The Information Machine

AMD Warrant Deals with Meta and OpenAI Function as Disguised Compute Subsidies

open · v1 · 2026-08-03 · 17 items

What

AMD struck performance-based warrant deals with OpenAI (late 2025) and Meta (early 2026), granting both companies warrants that vest as they purchase and deploy up to 6 GW of AMD Instinct GPUs, with vesting also contingent on AMD's share price crossing hurdles stepping up to $600 [1]. Financial analyst firm SemiAnalysis argues the conventional 'equity sweetener' framing misses the economic substance: because the warrants carry penny strike prices, they function more like compute rebates, with implied effective discounts of 85% to 105% at AMD stock prices near $600 [5][6]. AMD's stock has already risen from roughly $200 in early 2026 to $500-600 by August 2026, reducing Meta's effective GPU capital cost from ~$2.15/hr to ~$1.60/hr and OpenAI's from ~$2.50/hr to ~$1.75/hr [7]. SemiAnalysis projects that if AMD reaches $600-700, GPU-only capital cost would be effectively zero for both companies; above $1,000, the all-in cluster cost including servers, networking, and CPUs would also be zero [8].

Why it matters

If SemiAnalysis's analysis is accurate, AMD is subsidizing compute costs for the two largest AI labs using its own equity rather than cash — a structure that works only if the volume commitments themselves drive AMD's stock high enough to justify the dilution. This makes AMD's competitive position against Nvidia dependent not just on hardware and software, but on a self-reinforcing stock price dynamic.

Open questions

  • Is the circular logic self-sustaining — do Meta's and OpenAI's volume commitments actually drive AMD stock high enough that the implied warrant discounts don't destroy AMD shareholder value? [5]

  • At what AMD stock price do the warrants become materially dilutive to existing shareholders, and have sell-side analysts modeled this? [6][9]

  • Have Meta or OpenAI publicly characterized the effective economics of these deals, or does SemiAnalysis's rebate framing go uncontested by the parties themselves? [7]

  • Does AMD's software stack progress against Nvidia's CUDA ecosystem keep pace with its financial incentives — if not, will the volume commitments materialize to sustain the stock price that makes the warrant math work? [8][5]

Narrative

In late 2025 and early 2026, AMD struck large warrant agreements with OpenAI and Meta tied to compute commitments. Both companies receive performance-based warrants that vest as they purchase and deploy up to 6 GW of AMD Instinct GPUs, with vesting also contingent on AMD's share price crossing a series of hurdles stepping up to $600 [1]. When the OpenAI deal was announced, AMD stock surged more than 34%, and the Meta deal — publicly framed as a $60 billion, 6 GW AI chip partnership — drew comparable market attention [2][3][4]. Initial coverage treated both deals as equity sweeteners: a way for AMD to give its most important AI-lab customers a stake in AMD's upside in exchange for supply commitments.

SemiAnalysis argues that framing understates the economic substance. Because the warrants carry penny strike prices — effectively zero exercise cost — and vest in tranches tied to AMD's share price, they function as a purchase rebate rather than a conventional equity grant. SemiAnalysis calculates that at AMD stock prices near $600, the implied effective discount on GPU compute is between 85% and 105% depending on the tranche; at 105%, AMD is paying the customer to take its chips and funding the subsidy with its own equity [5][6]. The firm calls the arrangement circular: it works only if volume commitments drive AMD's share price high enough to justify the dilution, which is the same appreciation that makes the warrants valuable to Meta and OpenAI [5].

AMD's stock has already moved enough to make the math concrete. From roughly $200 in early 2026, AMD shares climbed to $500-600 by August 2026 [7]. That appreciation has already reduced Meta's effective GPU capital cost from approximately $2.15/hr/GPU to $1.60/hr/GPU, and OpenAI's from approximately $2.50/hr/GPU to $1.75/hr/GPU — assuming vesting is roughly linear relative to AMD's price level and that the full 6 GW capacity is deployed [7]. SemiAnalysis projects that if AMD stock reaches $600-700, the GPU-only capital cost would be effectively zero for both companies; if AMD exceeds $1,000, the all-in cluster cost including servers, networking, and CPUs would also reach zero [8].

The broader strategic context SemiAnalysis points to is AMD's effort to erode Nvidia's CUDA software moat. The warrant structure gives Meta and OpenAI a strong financial incentive to commit volume to AMD hardware, but the long-term viability of the arrangement depends on AMD making enough software progress that those customers would deploy at scale regardless of the financial subsidy. No public response from AMD, Meta, or OpenAI to SemiAnalysis's rebate framing has appeared in the sourced material.

Timeline

  • 2025-10: AMD and OpenAI announce a strategic partnership to deploy 6 GW of AMD GPUs; OpenAI is initially reported as seeking a ~10% stake in AMD through the deal. [2][13]
  • 2025-10: Latham Watkins advises AMD on the warrant agreement with OpenAI. [14]
  • 2026-02-24: AMD and Meta announce a $60 billion, 6 GW AI chip partnership; AMD stock surges on the announcement. [3][10][12]
  • 2026-early: AMD share price is approximately $200, establishing the baseline against which subsequent warrant-implied discounts are calculated. [7]
  • 2026-07-27: SemiAnalysis publishes a thread reframing AMD's warrant deals as compute rebates rather than equity sweeteners, calculating effective discounts of 85%-105% at AMD stock near $600. [5][6]
  • 2026-08-03: SemiAnalysis publishes a follow-up thread with realized cost-of-ownership calculations: with AMD stock at $500-600, Meta's effective GPU cost has dropped to ~$1.60/hr and OpenAI's to ~$1.75/hr, and projects zero GPU capital cost if AMD reaches $600-700. [1][7][8]

Perspectives

SemiAnalysis

AMD's warrant deals are structured as disguised compute rebates. At AMD stock prices near $600, the effective discount on GPU compute reaches 85%-105%, and AMD is funding customer compute costs with its own equity. The strategy is circular — it works only if the volume commitments sustain AMD's stock appreciation — and depends on AMD closing the software gap with Nvidia.

Evolution: Consistent and deepening across the thread; SemiAnalysis initiated this reframing and has published multiple installments with increasingly specific quantitative projections.

AMD (institutional)

Framed the deals publicly as strategic partnerships aligning AMD with its largest AI-lab customers through equity participation; no public rebuttal of the rebate characterization has appeared.

Evolution: Consistent with initial deal framing; no known engagement with SemiAnalysis's analysis.

Mainstream financial press (Reuters, Bloomberg, Barron's, Forbes)

Reported the deals as large compute supply agreements with warrant components; Barron's flagged potential warrant dilution concerns for AMD shareholders alongside the positive deal news.

Evolution: Initial coverage was broadly positive on deal scale; some skepticism about warrant dilution appeared but did not dominate coverage.

Tensions

  • SemiAnalysis argues the warrants are functionally compute rebates worth 85%-105% of GPU purchase price at current AMD stock levels; mainstream press and AMD's own framing treat them as equity sweeteners that align customer and company interests — the two framings carry different implications for AMD's financial health and shareholder dilution. [5][6][9]
  • SemiAnalysis describes the AMD warrant strategy as circular — the deal pays off only if volume drives stock, which is what makes the warrants valuable — but AMD has not publicly addressed whether this self-reinforcing logic is the intended design or a structural vulnerability. [5][1]
  • The full economic benefit to Meta and OpenAI is contingent on AMD making meaningful software progress against Nvidia's CUDA ecosystem; if AMD fails on software, the volume commitments may not materialize, and the circular logic breaks regardless of stock price. [5][8]

Status: active and growing

Sources

  1. [1] Late last year and earlier this year, AMD struck deals with Meta and OpenAI to provide them with performance-based warra… — SemiAnalysis Twitter (2026-08-03)
  2. [2] AMD and OpenAI Announce Strategic Partnership to Deploy 6 Gigawatts of AMD GPUs :: Advanced Micro Devices, Inc. (AMD) — reactive:amd-warrant-compute-deals
  3. [3] AMD and Meta announce $60B, 6 GW AI partnership – Jon Peddie Research — reactive:amd-warrant-compute-deals
  4. [4] AMD signs AI chip-supply deal with OpenAI, shares surge over 34% — reactive:amd-warrant-compute-deals
  5. [5] Push the warrant rebate to the top tranche — AMD at $600, everything vested, warrants exercised for a penny — and the im… — SemiAnalysis Twitter (2026-07-27)
  6. [6] AMD's warrant deals with OpenAI and Meta usually get described as equity sweeteners. Run the math, and these look more l… — SemiAnalysis Twitter (2026-07-27)
  7. [7] With AMD’s share price shooting up from the ~$200 range in early 2026 to around $500-600 range now, this means that Meta… — SemiAnalysis Twitter (2026-08-03)
  8. [8] If AMD’s share price rises to the $600-700 range – the capital cost for the GPUs only will effectively be zero for both!… — SemiAnalysis Twitter (2026-08-03)
  9. [9] AMD Stock Surges After Chip Deal With Meta. Too Bad About the Warrants. — reactive:amd-warrant-compute-deals
  10. [10] AMD clinches second mega chip supply deal, this time with Meta — reactive:amd-warrant-compute-deals
  11. [11] Meta to Spend Billions on AMD Gear, AI Scare Trade Continues | Bloomberg Tech 2/24/2026 — reactive:amd-warrant-compute-deals
  12. [12] Meta and AMD Agree To Massive AI Chips Deal — reactive:amd-warrant-compute-deals
  13. [13] OpenAI looks to take 10% stake in AMD through AI chip deal — reactive:amd-warrant-compute-deals
  14. [14] Latham Watkins Advises AMD on Warrant Agreement With Open AI — reactive:amd-warrant-compute-deals