AMD has two agreements under which OpenAI and Meta each intend to deploy up to 6 gigawatts of AMD Instinct GPUs, up to 12 gigawatts in total by our arithmetic. Each came with a warrant for up to 160 million AMD shares at $0.01, according to AMD’s 8-K filings of 6 October 2025 and 24 February 2026.
The 320 million shares equal about 19.5% of the 1,640 million AMD shares outstanding at the end of the third quarter, by our arithmetic. This analysis puts together how the deals are structured, how the warrants vest, how they reach AMD’s accounts and where the first deployments stand.
Business Pill · Concentration
An evergreen Business Engineer explainer on concentration, when a few big pieces make the total. It teaches the general idea only and does not discuss AMD.
If all 320 million warrant shares were issued on top of the 1,640 million AMD shares outstanding at 26 September 2026, OpenAI and Meta would hold 8.2%% each, 16.3%% together, ignoring other share changes. Source: AMD 8-Ks of 6 October 2025 and 24 February 2026 and its third-quarter 2026 10-Q; our arithmetic. Chart: FourWeekMBA.
The key insight: By our arithmetic from AMD’s filings, its OpenAI and Meta agreements cover up to 12 gigawatts (6 each) of Instinct GPUs and up to 320 million warrant shares, about 19.5% of shares outstanding; the shares vest only as purchases hit gigawatt milestones and AMD’s stock meets targets rising to $600, and none had vested by 26 September.
The OpenAI agreement, announced on 6 October 2025, covers 6 gigawatts across multiple generations of Instinct GPUs. AMD’s release said “The first 1 gigawatt deployment of AMD Instinct MI450 GPUs is set to begin in the second half of 2026”. The 8-K says OpenAI agreed to a binding commitment to purchase that first gigawatt.
The Meta agreement, announced on 24 February 2026, also covers up to 6 gigawatts. The release says the first deployment uses “a custom AMD Instinct GPU based on the MI450 architecture”, with shipments for the first gigawatt “scheduled to begin in the second half of 2026”, alongside 6th Gen EPYC CPUs codenamed Venice on the Helios rack-scale architecture. Meta likewise agreed to a binding commitment for the first gigawatt equivalent.
AMD chief financial officer Jean Hu said in the October 2025 release that the OpenAI partnership “is expected to deliver tens of billions of dollars in revenue for AMD”. The filings we read give no contract value.
AMD’s warrants for up to 320 million shares at $0.01: 160 million for OpenAI and 160 million for Meta, each tied to up to 6 gigawatts of Instinct GPU purchases; together about 19.5% of the 1,640 million shares outstanding at 26 September 2026. Source: AMD 8-Ks of 6 October 2025 and 24 February 2026 and its third-quarter 10-Q; the share is our arithmetic.
How the Warrants Vest
Both warrants vest in tranches tied to purchases of Instinct GPUs. For OpenAI, the 8-K says the first tranche vests “after the delivery of the initial one (1) gigawatt of AMD Instinct MI450 Series GPU products”; for Meta, “upon shipment of the initial one (1) gigawatt equivalent of certain AMD Instinct GPU products”. Full vesting of each 160 million shares requires purchases of 6 gigawatts.
Vesting also depends on AMD’s share price. The OpenAI 8-K cites “specified Company stock price targets that escalate to $600 per share for the final tranche and stock performance thresholds”; the Meta 8-K cites thresholds that also “escalate to $600 per share for the final tranche”. Each vested tranche must then meet technical and commercial conditions before it can be exercised.
The OpenAI warrant can be exercised until 5 October 2030 and the Meta warrant until 23 February 2031.
How They Reach AMD’s Accounts
AMD treats the warrants as consideration payable to a customer. The grant-date value of shares probable of vesting is recorded as a reduction to revenue as qualifying sales are recorded, and the liability is remeasured each quarter through other income (expense), the third-quarter 10-Q says.
At 26 September the warrants’ weighted-average fair value was $178.58 a share, against $14.25 at grant, about 12.5 times by our arithmetic, and the liability was $1.2 billion, as we reported from the 10-Q. The quarter’s remeasurement loss was $1,097 million, AMD’s first warrant charge, and the 10-Q says $4.5 billion of grant-date value would be recorded as a reduction of revenue as qualifying customer revenue is recognised.
The 10-Q says that as of 26 September “no warrant shares had vested or become exercisable, or been exercised, forfeited or expired”.
Where the Deployments Stand
AMD’s third-quarter release says “Anthropic, Meta, Microsoft, OpenAI and Oracle Cloud Infrastructure are running models and workloads on the first AMD Helios rackscale solutions”, powered by MI450 GPUs and 6th Gen EPYC CPUs. AMD also said OpenAI’s first custom AI chips run on EPYC host CPUs, as we reported on 3 November.
OpenAI is also building its own chips: its plan with Broadcom covers 10 gigawatts, about 1.7 times its AMD agreement by our arithmetic, as our follow-up found.
As we read the 8-Ks, each warrant has three gates: GPU purchases that start with the first gigawatt and run to six, AMD share-price targets that end at $600 for the final tranche, and technical and commercial conditions before any vested tranche can be exercised.
As we read the 10-Q, the deals already reach AMD’s accounts before any share vests: grant-date value comes off revenue as qualifying sales are recorded, and the liability moves with the share price each quarter.
AMD, Form 8-K, 6 October 2025
“specified Company stock price targets that escalate to $600 per share for the final tranche and stock performance thresholds”
Three Implications
FOR AMD’S REVENUE The 10-Q says $4.5 billion of grant-date value would be recorded as a reduction of revenue as qualifying customer revenue is recognised.
FOR THE SHARE COUNT Up to 320 million shares could be issued, about 19.5% of the 1,640 million outstanding at 26 September, by our arithmetic.
FOR DEPLOYMENT TIMING Both agreements start with a first gigawatt of MI450-based GPUs from the second half of 2026, according to AMD’s releases.
The framework says: “Big Tech’s hyperscalers are racing to build gigawatt-scale AI data centers, with five facilities set to exceed 1GW by 2026, each requiring a nuclear reactor’s worth of power.”
As we read it, AMD’s agreements price that build-out in gigawatts: each customer’s warrant vests in steps measured in gigawatts of GPUs purchased, from one to six. The framework does not discuss AMD’s agreements.
The filings do not give the number of shares in each tranche, the share-price target for each tranche before the last, or the gigawatt milestones between the first and the sixth. They do not split the warrant values between OpenAI and Meta, and they give no total contract value. Whether a first tranche has vested since 26 September is not stated.
This Week In The AI Supercycle
Anthropic’s Reality Check, The Capital Squeeze, & The New World Order
This week’s issue: for two years the AI build-out was priced as a growthstory. This week it started being priced as a credit story.
AMD’s OpenAI and Meta agreements cover up to 12 gigawatts of Instinct GPUs by our arithmetic, starting with a first gigawatt of MI450-based GPUs each from the second half of 2026, and carry warrants for up to 320 million shares, about 19.5% of shares outstanding by our arithmetic. The shares vest only as purchases reach milestones and AMD’s stock meets price targets that rise to $600 for the final tranche; none had vested by 26 September.
A note on sourcing. Agreement and warrant terms are from AMD’s Form 8-K filings of 6 October 2025 and 24 February 2026 and the press releases filed with them; accounting figures are from AMD’s 10-Qs for the second and third quarters of 2026; deployment status is from AMD’s third-quarter 2026 earnings release. Figures attributed to earlier pieces come from those pieces. Totals and shares are our arithmetic.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
