The Chip-Tier Verdict — AMD Hands OpenAI and Meta 320 Million Shares for a Penny Each, Arista Pays in Margin, the Capex Thesis Holds

The Lead

Buried in AMD’s 6 May 10-Q is a disclosure that doesn’t exist anywhere else in our 4,669-row historical sample: AMD has issued warrants to OpenAI (October 2025) and Meta (February 2026), each providing the right to purchase up to 160 million AMD shares at an exercise price of $0.01 per share. Total potential dilution: ~320 million shares — approximately 20% of AMD’s diluted share count at current pricing.

The warrants vest in tranches based on (a) AMD Instinct GPU purchase milestones achieved by OpenAI and Meta and (b) specified AMD stock-price targets. The OpenAI Warrant is exercisable through 5 October 2030; the Meta Warrant through 23 February 2031.

Read carefully: this is vendor financing structured as equity. AMD is willing to give its two largest potential AI customers the right to buy 20% of the company at a penny per share, conditional on them buying enough GPUs and on AMD’s stock performing. Two implications:

1. AMD believes it can win share at NVDA’s expense. Companies don’t dilute 20% to compete in markets they’re losing — they dilute to lock in customers in markets they’re winning.
2. OpenAI and Meta are not going to be NVDA-only buyers anymore. The warrant structure only pays out for them if they actually buy enough AMD chips to trigger the milestones. They’ve now signed a contract aligning their incentives with AMD’s silicon roadmap, not Jensen’s.

Falsifiable form: AMD’s Data Center segment revenue should grow at a materially higher rate than NVDA’s data center revenue over the next 4-6 quarters. If by Q4 2026 AMD’s DC revenue isn’t visibly accelerating beyond Q1’s already-strong base (DC was the primary driver of the +38% revenue print), the warrants don’t vest, the dilution doesn’t happen, and the strategic shift implied by this filing was overstated. If AMD DC accelerates while NVDA decelerates, this warrant structure was the inflection point and the Issue 03 “MSFT-OpenAI exclusivity is dead” thesis extends to “NVDA-OpenAI/Meta exclusivity is dead.”

Threads

AMD’s $0.01 warrants — vendor financing as equity

The exact disclosure: “In October 2025 and February 2026, the Company issued warrants to OpenAI OpCo, LLC (OpenAI) and Meta Platforms, Inc. (Meta)… Each warrant provides the holder the right to purchase up to an aggregate of 160 million shares of the Company’s common stock at an exercise price of $0.01 per share.” (AMD 10-Q, 6 May 2026, Note on Equity)

What’s unprecedented isn’t that vendors give discounts — that happens. It’s the structure. By issuing warrants at $0.01 strike rather than discounting GPUs directly, AMD:

• Defers the cost (warrants vest on future milestones, not at signing)
• Capitalises the customer-acquisition cost on the balance sheet rather than in COGS (preserves gross margin)
• Aligns customer incentives multi-year (vesting tranches mean OpenAI / Meta benefit if they keep buying)
• Creates equity-holders out of the two most strategically important AI customers

The 320M-share dilution potential is the cost. AMD’s Q1 2026 share count was ~1.6B diluted. 320M = 20%. At AMD’s current trading range (let’s say $200/share), that’s **$64B of equity given away** in exchange for guaranteed multi-year GPU purchasing commitments from the two best customers in the industry.

This is a meaningful number in dollar terms but a smart one in strategic terms. $64B of dilution to lock in (a) the largest AI training customer in the world and (b) Meta’s 6GW commitment is cheap — Meta’s 6GW alone, if it converts to ~$25-40B of GPU purchases over 5 years, more than pays for the dilution.

Lab-calibrated: AMD signal qualifies as future_bet × bullish, the strongest historical setup in our 4,669-row sample (+2.5% mean direction-adjusted alpha at 20d, 57.5% hit rate, n=365). High confidence framing earned.

Meta commits 6 gigawatts of AMD GPUs — half of AVGO’s TPU thunder, gone

Same 10-Q, same paragraph but in a different note: “In February 2026, we amended a master purchase agreement with Meta Platforms, Inc. (Meta) and Meta agreed to deploy up to 6 gigawatts of AMD GPUs, with the first gigawatt of capacity powered by custom AMD Instinct MI450-based GPU and 6th Gen AMD EPYC™ CPUs.” (AMD 10-Q, 6 May 2026)

To put 6GW in context: that’s roughly half of META’s total announced AI compute capacity through 2027. Meta has said it’s planning to deploy ~12GW total. So this single commitment puts AMD at ~50% share of Meta’s AI compute over the deployment period — a massive share-shift away from NVDA + away from Meta’s own MTIA silicon.

The first 1GW is custom MI450-based. That’s not off-the-shelf MI300X / MI350 — it’s a custom chip designed to Meta’s specifications. Custom silicon engagements are very sticky: switching costs are enormous once the customer has integrated their software stack to a specific architecture.

The read-through to AVGO is the second-order story. Issue 03 flagged AVGO as the off-watchlist beneficiary of GOOG’s TPU capex doubling (Q1 2026 +108% YoY). The same logic ran through META’s MTIA program — Meta worked with Broadcom on MTIA design. Meta committing 6GW to AMD is a partial substitute for what would have otherwise been MTIA volume. The MTIA-to-AVGO revenue line gets less than it would have, because some of that capex is now flowing to AMD instead.

This isn’t a thesis-killer for AVGO — GOOG’s TPU is the bigger story for AVGO and that one’s intact. But it does soften the META leg of the AVGO custom-silicon thesis.

Arista takes +35% with 180bps of margin compression

ANET’s Q1 numbers are rich and direction-confirming. (ANET 10-Q, 6 May 2026):

• Total revenue $2,709.0M vs $2,004.8M prior — +35.1%
• Product revenue $2,311.3M (+36.6%) + Service revenue $397.7M (+27.3%)
• Gross margin 61.9%, down from 63.7% — i.e., -180bps
• Cash from operations $1.7B vs $641.7M prior — +165%
• $8.9B non-cancellable purchase commitments ($7.6B within 12 months)
• Two end customers = 42% of FY25 revenue (26% + 16% — almost certainly MSFT and META)
• Non-Americas revenue dropped to 15.5% from 20.3% — concentration in US hyperscalers

The margin compression is the tell. ANET is explicit about the cause: “The decrease was primarily driven by an increased proportion of our sales to large end customers who generally receive higher discounts.” (ANET 10-Q, 6 May 2026, MD&A)

Translation: Arista is paying with margin to keep MSFT and META locked in. They’re taking +35% revenue growth at 180bps lower margin because the AI-network buildout volume is more important than per-unit margin. The $8.9B in purchase commitments (up materially YoY) shows them committing to continued buildout — they’re betting the volume keeps coming, even on tightening margins.

This is the same pattern AMD’s warrants tell from a different angle: the chip and networking suppliers are willing to pay (in margin or in dilution) for share of the AI capex spend. That’s a structurally bullish read on the demand: when suppliers compete this hard, demand is being rationed, not over-supplied.

Lab-calibrated: ANET signal qualifies as future_bet × bullish, same setup as AMD. Same +2.5% mean alpha calibration. The customer-concentration risk is real but offset by the demand strength.

The Issue 03 capex thesis holds — chip tier confirms

Issue 03 lead-thread thesis: “Combined hyperscaler Q1 capex annualises above $400B; the ‘is capex slowing?’ hypothesis is dead.” The chip-tier prints this week confirm:

Indicator	Issue 03 prediction	Issue 04 chip-tier confirmation
Capex flowing to AI compute	$124B+ in Q1 from 4 hyperscalers	AMD DC revenue +38% YoY; ANET commits $8.9B in purchase orders
Demand ramping into Q2-Q3	Backlog evidence in VRT, EQIX	ANET deferred + performance obligations $7.7B, 91% recognised over next 2 years
NVDA capturing $140-192B annualised	Industry-estimate based	AMD warrants suggest NVDA share gets compressed — exact magnitude unknown
AVGO as off-watchlist beneficiary	Flagged Issue 03	Meta 6GW to AMD softens AVGO’s META leg; GOOG/TPU leg stands

The most important refinement: NVDA’s effective share is now contested, but in a structured way. OpenAI and Meta both have explicit incentives to buy AMD. Microsoft has its own contested OpenAI exclusivity (Issue 03). Google has TPU. Amazon has Trainium. The “merchant GPU is NVDA’s monopoly” narrative is now breaking from all four directions simultaneously.

NVDA still wins more than half of the $400B+ FY26 capex stream — they’ve still got the best-in-class hardware and the most mature software stack. But the CEILING on their share of the AI compute spend is now visibly lower than the market priced six months ago.

13F drop Friday — the convergence-detection moment

Q1 2026 13F filings are due Friday 15 May. Nine funds we track all filed Q4 2025 on the deadline (17 February) — i.e., they’re deadline-day filers. Expect a cluster of late-Friday drops:

• Atreides Management (Gavin Baker) — CIK 0001777813
• Altimeter Capital (Brad Gerstner)
• Coatue Management (Philippe Laffont)
• D1 Capital Partners (Dan Sundheim)
• Dragoneer Investment Group (Marc Stad)
• Duquesne Family Office (Stanley Druckenmiller)
• Durable Capital Partners (Henry Ellenbogen)
• Lone Pine Capital (Steve Mandel)
• Whale Rock Capital (Alex Sacerdote)

The convergence-detection logic will fire if ≥3 of these funds buy the same off-watchlist ticker. Names worth watching given the Issue 03 + Issue 04 context:

• AVGO (would confirm the off-watchlist primary-beneficiary thesis)
• BE (Bloom Energy — already #1 in Aschenbrenner’s portfolio per the Chamath thesis discussion)
• CRWV (CoreWeave — already in our watchlist; would confirm continued institutional inflow)
• CORZ / IREN / APLD (the crypto-pivot neoclouds Aschenbrenner concentrated in)
• EQT / VST / TLN / CEG (power-side beneficiaries we don’t currently track — if 3+ of these funds buy them, that’s an explicit “follow Aschenbrenner” signal)

The system is set up to send a watchlist-promotion alert when convergence fires — that alert next Friday is the highest-information event of the next two weeks.

Watchlist Updates

Ticker	Direction	Driver	Lab-calibrated reading
AMD	bullish	OpenAI + Meta warrants for 320M shares at $0.01; Q1 +38% revenue	Strong — bullish future_bet +2.5% mean alpha (n=365). Highest-conviction setup of the week.
ANET	bullish	Q1 +35% revenue, $8.9B purchase commitments, customer concentration with MSFT/META	Solid — same calibration, but watch the margin trajectory next 1-2 quarters
NVDA	neutral-to-bearish on margin	Two largest AI customers (OpenAI, Meta) now structurally aligned with AMD	Modest — bearish future_bet has -1.3% mean alpha; don’t oversell. Share dominance still intact for now.
META	bullish	6GW AMD GPU deployment confirmed (first 1GW custom MI450)	Reinforces the Issue 03 META thesis
AVGO	mixed (still off-watchlist; backfill in progress)	GOOG TPU thesis intact; META leg slightly softened by AMD share-grab	Tracking — 13F convergence on Friday is the trigger
MSFT	neutral	Indirect read: customer-concentration data point in ANET’s filing supports MSFT continuing as a top-2 ANET buyer	No direct change

The Week Ahead

• Tuesday 12 May: No major earnings on the watchlist
• Wednesday 13 May: TSM Q1 sales report (monthly revenue update, not full earnings) — read on whether AI accelerator volumes are ramping at TSMC
• Thursday 14 May: ARM Q4/FY26 results (foreign-private-issuer 6-K — earnings call commentary is the main read for the IP-royalty trajectory)
• Friday 15 May: 13F deadline. Nine tracked funds expected to file. Convergence-detection runs at end of day.
• Sat-Sun 16-17 May: Issue 05 production cycle.

What would change next week’s thesis: a NVDA pre-announcement that contradicts the AMD share-shift implication (low probability, but historically NVDA has used pre-announcements when narratives turn against them). More likely: silence from NVDA on the OpenAI / Meta side until their own earnings (28 May, after Issue 06 ships).

Methodology and disclosures

Filings Intel Digest is a bona fide financial publication. Nothing here is personalised investment advice. The editor may hold positions in companies discussed; current positions and policy are at /about. All claims are sourced to publicly filed documents linked inline. Backtested or historical figures are direction-adjusted and calibrated against the signal lab; calibration is partial — treat any forward statement as a hypothesis, not a forecast.

Frozen as of close-of-trading 8 May 2026 (Friday). The 13F deadline (15 May) is the next major information event.

Issue 03’s bearish read on NVDA was modest — it’s now slightly more bearish given the AMD warrants disclosure, but lab calibration on bearish future_bet setups (-1.3% mean alpha, 51.7% hit rate, n=180) means the share-price translation is poor even when the directional read is correct. Treat as hypothesis, not call.

The “AMD is willing to dilute 20% to lock OpenAI and Meta” framing is the editorial reading of the warrant disclosure; the underlying disclosed facts are the share counts (160M each), strike price ($0.01), and milestone structure. The strategic interpretation is editorial.