AMD's OpenAI Warrant: 160M Shares, a $600 Target and What Must Be Hit

AMD has promised OpenAI up to 160 million of its own shares for $0.01 each. The AMD OpenAI warrant pays out only if OpenAI buys 6 gigawatts of AMD GPUs and AMD's stock climbs to $600. At that price the shares are worth $96 billion. AMD then wrote the same warrant for Meta, and none of either had vested by 27 June 2026.

The short answer. The AMD OpenAI warrant lets OpenAI buy up to 160 million AMD shares at $0.01. Shares vest in tranches as OpenAI buys AMD GPUs, starting at 1 gigawatt of MI450 chips and ending at 6 gigawatts, and only if AMD's stock hits targets rising to $600. It expires on 5 October 2030.

Reader for this post: founders and finance leads who price vendor deals. The warrant is the clearest public example of a supplier paying a customer in its own equity, and you will meet smaller versions of it in your own contracts.

What AMD's 8-K says the OpenAI warrant actually is

A warrant is a right to buy shares at a fixed price. This one has an exercise price of $0.01, so in practice it is a promise of free stock. The economics sit in the conditions.

AMD disclosed the deal in a Form 8-K dated 5 October 2025. The holder is OpenAI OpCo, LLC. The warrant was issued "in connection with and concurrent with" a product purchase agreement covering AMD Instinct GPUs. OpenAI made a binding commitment, at signing, to buy the initial 1 gigawatt of MI450 Series products.

That is the only binding volume in the filing. The other 5 gigawatts are what OpenAI must buy for the remaining tranches to vest. They are an incentive, not an order.

The two conditions that must both be met

The first condition is purchases. The 8-K says the first tranche vests "after the delivery of the initial one (1) gigawatt" of MI450 GPUs, and full vesting is "contingent upon" purchases of 6 gigawatts. Purchases count whether OpenAI buys directly, through affiliates, or through authorised third parties.

The second condition is AMD's share price. Vesting is "further subject to achievement of specified Company stock price targets that escalate to $600 per share for the final tranche and stock performance thresholds". Even then, each vested tranche must meet "certain other technical and commercial conditions prior to exercise".

So the warrant has three gates, not two. Buy the chips, see the stock rise, then satisfy conditions AMD has not described in public. I read that third gate as AMD's protection against paying for gigawatts that are delivered but never switched on.

What the public filings leave out

The warrant itself, filed as Exhibit 4.1, places the vesting schedule in an exhibit labelled E and the exercise conditions in an exhibit labelled F. The public text does not give per-tranche share counts or the intermediate price targets. Only the first gigawatt, the sixth gigawatt and the $600 endpoint are public.

The exhibit does state that unvested shares are cancelled automatically if AMD terminates the supply agreement for cause, or at expiry. It also defines VWAP, the volume weighted average price, using Bloomberg data. That is a sign the price targets are measured over trading periods rather than on a single closing print, though the filing text I read does not tie VWAP to the targets directly.

The Meta warrant doubles the AMD share count at stake

Most coverage of the AMD OpenAI warrant treats it as one deal. It is now one of two. On 23 February 2026, AMD issued a warrant to Meta Platforms with the same headline terms, disclosed in a Form 8-K dated 23 February 2026.

Meta can buy up to 160 million AMD shares at $0.01. The first tranche vests on the initial gigawatt of Instinct GPUs, and full vesting requires 6 gigawatts of purchases. Price targets again escalate to $600 for the final tranche. Meta's warrant runs to 23 February 2031, about 5 months after OpenAI's expires.

AMD's own 10-Q now reports the two together. Note 12 of the June 2026 quarterly filing says AMD "issued warrants to OpenAI OpCo, LLC (OpenAI) and Meta Platforms, Inc. (Meta), each entitling the holder the right to purchase up to 160 million shares". It adds that "the vesting of the OpenAI Warrant is also subject to stock-performance thresholds," a condition it does not attach to Meta's.

A shareholder who priced in 160 million shares in October 2025 is now holding a company that has committed up to 320 million. I think analysts who still quote "roughly 10%" are reading last year's filing.

AMD OpenAI warrant dilution: the share-count math

AMD's Q2 2026 10-Q cover page reports 1,632,475,042 shares outstanding as of 29 July 2026. That is the base for every percentage below. It is my own calculation, and it assumes full vesting and full exercise.

The 160 million OpenAI shares equal 9.8% of today's count. Once issued, they would be 8.9% of the larger total, because the new shares also enlarge the denominator. The second number is the one that describes OpenAI's actual ownership.

Add Meta and the figures double. 320 million shares equal 19.6% of the current count and 16.4% of the enlarged base of roughly 1.95 billion shares. Every existing holder's claim on AMD's future profits would shrink by about one sixth.

Author's calculation from AMD's 10-Q cover share count. The first row is left blank on purpose. Tranche sizes sit in an exhibit whose schedule is not public, and averaging 160 million across 6 gigawatts would invent a number the filings do not support.

Buybacks are the obvious offset, and they cost cash

AMD can repurchase shares to absorb some of this. That does not make the dilution free. It converts a share-count cost into a cash cost, paid at whatever the market price is when the warrant shares arrive. If the warrant has vested, by definition that price is high.

The share count already drifts. It rose from 1,630,410,843 in January 2026 to 1,632,475,042 in July 2026, per AMD's 10-K and 10-Q covers. That is ordinary employee equity, so the warrant sits on a count that was already rising.

What 160 million AMD shares are worth at each price

Because the exercise price is a cent, the value of the warrant shares is close to the share price times 160 million. The hard part is choosing which price matters.

Four prices tell the story. AMD closed near $165 on 3 October 2025, the last trading day before the deal. Press reports put the close at about $197 on 23 February 2026, the day of the Meta deal. The stock closed at $504.80 on 30 October 2026. The final tranche needs $600.

The $600 target is a $1 trillion company

At 1.63 billion shares, $600 implies a market value of about $980 billion before the warrant shares are added. UBS analyst Timothy Arcuri reached the same conclusion, telling clients the final tranche needs a market cap of roughly $1 trillion to vest, as TechCrunch reported on 7 October 2025.

The stock has already been there briefly. AMD traded near $649 intraday on 6 October 2026, after a run through $600 in late September. Nothing vested, because the gigawatt conditions were not met. The price gate is the easier one.

Contra-revenue: how the warrant reaches AMD's income statement

Contra-revenue is a deduction recorded against sales rather than as an expense. When a supplier pays a customer, in cash or in kind, accounting rules usually treat that payment as a discount on the goods sold. The supplier reports less revenue, not more costs.

AMD said this is how the warrant will work. Its 10-Q for the quarter ended 27 September 2025 states in its subsequent events note: "The grant date fair value of the warrant will be recognized as a reduction to revenue when revenue is recognized".

Grant date value hits revenue, later swings hit other income

The same note says AMD "expects to account for the warrant as a liability-classified financial instrument, unless the conditions for equity classification are satisfied". While it is a liability, it "will be remeasured at fair value at each reporting period, with changes in fair value to be recorded within Other income (expense), net".

That split is the key to reading future AMD results. The revenue reduction is fixed at the value on the grant date, when AMD's stock was near $165. Every dollar of value gained since then is a below-the-line charge, outside gross margin and outside the Data Center segment.

I think this is the single most under-discussed fact about the deal. If AMD's stock keeps rising, the warrant's growing cost shows up in other expense, which many readers skip and which non-GAAP figures commonly exclude. The revenue line carries only the October 2025 value.

The ceiling on the revenue reduction

AMD has not disclosed the grant date fair value for either warrant. No 10-Q through June 2026 gives a figure. It can be bounded, though. A fair value cannot exceed the shares' worth with every condition already met, which on the grant date was about $26.3 billion for OpenAI and $31.6 billion for Meta.

The real figure will be lower, because the price targets and gigawatt milestones were far from certain at grant. How much lower depends on an option model AMD has not published.

The ceilings are intrinsic values on the grant date, not AMD's fair value. The combined ceiling is 3.5 times fiscal 2025 Data Center revenue. It will be spread across the revenue from up to 12 gigawatts of shipments, not taken in a single year.

The deduction is recognised as the related revenue is recognised, so it is matched against chip sales over years. AMD told investors the OpenAI deal alone was expected to deliver "tens of billions of dollars in revenue", which is the right denominator.

Where the AMD OpenAI warrant becomes circular financing

Circular financing is any arrangement where a supplier funds, directly or indirectly, the purchases a customer makes from it. The common versions are loans, equity stakes and guarantees. The wider map of who pays whom in the AI circular deals covers the cash versions. AMD's version uses its own stock.

The mechanism works like this. OpenAI buys a gigawatt of GPUs. A tranche of AMD shares vests, worth more if AMD's stock has risen on the news of those purchases. OpenAI can sell those shares, since the 8-K says vested warrant shares are "freely tradeable, subject to securities laws and specified limitations". The proceeds can fund the next gigawatt.

Arcuri described exactly this in October 2025, saying OpenAI was more likely to sell shares along the way to fund its AMD purchases than to hold the whole stake. In that reading, AMD shareholders are partly paying for AMD's own revenue, through dilution.

It differs from the Nvidia structure in one way that I think favours AMD. Nvidia's arrangement with OpenAI involves cash invested in the customer, which is traced in the filings trace of Nvidia's OpenAI investment. AMD lends nothing and holds no OpenAI equity. If OpenAI stops buying, AMD has no loan to write off. Unvested shares are simply never issued.

The cost moves rather than disappears. With a loan, the vendor bears credit risk. With a warrant, the shareholder bears dilution risk, and only in the scenario where the stock has risen a long way. The history of vendor financing at Lucent and Nortel is a story about the first kind of risk. This deal is mostly the second kind.

Auditors look at these structures through a specific lens: whether the payment to the customer exceeds the fair value of anything distinct the supplier receives. The revenue tests auditors apply to circular financing set out why that test decides whether a payment is a discount or something else. AMD's filing has already answered it: a discount.

Where this argument is weakest

The dilution numbers above are maximums, and that framing can mislead in three ways.

Dilution only arrives alongside success

The final tranche needs AMD at $600, close to a $1 trillion valuation, and 6 gigawatts of shipped GPUs. In that world a holder who bought at $165 has gained roughly 3.6 times. Giving up 8.9% of a company that much larger is a cost most shareholders would accept. Dilution in a failure scenario is close to zero, because nothing vests.

That asymmetry is the strongest case for the structure. AMD issues equity only in the states of the world where it can most afford to. It is closer to a success fee than to a giveaway.

A price cut might have cost more, and the ceiling is not the number

The alternative to a warrant was a lower chip price. A cash discount hits gross margin on every unit from the first shipment, whether or not the stock rises. The warrant defers the cost and makes most of it conditional. For a challenger fighting Nvidia for its first hyperscale-sized customers, that may have been the cheaper route.

My contra-revenue ceilings are also loose. Option models applied to deep out-of-the-money price targets often produce fair values well below intrinsic value. If AMD's booked figure comes in at a fraction of $26.3 billion, the revenue impact per gigawatt is modest. I cannot rule that out, because AMD has not published the number.

Finally, the third gate is opaque in both directions. The undisclosed "technical and commercial conditions" could make exercise harder than the headline suggests. They could also be routine. Anyone claiming to know which is guessing.

What to read in AMD's Q3 2026 report

AMD reports its fiscal third quarter on 3 November 2026, after the market close. Its Q2 release guided to revenue of about $13.0 billion, plus or minus $0.3 billion, and said Helios rack deployments with customers including OpenAI begin to ramp in the second half. That is the quarter in which the first tranche could start to matter.

Three lines in the 10-Q will tell you more than the headline revenue. Whether Note 12 still says no warrant shares have vested. Whether any revenue reduction is reported, and its size, since that reveals the grant date fair value for the first time. Whether other income carries a remeasurement charge.

Watch the OpenAI side for financing strain too. The 6 gigawatts depend on OpenAI raising and spending at a scale set out in its compute commitments against its revenue. If later tranches lapse, the dilution never happens, and neither does the revenue.

Frequently asked questions

How to use this before your next vendor deal

If you hold AMD, open the Q3 10-Q when it lands and search for "Warrants". Write down three numbers: shares vested, any revenue reduction, and any remeasurement in other income. Compare them with the ceilings in this post, and you will know within ten minutes whether the cost is landing high or low.

If you run a company that is offered equity, credits or rebates by a supplier, ask the question AMD's filing answers. Is this a discount on what I buy, and where will each side book it? A vendor who cannot tell you how the incentive hits their own revenue has not finished pricing the deal. Neither have you.

Related reading