributes to the grouping is measured.
“The Compute & Networking segment is comprised of our Data Center accelerated computing platforms and end-to-end networking platforms including Quantum for InfiniBand and Spectrum for Ethernet; our NVIDIA DRIVE automated-driving platform and automotive development agreements; Jetson robotics and other embedded platforms; NVIDIA AI Enterprise and other software; and DGX Cloud software and services.”
Physical AI revenue, fiscal 2026 (a floor) · FY2026
Physical AI revenue, trailing twelve months (a floor) · TTM as-of 2026-04-26
On-premises data center revenue in the automotive vertical, trailing twelve months · TTM as-of 2026-07-26
What else could explain it
relabel: Physical AI groups data center, automotive and robotics sales the company already made.
other: No physical AI level this quarter; the estimate rolls forward from the last stated level.
Quotes
“In enterprise, on a trailing 12-month basis, on-prem revenue in the automotive vertical reached $8 billion, while financial services, manufacturing, and healthcare combined contributed $7 billion in revenue.”
AI workstations and deskside systems (RTX PRO workstations, DGX Spark, DGX Station)
0.27% to 1.7% of the quarter’s revenue
Incremental total: counts at zero.
our inferencedisclosure: described· motive: exploratory· before LLMs: relabelled· layer: hardware
Edge Computing revenue was , driven by strong workstation sales while consumer PCs slowed on memory prices (claim c24). The size, , is the ledger’s ballpark.
Evidence: 1 quote, 1 figure, 2 confounds, 2 from before coverage
Workstation GPUs and deskside AI systems that developers and enterprises purchase to build and run models locally, which the company says generative and agentic AI is expanding. Revenue is inside Professional Visualization (fiscal 2026) and then Edge Computing, which also holds gaming PCs, consoles, automotive and robotics. No AI share is given. The buyers are enterprises and individual developers, so the counterparty is mixed.
Why this motive
Carried from the prior quarters: workstation growth credited to Blackwell with no AI measure.
Before LLMs: relabelled
At the anchor the annual report already said generative AI was expanding the market for workstation GPUs, and the CFO called it a major new workload for them; Professional Visualization revenue was in Q1 of fiscal 2024. With no measure of AI-attributed demand, the tie-break reads the channel as relabelled.
“In addition, generative AI is expanding the market for our workstation-class GPUs, as more enterprise customers develop and deploy AI applications with their data on-premises.”
mix shift: Workstation demand includes design and content creation.
other: Blackwell product cycle: the company credits the growth to Blackwell demand beside its AI systems, and no measure separates AI’s part. Under the rule for AI named beside another cause no part of the growth is credited to AI; the size is a ballpark of the level of AI workstation revenue, an assumed share of the line, and the channel is relabelled, so it adds nothing to the incremental total.
Quotes
“Edge Computing revenue for the second quarter was $7.2 billion, up 27% from a year ago and up 13% sequentially. The increases were driven by strong sales of Blackwell workstations, partially offset by slower consumer PC sales that were tempered by elevated memory and systems prices.”
Gains and losses on equity stakes in AI model makers, AI clouds and other ecosystem companies
0.81% to 6.5% of the quarter’s revenue; a non-operating gain, not added into totals
our inferencedisclosure: described· motive: exploratory· before LLMs: expanded
Net gains from equity securities were (claim c34); equity investments held were with committed, focused on AI model makers and infrastructure financiers (claims c20, c33). Warrants on publicly traded stock received in the quarter, from issuers the filing does not name, were valued at at the quarter end (claim c56). The AI part, , is the ledger’s ballpark, non-operating and out of totals.
Evidence: 6 quotes, 4 figures, 2 confounds, 2 from before coverage
Realized and unrealized gains and losses on the company’s publicly held and non-marketable equity securities, recognized in other income: stakes in AI model makers, AI clouds, infrastructure financiers and other private companies, and in partners such as Intel. Not revenue and not cash: a non-operating channel, traced and left out of flow totals. The filings do not split the gains by investee, so the AI part is an assumed share. The layer is end-use: the investees span model makers, AI clouds and financiers, and the filings do not split the marks among them.
Why this motive
Carried: ecosystem stakes, now focused on AI model makers and infrastructure financiers (claim c33).
Before LLMs: expanded
Strategic equity stakes predate the build: of non-marketable equity securities at the anchor, and, as context only, an unrealized gain of on a private company in Q4 fiscal 2024, after ChatGPT. That gain is not the quarter’s total and its investee may itself be an AI company, so no quarter before AI is traced as a baseline. The size is the whole of an activity that existed before.
“The carrying value of our non-marketable equity securities totaled $1.3 billion and $288 million as of January 28, 2024 and January 29, 2023, respectively.”
“In the fourth quarter of fiscal year 2024, one of our private company investments completed a secondary equity raise that resulted in an unrealized gain of $178 million.”
Fair value of equity derivatives (warrants received in the quarter) · as-of 2026-07-26
What else could explain it
other: Gains are not split by investee.
one time item: Marks move with share prices and financing rounds.
Quotes
“We have made, and may continue to make, investments and commitments in our ecosystem to enhance our growth opportunities, cultivate our ecosystem, and strengthen our competitive position. These include equity investments of $99 billion and equity investment commitments of $25 billion as of July 26, 2026.”
“Equity investments – We committed to make certain equity investments in AI model makers, infrastructure financiers, and other private companies, subject to certain contingencies.”
“Now, having said that, taking a step backwards, investing in these two companies, or there are several AI labs that we've invested in, investing in these companies are a once-in-a-generation opportunity. I think the only regret that I have is that I didn't invest more and sooner. Two of the companies will likely go public soon, and others will follow.”
“In the second quarter of fiscal year 2027, we received warrants to purchase shares of publicly-traded common stock with terms of three to five years.”
Q3 2026. Growing slower than revenue: cost of revenue (+86.8%), research and development (+64.4%), sales, general and administrative (+20.7%), total operating expenses (+55.3%). A displaced cost shows up as a line that stays under the dashed revenue line. These are the audited lines, as first reported; nothing here is attributed to AI by the filing. Not drawn: cost of revenue, research and development, which one-time items move by more than 60% in a quarter; the values are in the table below.
Sales, general and administrativeTotal operating expensesRevenue