VC
Value Add VC
⚡HomePulse⚡Helpful Apps📝Blog🤝Partner
Illustration for: Nvidia's $51 Billion Bet Shows AI's Money Never Leaves
Value Add VC/Pulse/BIG TECHDEEP DIVE

Nvidia's $51 Billion Bet Shows AI's Money Never Leaves

Nvidia's newly disclosed $21 billion SpaceX stake and $30 billion Intel position -- 80% of its entire public portfolio -- are the clearest evidence yet that AI-era capital keeps circulating through the same handful of companies rather than spreading out.

By the Numbers

~$51B
Nvidia's combined stakes
~80%
Share of Nvidia's portfolio
$2.2B
Harvard's SpaceX stake
$11.5B
Anthropic Q2 revenue
40
New unicorns, July 2026
TC
By the Markets Desk
Edited by Trace Cohen · Early-stage VC & angel · Founder, New York Venture Partners
August 16, 2026
3 min read
ShareXLinkedInEmail

THE RUNDOWN

1

Nvidia's August 14 regulatory filing put a number on its SpaceX position for the first time -- roughly $21 billion -- and showed its 2025 Intel stake has swelled to about $30 billion, per [Bloomberg](https://www.bloomberg.com/news/articles/2026-08-14/nvidia-has-21-billion-spacex-stake-30-billion-in-intel-shares)

2

Intel and SpaceX together account for roughly 80% of Nvidia's $63.44 billion U.S.-listed equity portfolio as of June 30, per [Fortune](https://fortune.com/2026/08/15/nvidia-21-billion-spacex-stake-30-billion-intel-shares) -- both companies have separately pledged to make Nvidia their exclusive chip supplier

3

The filing lands in the same week Harvard's endowment disclosed a $2.2 billion SpaceX stake and Anthropic reported $11.5 billion in Q2 revenue -- three unrelated filings that all point at the same small circle of AI-adjacent winners

4

It also lands one day after BCA Research's Dhaval Joshi warned AI markets are a 'rolling sequence' of bubbles, not one -- Nvidia's stakes are the mechanism, not just a symptom, of the capital concentration he's describing

TC

The VC Read · Trace's Take

Trace Cohen

The diligence item that matters here isn't the size of Nvidia's stakes, it's the accounting treatment. Ask any GP with AI infrastructure exposure whether portfolio company revenue includes credits redeemable only for more compute, or strategic equity booked at a circular valuation rather than cash. Comparable deal to watch: Microsoft's OpenAI arrangement, which took years of investor pressure before the vendor-financing structure got real scrutiny. Nvidia's 80%-concentrated portfolio is the same trade at a bigger scale.

→ →

Analysis

Nvidia's quarterly regulatory filing, made public Friday, gave a number to something the market had only estimated until now: a roughly $21 billion stake in SpaceX, the company's first disclosed valuation of that position, and a $30 billion stake in Intel, according to Bloomberg.

Together, the two positions account for roughly 80% of Nvidia's disclosed U.S.-listed equity holdings as of June 30 -- meaning four-fifths of the chipmaker's public investment book now sits in two companies that also happen to be two of its largest customers.

The Intel stake traces back to a $5 billion purchase Nvidia made in September 2025 as part of a strategic AI-infrastructure partnership; that position alone has generated close to $25 billion in paper gains as Intel's turnaround narrative gained traction, per Tom's Hardware. The SpaceX position dates to Nvidia's earlier investment in xAI, which was folded into SpaceX ahead of its record-setting public debut in June. Nvidia had never before disclosed a value for that stake.

“Watch Q3 filings from Intel and Microsoft for that distinction; right now, most of it is still bundled into one number.”

## The Exclusive-Supplier Trade What makes both positions more than passive investing is the strings attached: SpaceX and Intel have each separately pledged to make Nvidia their exclusive chip supplier. That turns equity ownership into a retention mechanism as much as a financial bet -- Nvidia isn't just capturing upside in two AI-adjacent winners, it's using its balance sheet to lock in two large customers against the pull of in-house silicon (Intel's own fabs, SpaceX's Starlink compute needs) or rival chipmakers like AMD and Broadcom, both of which have been winning custom AI-accelerator contracts with hyperscalers this year.

The pattern isn't limited to these two names. Nvidia has previously committed capital and compute deals across OpenAI, CoreWeave and a widening circle of AI infrastructure buyers, and Microsoft, Amazon and Google run comparable arrangements with the labs and startups they back. What's different about the SpaceX and Intel disclosures is the scale relative to Nvidia's own portfolio -- this isn't a side bet, it's four-fifths of it.

## Numbers In Context Harvard's endowment, one of the most sophisticated institutional investors in the world, disclosed its own SpaceX stake this week at $2.2 billion -- its largest single public position, built over more than a decade. Nvidia's SpaceX stake alone is roughly ten times that size, accumulated far more recently and directly tied to a supply relationship rather than a passive return bet. Anthropic, meanwhile, reported $11.5 billion in Q2 revenue this week -- real, recognized revenue that stands in some contrast to the paper gains driving headlines like Nvidia's.

July alone minted 40 new unicorns worth a combined $49 billion in decacorn value, per Crunchbase -- a reminder that capital is still forming new companies even as the biggest checks concentrate around a handful of already-enormous names.

## What the Circularity Doesn't Show None of this means the underlying businesses are fake. Intel's fabs are real, SpaceX's launch cadence is real, and Nvidia's chips are genuinely the scarce resource in the AI buildout. But equity stakes exchanged for supply commitments make it harder to separate organic demand from vendor-financed demand -- if a customer's balance sheet is partly filled with its supplier's goodwill, a slowdown in one shows up as a wobble in the other. BCA Research's Dhaval Joshi flagged a related concern this week, warning that AI-adjacent markets aren't one bubble but a rolling sequence of them, with hyperscaler capex on pace to exceed free cash flow by 2027.

The diligence question for anyone underwriting exposure to this trade isn't whether Nvidia's stakes are real -- they are, and they're growing. It's whether the next round of hyperscaler and lab financial disclosures start breaking out vendor-financed revenue and strategic equity separately from organic sales, the way accountants eventually forced telecoms to do after the dot-com-era vendor-financing wave. Watch Q3 filings from Intel and Microsoft for that distinction; right now, most of it is still bundled into one number.

ShareXLinkedInEmail

More on

SpaceX →Nvidia →

Reported by Value Add Pulse Analysis · Analysis by Value Add Pulse.

← Back to Pulse

THE WIRE in your inbox— Tech, startup & VC news with Trace's take. Free, no spam.

Read Next

BIG TECH· Aug 17, 2026

Alibaba's $1.5B Gaming Sale Bankrolls Its AI Pivot

Illustration for: Alibaba's $1.5B Gaming Sale Bankrolls Its AI Pivot
BIG TECH$1.5B+

Alibaba's $1.5B Gaming Sale Bankrolls Its AI Pivot

Alibaba is selling its gaming studio Lingxi Games to private equity firm Trustar Capital for more than $1.5 billion, the clearest sign yet that CEO Eddie Wu is stripping the conglomerate down to AI and cloud.

BIG TECH· Aug 16, 2026

Stripe Buys OpenRouter in $7B AI Gateway Deal

Illustration for: Stripe Buys OpenRouter in $7B AI Gateway Deal
BIG TECH$7B+ acquisition

Stripe Buys OpenRouter in $7B AI Gateway Deal

Stripe has agreed to acquire OpenRouter, the routing layer that lets developers switch between AI models through one API, for more than $7 billion -- roughly five times the $1.3 billion valuation OpenRouter carried in May.

BIG TECH· Aug 16, 2026

Zuckerberg's 6,500-Word AI Manifesto Isn't Landing

Illustration for: Zuckerberg's 6,500-Word AI Manifesto Isn't Landing
BIG TECH

Zuckerberg's 6,500-Word AI Manifesto Isn't Landing

Mark Zuckerberg's essay 'The Future is for Everyone' pitched Meta's Glimmer and Muse Spark models as universal personal AI, but critics questioned both the messenger and whether hardware requirements make the vision accessible at all.

@Trace_Cohen·t@nyvp.com