AMD is handing OpenAI and Meta up to 20% of the company to lock in their AI chip orders

Every figure in this article is drawn from AMD's earnings releases and investor announcements, or from Reuters and market-data coverage as attributed. Warrant terms beyond AMD's own releases are attributed to press reporting of the securities filings. Sections marked as analysis are identified as such.

For thirty years, AMD's identity was the other chipmaker — the discount alternative, perpetually one node and one software stack behind. Its June quarter reads like the identity finally cracking: revenue of $11.5 billion, up 50 per cent, with data-centre revenue of $6.7 billion — more than doubled in a year — net income up 163 per cent, and guidance to roughly $13 billion next quarter. "We delivered an excellent quarter, with record revenue and profitability as Data Center revenue more than doubled year-over-year," chief executive Lisa Su said in the release.

The stock has answered: up roughly 126 per cent this year, a market value around $760 billion — a company worth about $3 billion when Su took over in 2014. And the engine behind the re-rating is a pair of deals unlike anything else in the industry, in which AMD's biggest customers are also, potentially, among its biggest shareholders.

The warrant machine

Last October, AMD and OpenAI announced a partnership to deploy six gigawatts of AMD Instinct GPUs, beginning with a gigawatt of the new MI450 systems in the second half of this year — "tens of billions of dollars in revenue," in AMD's words. Attached to it: a warrant granting OpenAI up to 160 million AMD shares — roughly 10 per cent of the company — vesting in tranches tied to deployment milestones, AMD's share price, and OpenAI's own commitments, at a reported one-cent exercise price with the final tranche tied to a $600 share price. The stock leapt by roughly a quarter on the announcement.

In February, the structure was duplicated: Meta signed for six gigawatts of AMD GPUs — custom MI450-based silicon plus AMD's Helios rack systems, first gigawatt shipping this half — with a second warrant for up to another 160 million shares on the same milestone logic. AMD disclosed no dollar value; the multibillion-dollar estimates circulating are press arithmetic, not company figures. "I expect AMD to be an important partner for many years to come," Mark Zuckerberg said in the announcement.

Around the two anchors, a genuine customer list has formed: Oracle taking 50,000 MI450 chips from this autumn, Anthropic signing for up to two gigawatts alongside an AMD equity investment of as much as $5 billion in the lab itself, Microsoft deploying next-generation Instinct systems, and the Helios rack — 72 GPUs and AMD's 2-nanometre EPYC processors per cabinet — now in full production, with AMD claiming up to 30 per cent more inference output per dollar than Nvidia's equivalent. The claim is the vendor's own; the order book is not.

The gap that remains

Honest scale-keeping matters here. AMD's record AI quarter of $6.7 billion in data-centre revenue landed in the same season Nvidia reported $81.6 billion of total revenue — AMD's entire triumph is roughly a tenth of its rival's business. AMD's own long-term targets, set at its November analyst day, frame the ambition: data-centre AI growing more than 80 per cent a year against a "$1 trillion compute market," with earnings per share above $20. The pitch to customers remains price-performance rather than supremacy — tokens per dollar, open software, a credible second source — which is precisely why the customers desperate to discipline Nvidia's pricing are willing to pay for AMD's success in warrants.

China, meanwhile, remains a taxed and flickering market: an $800 million inventory write-down when export controls hit its MI308 chip in 2025, a licensing regime that reportedly hands 15 per cent of China chip revenue to the U.S. government, and — per AMD's own releases — no MI308 China revenue recognised in recent reported quarters.

The market's new demand

The sharpest recent signal came after the record quarter: the stock fell for two straight sessions in mid-August, as coverage converged on a single anxiety — the customers AMD signed are the same companies designing their own chips. Meta builds in-house silicon and rents Google's; OpenAI is co-developing custom accelerators with Broadcom. At a $760 billion valuation, a 50 per cent growth quarter and a $13 billion guide no longer guarantee applause. The expectations treadmill that has long governed Nvidia now runs under AMD too.

Analysis: what the equity actually bought

The following section is analysis, drawn from the reported facts above.

The warrant structure deserves to be evaluated as the genuine innovation it is. AMD faced the second-source paradox: customers wanted a credible Nvidia alternative but would not commit the volumes that make one credible. Su's answer was to sell them the upside — if OpenAI and Meta deploy the gigawatts and AMD's stock reaches the milestone prices, they collect up to a fifth of the company between them, at a negligible strike. Demand certainty was bought with dilution. Every AMD shareholder's stake shrinks a little so that the order book could exist at all.

The bargain is defensible arithmetic: 20 per cent of a company transformed by anchored, multi-year AI demand is worth vastly more than 100 per cent of the perpetual runner-up. The tranche design aligns everyone — the warrants pay out only if the buyers actually deploy and the stock actually rises. But it also binds AMD's fate to two customers whose in-house silicon programmes are explicitly designed to need less of what AMD sells, and it caps the fairy-tale scenario: if AMD truly wins, a meaningful slice of the winnings was pre-sold at a penny.

Su's decade has been the industry's best rebuttal to inevitability — Nvidia's dominance looked total in 2023, and AMD now ships gigawatts to the most demanding buyers on Earth. The record quarter is real, the products are competitive, and the doubters have been wrong about her before, expensively. What the next two years decide is narrower: whether the second source becomes a true peer, or whether AMD has brilliantly monetised a moment in which its customers needed leverage more than they needed AMD. Chips for shares is a trade both sides entered with open eyes. The market is still arguing about who got the better of it — and that argument is Lisa Su's greatest achievement to date.

Sources

• AMD Q2 2026 earnings release, August 4, 2026 — revenue $11.54B (+50%), data-centre $6.72B (+107%), GAAP net income $2.30B (+163%), margins, Q3 guidance ~$13.0B; Su quote.
• AMD Q4/FY2025 release, February 3, 2026 — fiscal 2025 revenue $34.6B, data-centre $16.6B; Financial Analyst Day release, November 11, 2025 — long-term targets (>35% revenue CAGR, >80% data-centre AI CAGR, >$20 EPS, $1T compute market).
• AMD press release, October 6, 2025 — the OpenAI 6GW partnership, first MI450 gigawatt in H2 2026, 'tens of billions' revenue expectation, warrant for up to 160M shares with milestone tranches; exercise-price and $600-tranche details per press reporting of the securities filings; the ~24% share-price reaction as widely reported.
• AMD press release, February 24, 2026 — the Meta 6GW partnership and second 160M-share warrant; Zuckerberg quote; no company dollar value disclosed.
• Reuters (via syndication), October 14, 2025 — Oracle's 50,000-MI450 deployment from Q3 2026; AMD releases, July 20–23, 2026 — the Microsoft deployment, the Anthropic 2GW agreement with up to $5B equity investment, and the Helios launch with AMD's stated performance-per-dollar claims.
• AMD 2025 quarterly releases — the $800M MI308 export-control charge and statements that recent quarters included no MI308 China revenue; the 15% China revenue-share arrangement as widely reported in August 2025.
• Market data as of August 19, 2026 — share price ~$465, market value ~$759B, ~+126% year to date, and the two-session post-earnings decline; Nvidia's $81.6B quarter (ended April 26, 2026) from its own reported results, for scale.

Figures are in U.S. dollars as reported. Warrant exercise-price and final-tranche terms come from press accounts of the filings rather than AMD's releases. The Meta arrangement's dollar value is undisclosed; circulating estimates are excluded. AMD-versus-Nvidia comparisons use overlapping but offset fiscal quarters and are indicative of scale, not precise share.