Nvidia to Buy $3B of Meta as Supplier Turns Shareholder

By Daniel Rao

Meta spent Monday selling its own stock to the world. On Tuesday, its chip supplier decided to buy some. That is the whole story, and it is stranger than it sounds: Nvidia, the company whose processors Meta must have to build its AI future, is about to become a shareholder in the customer whose spending keeps its own factories humming. When the seller and the supplier start swapping places, the AI boom has entered a new phase — one where the money flows in a circle, and everyone in the circle insists it is not a circle at all.

The announcement itself was terse, as these things always are. Nvidia and Meta said on Tuesday they would deepen their partnership on Muse, Meta's personal AI assistant, and that Nvidia would invest $3 billion in Meta's equity offering. It is a small number next to the sums Meta is throwing around this month. It is a very large statement about who believes in whom.

One wrinkle first: some wire copies of Nvidia's release carry a 26 October date; the news itself broke on Tuesday, in step with Reuters and Dow Jones reporting the same day. If you see both dates floating around, that is why.

What changed on Tuesday — and what did not

We covered Meta's $95 billion plan on Monday, when it launched about $35 billion of public offerings and sketched a further $60 billion programme for later. Tuesday did three new things: it put a firm Nvidia number on the table, it priced and upsized the offerings underneath that plan, and it tied the money explicitly to Muse — the product on which Mark Zuckerberg has staked Meta's next decade.

What Tuesday did not do is just as important. It did not raise the total. Nvidia's $3 billion sits inside Meta's offering, not on top of it. It did not turn AMD or Broadcom into buyers — more on that below, because the distinction matters. And it did not, by itself, sell a single extra chip. A dollar invested in Meta's stock is not a dollar of Nvidia chip revenue, however tempting it is to add the two together in a headline.

“AI agents represent a new computing paradigm — software that understands our goals, reasons, uses tools and gets work done on our behalf,” said Jensen Huang, founder and chief executive of Nvidia, in the joint statement. “NVIDIA and Meta are expanding our partnership to optimize the computing infrastructure end to end for capable, efficient and secure AI agents, so Muse can empower billions of people to achieve more and pursue greater ambitions.”

“Muse is how we bring personal superintelligence to everyone,” said Mark Zuckerberg, founder and chief executive of Meta, in the same statement. “We're excited to keep working with NVIDIA to deliver the best personal agents and frontier models for billions of people.”

Huang is selling a computing shift; Zuckerberg is selling a destiny. Both need the other to be right.

Photo: Jensen Huang, Nvidia's founder and chief executive. File photo — Peter Dasilva / European Union, via Wikimedia Commons, CC BY 4.0. This is a 2024 portrait, not a photograph of Tuesday's announcement.

The $97.5 billion, taken apart

The numbers deserve a slow read, because the headline figure is doing a lot of heavy lifting. On Monday 26 October, Meta launched roughly $35 billion of public offerings. On Tuesday, the priced version of that launch came in bigger: $25 billion of Class A common stock, upsized from $22.5 billion, sold at $625 a share through forward sale agreements with Morgan Stanley and Goldman Sachs, plus $12.5 billion of depositary shares — 250 million of them at $50 each — representing mandatory convertible preferred stock. Together, about $37.5 billion. The depositary shares are expected to trade on Nasdaq under the ticker METAM, pay a 6.50% annual dividend, and convert into Meta stock around August 2029.

Then comes the rest of the $97.5 billion, and here the language matters. Meta plans a separate at-the-market programme of up to $60 billion, using new non-voting Class C shares that are not expected to begin trading until the first quarter of 2027. No agreement for that programme has been signed, and its timing, size and terms have not been finalised. So the honest arithmetic is $37.5 billion priced plus $60 billion planned. The $97.5 billion is the largest fundraising plan in Meta's history — Reuters calls it that — but only the first tranche is a deal that has actually happened.

The plumbing has one more quirk. Both priced offerings are expected to close on Thursday 29 October, but Meta does not receive the common-stock cash at closing: the $25 billion of Class A stock is sold through forward agreements, and Meta gets paid when it settles them, on dates of its choosing, by 19 November 2027 at the latest. Meta expects about $24.9 billion of net proceeds from that leg when it settles. The preferred stock, by contrast, is cash at closing.

Why does Meta need any of this? Look at the third quarter. Revenue of $64.6 billion, up 26% on a year earlier. Capital spending of $34.91 billion against operating cash flow of $32.79 billion, leaving free cash flow at negative $2.12 billion. A cash pile of $82.45 billion at 30 September. And a full-year capital spending guide of $135 billion to $145 billion. A company can be enormously profitable and still run out of room — that is the whole paradox of the AI buildout, and it is why the equity market, not just the bond market, is now funding it.

The supplier becomes a shareholder

Now the part of the story that will travel furthest. When Meta launched its offerings on Monday, it named Nvidia, AMD and Broadcom as potential cornerstone investors, with non-binding indications of interest in buying up to $5 billion of the Class A shares between them, at the same price and on the same terms as everyone else. Non-binding is doing real work in that sentence. Any of the three could buy more, fewer, or none. The underwriters could sell them more, fewer, or none.

Tuesday turned one corner of that possibility into something firmer. Nvidia's $3 billion is an announced investment in the offering, subject to closing conditions, as the company's own release puts it. Read the release to the bottom and the tone changes: pages of forward-looking-statements boilerplate warning that the investment, its completion and its conditions are all projections, not guarantees — a reminder that “will invest” and “has invested” are not yet the same sentence.

What Tuesday did not do is confirm AMD or Broadcom: Nvidia's release does not mention them, and the pricing reports we checked do not say how much, if anything, either company bought. Anyone telling you the $5 billion cornerstone is fully subscribed is getting ahead of the filings.

We checked the release against Meta's pricing terms and the Reuters and Dow Jones reports before writing this, because the distinction between “inside the offering” and “on top of it” is where most of the early coverage goes wrong. Nvidia is buying into the same $37.5 billion that everyone else is buying into. It is not writing a separate cheque on top. The symbolism survives that correction intact. The arithmetic does not need exaggerating.

Reuters noted on Monday that the chipmakers' funding interest “adds to concerns about an increasingly circular flow of capital in the AI industry, as chip suppliers help finance the expansion of one of their biggest customers.” That is the right word, and it is worth sitting with. Meta buys Nvidia's chips. Nvidia buys Meta's stock. Meta uses the proceeds, in part, to buy more of Nvidia's chips. Each step, on its own, is defensible. Draw the whole loop and it looks like an industry funding itself — which is fine while the music plays and is exactly the arrangement that deserves scrutiny before it stops.

Keep the proportions honest: $3 billion is small next to $37.5 billion priced, let alone $97.5 billion planned. This is not a bailout. It is a flag in the ground — Nvidia telling every other investor what the best-informed player in the room thinks.

Photo: The Meta sign at 1 Hacker Way, Menlo Park. File photo — Nokia621, via Wikimedia Commons, CC BY-SA 4.0.

Muse is the reason, not the excuse

This is not just a financing story with an AI label stuck on it. Muse, known internally as Hatch, is the assistant Meta launched in September as the consumer face of Zuckerberg's “personal superintelligence” plan. Modelled on the open-source agent OpenClaw, it is designed to reach into a user's email, calendar, payments, health, shopping and smart-home apps — with the user choosing the connections and able to revoke access at any time. A basic version is free; heavier use costs $20 or $100 a month.

Meta has pointed to more than 10 million Muse users in the US and Canada in its first weeks — the company's own claim. But it explains the urgency: an agent that does things, rather than merely answering questions, burns far more computation per user than a chatbot, and every task it completes is inference work on somebody's hardware. Muse's economics are, in the end, a hardware bill with a subscription attached.

Tuesday's announcement is about making that bill smaller. Meta's Muse models and agents already run on Nvidia infrastructure. The two companies say they will now work more closely on optimising training and inference across workflows, to maximise performance and improve total cost of ownership — the kind of unglamorous sentence that decides whether an agent business makes money or merely makes headlines. Meta intends to expand deployment of Nvidia's Vera Rubin platform and Vera CPUs for next-generation agents, plus Spectrum-X and BlueField-4 networking. They will also consider working together on Nvidia's OpenShell framework for safe software deployment and a tool called Sentry for monitoring how agents behave — both still at the “considering” stage, not the committed stage, and the distinction is worth preserving.

None of this is a new relationship. In February 2026, the two companies announced a multiyear, multigenerational partnership spanning Meta's on-premises and cloud infrastructure, including millions of Blackwell and Rubin GPUs, Grace CPUs, Spectrum-X networking and confidential computing for WhatsApp. “We're excited to expand our partnership with NVIDIA to build leading-edge clusters using their Vera Rubin platform to deliver personal superintelligence to everyone in the world,” Zuckerberg said at the time. Tuesday is that partnership growing a financial limb. The technology came first. The equity is the vote of confidence on top of it.

What Nvidia actually gets

It is worth asking what Nvidia is buying, beyond a $3 billion stake whose exact tranche — common, preferred, or both — the release does not disclose. For scale, the offering’s common stock was priced at $625 a share, a discount to Friday's $642.62 close, with Meta shares hovering near $631 on Tuesday morning. The first answer is alignment. A customer that has committed, by our earlier reporting, to roughly $825 billion of future spending — mostly data centres, AI infrastructure and cloud capacity — is not a customer you leave to chance. Owning a piece of it aligns Meta's success with Nvidia's balance sheet.

The second answer is influence over the cost curve. Agents are where Nvidia's next decade of demand will come from, if the agent thesis holds — and nobody learns what agents actually cost to run faster than the company running one of the largest consumer deployments. The joint optimisation work, the Vera Rubin deployment, the OpenShell and Sentry experiments: these are Nvidia shaping the platform its hardware will run on, with its biggest agent customer as the laboratory.

The third answer is the one Nvidia will not say out loud. The AI infrastructure trade has spent October under a microscope — we wrote last week about a neocloud borrowing $3.5 billion for a single year despite $40 billion of contracts — and every vendor-financing arrangement now gets read as a stress signal. Nvidia buying Meta stock cuts the other way: it says the vendor with the fullest order book is happy to own the customer, not merely lend to it. Equity is a stronger statement than credit, and one Nvidia's hardware margins can comfortably afford.

What is still unconfirmed

What we do not know, plainly stated. First, AMD and Broadcom: unconfirmed as buyers, in any amount. Second, the final shape of Nvidia's $3 billion: which tranche it sits in, how many shares it represents, and what percentage of Meta it leaves Nvidia holding — none of that is in the release. Third, the $60 billion at-the-market programme: unsigned, unpriced, and not expected to begin until Class C shares trade in early 2027. Fourth, the OpenShell and Sentry work: under consideration, not committed. Fifth, the Muse performance claims: the release speaks of “frontier-level benchmark performance” and “surging consumer downloads,” which are the companies' characterisations, not independent measurements. Treat them as such.

Does this make Meta's spending safer? It makes its funding story stronger — a $37.5 billion priced deal with its own supplier buying in is not a company struggling to find buyers. It does not make the spending itself smaller, cheaper, or more certain to pay off. The fourth quarter still implies roughly $49 billion to $59 billion more capital spending before Christmas on Meta's own full-year guidance. Nvidia's cheque does not change any of that. It changes who is standing next to Meta while it spends.

Our take: alignment is good, circularity is the question

So is this a smart deal or a warning sign? It is both, and the honest answer depends on which seat you sit in. For Meta, it is good news: the deepest-pocketed, best-informed supplier in AI just validated the spending programme and the product at its centre, in public, with money. For Nvidia, it is a shrewd piece of vertical diplomacy: a small equity stake that buys alignment, insight and a front-row seat in the agent race, at a price below last week's close. Neither company is behaving irrationally.

The discomfort is real and should not be waved away. When suppliers become shareholders, the market's usual discipline gets softer: a customer that owns its vendor's favour, and a vendor that owns its customer's stock, have fewer reasons to ask each other hard questions about price, volume and returns. The AI industry has spent this year discovering that its financing loops — debt, leases, commitments, now equity — are more tangled than anyone admitted. This deal adds a new strand to the tangle, at the very top of the food chain.

But let us not overclaim in the other direction either. Three billion dollars does not make Meta dependent on Nvidia's goodwill, and it does not make Nvidia dependent on Meta's share price. Both companies are vastly larger than this transaction. The circularity here is a shape to watch, not a scandal to declare — the scandal, if one comes, will be in deals where the circle is the whole business, where a vendor's revenue exists only because it financed the customer. This is not that deal.

What to watch next is concrete. First, the filings: whether AMD and Broadcom appear as buyers, and what Nvidia's final holding looks like. Second, the 29 October close. Third, the Class C timetable — whether those shares really trade in early 2027 and whether the $60 billion programme is ever signed. Fourth, Muse itself: whether the optimisation work shows up as lower cost per task, or merely as better press releases. And fifth, January's capital spending guidance.

If the agent race is decided by cost per task — and we believe it will be — then Tuesday's deal is best read as a cost play dressed as a financing play. Nvidia is not just betting that Meta will keep buying chips. It is betting that the two of them, working on the same stack, can make each agent interaction cheap enough that billions of people will use one every day. It is also, as of Tuesday, a bet Nvidia has made with its own money. That, more than any press release, is the news.

Sources: Nvidia–Meta joint announcement via GlobeNewswire (reported Tuesday 27 October 2026; some wire copies carry a 26 October date); Reuters on the partnership and Meta's upsized offering (27 October 2026); Dow Jones Newswires via Morningstar (27 October 2026); Meta's offering announcements and pricing terms (26 October 2026); Nvidia–Meta infrastructure partnership announcement (17 February 2026); Meta's third-quarter 2026 results, as covered in TechPulse USA's 27 October $95 billion-plan post. The chipmakers' $5 billion interest is non-binding; AMD and Broadcom purchases are unconfirmed; the $60 billion programme has no signed agreement.

By Daniel Rao — TechPulse USA covers the collision of AI, power and policy. Tips: we read every message.