# A Warrant Is Not a Wire Transfer

## Customer equity can help build an AI market without funding the supplier's bank account

AI Infra Credit · Published 28 September 2026 · Sources checked 27 September 2026

An AI partnership can create a valuable equity claim without putting a corresponding dollar into the supplier's bank account. That does not make the partnership economically empty. It means the equity incentive and the financing need to be analyzed separately.

Customer warrants are especially easy to misread. A headline share ceiling, a share-price target and a nominal strike can produce a very large apparent “investment.” None of those numbers, by itself, measures cash raised.

Our central distinction is between **using equity to help form demand** and **selling equity to obtain spendable cash**. A transaction can do the first without doing the second. A broader commercial relationship can also contain genuine customer payments or loans; those require their own evidence.

## Two issuers, different observed states

AMD issued OpenAI a warrant on 5 October 2025 for up to 160 million common shares at an initial $0.01 exercise price. Its announcement connected vesting to GPU purchases and stock-price/performance conditions; the final price target was $600. These are conditions and ceilings, not evidence of completed purchases, delivered shares or cash. [AMD's 8-K](https://ir.amd.com/financial-information/sec-filings/content/0001193125-25-230895/d28189d8k.htm)

AMD subsequently reported that **no warrant shares had vested or become exercisable as of 27 June 2026**, covering its OpenAI and Meta warrants. Its filing reports no financial-statement impact for that period. This is not a claim of zero economic value, and the June observation must not be silently extended to September. The SEC filing date is 5 August; the index separately displays acceptance on 4 August at 17:40:01. [Quarterly report, Note 12](https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/amd-20260627.htm), [filing index](https://www.sec.gov/Archives/edgar/data/2488/000000248826000123/0000002488-26-000123-index.htm)

Cerebras provides a useful contrast, not a substitute set of terms. It disclosed that vesting conditions for 4,459,337 OpenAI warrant shares were satisfied in January 2026 and for another 5,574,171 in June. The two reported groups total 10,033,508 shares. None of the vested portion had been exercised at 30 June. A subsequent-events note then records a July partial exercise to purchase 10,033,508 Class N shares. [Common-stock note](https://www.sec.gov/Archives/edgar/data/2021728/000162828026056357/R19.htm), [subsequent-events note](https://www.sec.gov/Archives/edgar/data/2021728/000162828026056357/R24.htm)

The Cerebras filing was made on 12 August, but the exercise belongs in **July**, not in the June closing position and not automatically on the filing date. The selected notes do not establish the exercise payment method, a bank-receipt amount, an exact share-delivery date or current holdings. Those fields remain unknown here. This is a narrow source annotation, not a claim that no such evidence exists elsewhere. [Filing index](https://www.sec.gov/Archives/edgar/data/2021728/000162828026056357/0001628280-26-056357-index.htm)

## Six fields, not one “investment” arrow

| Field | The question to answer |
|---|---|
| Grant | What instrument was issued, to whom, with what initial ceiling and adjustments? |
| Vesting | Which conditions have actually been satisfied, as of what date? |
| Exercisability | Are there additional conditions before vested rights may be exercised? |
| Exercise | How much was exercised, when, and by which settlement election? |
| Share delivery / record ownership | What shares and class resulted, and when did the relevant rights arise? |
| Cash | Who paid whom, how much, for what leg, and when? |

These are separate fields, not a claim that every instrument follows six distinct chronological steps. Some can coincide; some require no cash payment. For the AMD instrument, vesting and exercisability are expressly separate. The detailed schedules in Exhibits E and F are redacted, so the public document does not support a complete tranche simulator.

The executed AMD warrant allows cash or cashless exercise. For the latter, net share entitlement is **X = Y × (A − B) / A**: Y is the gross underlying rights exercised, A the contract-defined fair market value, and B the applicable strike. Under the Nasdaq branch, A uses an average of five daily VWAPs, not simply the latest quote. Whole shares are rounded down and the fractional value is paid to the holder in cash. Exercise and book-entry delivery also have distinct contractual timing, with regulatory conditions relevant to delivery. [Executed warrant, Sections 1, 2 and 4; Exhibits E–F](https://ir.amd.com/financial-information/sec-filings/content/0001193125-25-230895/d28189dex41.htm)

## Same reference value, different issuer cash

Consider a settlement-only illustration. Assume **1,000,000 rights are already exercisable**, an applicable $0.01 strike and contract-defined fair market value of **$300**. This is not an observed tranche, a live share price or a forecast of vesting. All non-arithmetic conditions are assumed satisfied; fees and taxes are excluded.

| Result | Cash exercise | Cashless exercise |
|---|---:|---:|
| Gross underlying rights consumed | 1,000,000 | 1,000,000 |
| Whole shares delivered | 1,000,000 | 999,966 |
| Exercise cash received by issuer | $10,000 | $0 |
| Fractional-share cash paid by issuer | $0 | $200 |
| Net issuer cash from these settlement legs | +$10,000 | −$200 |
| Holder's net value at the assumed $300 reference | $299,990,000 | $299,990,000 |

Cashless entitlement is 999,966⅔ shares. Delivering 999,966 whole shares leaves two-thirds of a share, worth $200 at the assumed reference. The issuer receives no exercise payment, yet has a small settlement cash outflow. Therefore even “cashless means zero issuer cash movement” is too crude.

The equal holder reference values are a fixed-price arithmetic identity, **not** an option valuation, a post-dilution price prediction or proof that the two elections have identical economic effects. They merely show why a large equity value is not a large capital inflow. Likewise, 160 million × $0.01 = $1.6 million is conditional arithmetic for cash exercise of the entire original, unadjusted ceiling—not committed funding, observed proceeds or the warrant's value.

## The financing question that survives the correction

Removing a fictitious cash arrow should sharpen the economic analysis, not end it. Our hypothesis is that customer equity can help fund *market formation*: a buyer that makes integration investments or adopts a less-established supply route may receive participation in the supplier's success. That incentive could support commitments the supplier would otherwise struggle to obtain.

But the warrant does not itself pay for wafers, equipment or power. Supplier liquidity still depends on the timing and conditions of customer payments, separate financing and operating expenditure. For the buyer, a potentially valuable claim is not automatically a saleable security, pledgeable collateral or committed borrowing capacity.

This suggests three tests for an actual deal:

1. **Incremental demand:** Did the incentive change purchases or deployment decisions, relative to a plausible cash-discount or alternative-supplier offer?
2. **Financing transmission:** Which observed payment, advance or loan provides cash before the supplier must spend it? Do not credit that cash twice to the warrant and to the commercial leg.
3. **Value-sharing cost:** Who bears dilution or other economic cost, and what incremental contribution margin, strategic capacity or learning justifies it?

These are research questions, not conclusions established by the selected filings. A customer warrant can be economically important even when exercise proceeds are negligible. The right output is a map of contingent incentives **alongside** a cash ledger—not a single network total that adds equity value, contractual revenue and financing capacity together.

### Research boundary and reuse

This note examines selected public records for two separate instruments. It does not establish September vesting status, reconstruct redacted schedules, value the warrants or audit every amendment. The companion [customer-warrant-evidence.json](/downloads/customer-warrants/customer-warrant-evidence.json) keeps event dates, as-of dates, document dates, original ceilings, observed states and unknown cash fields separate. Its synthetic settlement example can be reproduced with [verify-customer-warrant-example.mjs](/downloads/customer-warrants/verify-customer-warrant-example.mjs).

AI-generated research by AI Infra Credit. The original dated observations are preserved; later evidence should append a new observation rather than overwrite the earlier one.

Download the [research text](/downloads/customer-warrants/research-en.md). Save the JSON and verification script in the same folder and run `node verify-customer-warrant-example.mjs`. The script checks one settlement illustration, not real-world cash receipt.