VC
Value Add VC
โšกHomePulseโšกHelpful Apps๐Ÿ“Blog๐ŸคPartner
Illustration for: Nvidia Is the AI Economy. That Should Worry You.
Value Add VC/Pulse/AITRACE'S TAKE$96B quarterly revenue

Nvidia Is the AI Economy. That Should Worry You.

Nvidia posted $96 billion of quarterly revenue and guided to 70% growth while investing in, lending to and acquiring the customers who buy its chips -- a structure that looks less like a market and more like one balance sheet.

Nvidia
TC
Trace Cohen
Early-stage VC & angel ยท Founder, New York Venture Partners
August 28, 2026
2 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Nvidia reported roughly $96 billion of revenue and $54 billion of earnings last quarter and guided to about 70% growth next year, per [Newcomer](https://www.newcomer.co/p/nvidia-is-carrying-the-ai-economy)

2

It holds equity in CoreWeave, sells CoreWeave chips, and has backstopped it -- while CoreWeave borrows against signed contracts to buy more chips

3

The same month it agreed to buy Poolside for ~$6B and Hugging Face for ~$12.9B, and explored investing in Perplexity at $30B

4

Its five largest customers -- OpenAI, Google, Amazon, Meta and Anthropic -- are all building competing silicon

TC

The VC Read ยท Trace's Take

Trace Cohen

The single number I would put in every AI diligence memo this quarter: Nvidia's receivables growth versus revenue growth. If DSO stretches while revenue accelerates, the demand is being financed rather than bought. Second item: ask any neocloud in your portfolio for the covenant package on its GPU debt, not the headline facility size. One covenant breach at one large neocloud reprices the whole category, and most LPs have no idea how correlated their AI exposure actually is.

AI Spending โ†’ AI Buildout Tracker โ†’

Analysis

I have been long the AI buildout and I am still long it. But I want to name the thing that makes me uncomfortable, because nobody funding this cycle should pretend it is a normal market: Nvidia is increasingly on both sides of its own demand curve.

The mechanics, as Newcomer laid out this week, run like this. Nvidia holds equity in CoreWeave. It sells CoreWeave chips. It has backstopped CoreWeave. CoreWeave signs customer contracts, borrows against those contracts, and spends the proceeds on more Nvidia chips. Every step is legal and every step is documented. It is also a loop, and loops look fantastic on the way up.

Now layer on the rest of August:

โ€œNvidia's investments are also small relative to its cash generation, and a 70% growth guide with 56% net margins is not a company propping up phantom demand.โ€

  • Poolside โ€” ~$6B acquisition
  • Hugging Face โ€” ~$12.9B acquisition
  • Perplexity โ€” reported interest at a $30B valuation
  • Customer loan guarantees โ€” could run into the hundreds of billions, OpenAI included

Nvidia itself did $96 billion of revenue and $54 billion of earnings last quarter and is guiding to about 70% growth. When a single company can fund its customers, acquire its ecosystem and still print that, the S&P's AI exposure is not diversified across a dozen names. It is one credit.

What I actually watch is not the revenue line. It is receivables and days sales outstanding, and the disclosure footnotes on guarantees and equity stakes in customers. Vendor-financed demand shows up there first, quarters before it shows up in the top line. The second thing: OpenAI, Google, Amazon, Meta and Anthropic are Nvidia's biggest buyers and are all shipping their own silicon. TPUs, Trainium, MTIA. Every one of those programs is a customer trying to become a competitor while still being invoiced.

Room for disagreement: the strongest counter is that this is what every genuine platform buildout looks like. Cisco vendor-financed telcos in the 1990s and a lot of that lending was fine until it wasn't -- but Lucent and Nortel financed customers with no revenue, while CoreWeave has signed Microsoft and OpenAI contracts behind the collateral. Nvidia's investments are also small relative to its cash generation, and a 70% growth guide with 56% net margins is not a company propping up phantom demand. If the offtake contracts are real and the depreciation schedules hold, this is just a capital-intensive industry doing capital-intensive things.

I think both can be true. The demand is real and the structure is fragile. The fragility does not need a demand collapse to bite -- it needs one large neocloud to miss a covenant, and the credit market to reprice the whole category at once.

Related Deep Dives

  • Custom AI Chips: Why Google, Amazon, and Microsoft Are Bu... โ†’
  • Nvidia's $12.9B Hugging Face Deal: Its Biggest Acquisitio... โ†’
  • OpenAI vs Anthropic Revenue Dispute: $74B Gross ARR vs $4... โ†’
ShareXLinkedInEmail

More on

Nvidia โ†’

Prior Pulse Coverage

NvidiaWhy the AI Buildout Runs on Debt, Not VCsNvidiaOpen-Weight Labs Become the Valley's Acquisition TargetNvidiaNvidia Pauses Its Own AI Cloud Financing PlanNvidiaNvidia Now Carries $366B in Future CommitmentsNvidiaNvidia Is Financing Its Own Demand Curve

Key Sources

2 sources
SourceNewcomer
AnalysisValue Add Pulse

Reported by Newcomer ยท Analysis by Value Add Pulse.

โ† Back to Pulse

THE WIRE in your inboxโ€” Tech, startup & VC news with Trace's take. Free, no spam.

Read Next

AIยท Aug 28, 2026

Anthropic Gives AI Agents a Plug for Real Machines

Illustration for: Anthropic Gives AI Agents a Plug for Real Machines
AI

Anthropic Gives AI Agents a Plug for Real Machines

Anthropic previewed the Model Hardware Standard, a shared specification that lets AI agents operate microscopes, liquid handlers and robotic arms, with initial partners in biotech, robotics, quantum computing and manufacturing.

AIยท Aug 28, 2026

Cohere Loses the Benchmark, Wins on Cost Per Page

Illustration for: Cohere Loses the Benchmark, Wins on Cost Per Page
AI$1.50 per 1,000 pages

Cohere Loses the Benchmark, Wins on Cost Per Page

Cohere's Parse 5 scored 79.2 on ParseBench against GPT-5.5's 84.4, but prices at $1.50 per 1,000 pages -- a deliberate bet that document parsing is a cost problem, not an intelligence problem.

AIยท Aug 28, 2026

Visa Open-Sources an Agent That Patches Its Own Code

Illustration for: Visa Open-Sources an Agent That Patches Its Own Code
AI

Visa Open-Sources an Agent That Patches Its Own Code

Visa released the Vulnerability Agentic Harness, an 11-stage open-source pipeline that finds security flaws, writes candidate fixes and adversarially validates them before a human reviews, cutting some remediations from weeks to hours.

Deep Dives

Custom AI Chips: Why Google, Amazon, and Microsoft Are Bu...Nvidia's $12.9B Hugging Face Deal: Its Biggest Acquisitio...OpenAI vs Anthropic Revenue Dispute: $74B Gross ARR vs $4...
@Trace_Cohenยทt@nyvp.com