The Meta-AMD deal disclosed Friday was supposed to be a contract. By Sunday morning the only thing the AI sell-side agreed on was that nobody knew how to price the warrant attached to it. JPMorgan's Harlan Sur put the warrant's intrinsic equity value in a range of $30 billion to $100 billion. [1] Goldman's Toshiya Hari put the gross-margin drag at 200 to 300 basis points; Bank of America's Vivek Arya put it at 150. The spread on the same filing, on Day 4, is wider than the spread on most acquisitions.

The structure that caused it is, on paper, simple. AMD granted Meta a warrant for up to 160 million AMD shares — approximately 10 percent of the share count — at a $0.01 strike. The warrant vests in tranches conditioned on Meta's cumulative purchases hitting a 6-gigawatt threshold of AMD AI accelerators, AND on AMD's stock price reaching $600. [2] The intermediate vesting tranches are described in the 8-K but not enumerated; the filing reserves the schedule for an exhibit yet to be published. [3] The paper's Friday read had the JPM-GS-BofA spread tightening at the cohort level; Day 4 shows the spread on the underlying valuation has not.

That single missing exhibit is the reason JPMorgan's range is what it is. If Meta hits the 6-gigawatt threshold and AMD shares clear $600, the warrant prints intrinsic value of roughly $96 billion at the threshold price, on the order of $100 billion at any meaningful overshoot. If Meta only hits half the threshold and the share price plateaus at $400, the warrant prints closer to $30 billion of value at issuance, less the unvested tranches. The sell-side cannot draw the chart without the schedule.

The bull argument runs that the 6-gigawatt purchase commitment is a real contract, the warrant is non-cash and vests only if Meta delivers, and the dilution is more than offset by the multiple expansion the contract justifies. JPMorgan made the case in two paragraphs and conceded in the third that the math depends on a tranching schedule the bank does not have. [1] The bear argument is structural. Meta was going to sign a chip-supply contract with AMD or Nvidia regardless; the warrant gave AMD the contract by transferring 10 percent of the company to the customer.

The cohort is now two: AMD-OpenAI and AMD-Meta. Both warrants are penny-strike, conditioned on volume thresholds and share-price thresholds. Together they encumber more than 18 percent of AMD's fully-diluted share count at maximum vesting. The AMD investor-relations team has not, in either disclosure, published a combined-encumbrance figure. The math has to be done by the reader.

Goldman's narrower 200-to-300 bps drag assumes the Meta volumes carry a margin profile within 100 bps of the rest of the AI accelerator book. JPMorgan's wider range prices in customer-concentration risk and tranche compression. BofA's narrowest reading discounts the gross-margin drag by tranche probability — the warrant is so unlikely to vest, on its model, that the drag is partly imaginary. Three banks, one filing, three probability sets.

What the sell-side wants this week is the tranche-vesting schedule. The AMD CFO is at Bernstein Strategic Decisions on May 28 — too late to settle the question before Cerebras prices Wednesday and SpaceX's S-1 lands the following Monday. Meta's 10-Q is due May 23 and will disclose the 6-gigawatt commitment in the related-party section; it will not disclose Meta's internal probability model on $600.

What is true on Day 4 is that AMD's stock has rallied 18 percent since the deal print, the option market has steepened through $600 on 2026 strikes, and the $600 threshold is no longer aspirational. It is the strike at which approximately $80 billion of AMD-Meta equity value enters the float.

-- THEO KAPLAN, San Francisco