Playbook › Playbook
Equity issued to customers is contra-revenue, and adding it back manufactures both growth and margin
Claim
The accounting is not obscure and it is not a judgment call. Equity granted to a customer is consideration paid to the customer, so ASC 606 nets it against the transaction price. It lands on the balance sheet as a customer warrant asset and unwinds through the revenue line, in proportion to the related revenue recognised.
The trap is that management reports around it, and the add-back flatters three metrics at once.
Evidence — CBRS, Q2 2026
Cerebras issued warrants to OpenAI, AWS and G42 and capitalised $1.128B of customer warrant assets, amortising through October 2031. In six months it issued $1.022B of equity to its own customers — $822.9M OpenAI, $149.4M G42, $50.0M AWS — against H1 revenue of $373.5M. 2.7x its half-year revenue, paid in stock, to secure commitments.
| Q2 2026 | GAAP | "Core" (company non-GAAP) |
|---|---|---|
| Revenue | $180.1M (+74%) | $209.9M (+103%) |
| Gross margin | 14.2% | 41% |
| Operating margin | −265% | −16% |
The sign inversion. GAAP hardware revenue fell 23% year over year, and the whole financial press reported a chip company whose chip sales were collapsing. The 10-Q says the $16.2M decline "was primarily attributable to $28.0 million in amortization of customer warrant assets." Ex-warrant, hardware revenue grew ~17%. The add-back was directionally correct here — but only visible to someone who read the MD&A. The metric is wrong in both directions and neither error is detectable from the headline.
The double charge. The warrants suppress reported revenue and dilute the share count. At CBRS the OpenAI warrant alone covers up to 33,445,026 shares at $0.00001 — ~14% of the company. The shareholder funds the customer acquisition twice: once through the revenue line, once through ownership.
How to run the check
- Search the filing for
customer warrant,contra-revenue, orreduction of revenue. The balance-sheet line is usually a named current + non-current asset pair. - Take the remaining unamortised balance — that is committed future contra-revenue, and it scales with revenue, so it does not decay away as the company grows.
- Read the per-segment MD&A attribution before believing any segment growth rate.
- Add the warrant shares to the diluted count independently. The revenue effect and the dilution effect are disclosed in different notes and are rarely presented together.
- Value on GAAP revenue. Then ask separately whether backlog bought with equity renews at a price, or only renews with another grant.
Why this will recur
Warrant-for-commitment structures spread across AI infrastructure through 2025–2026 as suppliers competed for anchor tenants. Wherever a vendor needs a marquee customer more than the customer needs the vendor, this structure appears. Related: [[pattern-non-gaap-definition-change-leaves-a-restatement-fingerprint]], [[pitfall-contra-revenue-reclass-inflates-revenue-growth]], and [[pitfall-rpo-inflated-by-pass-through-and-termination-rights]] — the same CBRS filing produced both.
The one-line test: a company that pays its customers in equity to take the product does not have pricing power, whatever the growth rate says. See [[principle-story-vs-revenue]].
✅ Amendment 2026-08-27 — AMD is the megacap instance, and it is the control case
Found the same day as the original note, on a $778B company rather than a $12B one. The structure is identical; the accounting stage is one step earlier, which is what makes it valuable.
| CBRS | AMD | |
|---|---|---|
| Warrants | to OpenAI, G42, AWS | 160M to OpenAI (Oct-2025) + 160M to Meta (Feb-2026) |
| Strike | nominal | $0.01 |
| Size vs company | ~14% held by OpenAI | 320M shares = 16.2% of the post-issue company (~$152B at spot) |
| Vesting | purchase milestones | purchase milestones AND specified stock-price targets |
| GAAP classification | customer warrant asset, amortised to contra-revenue | liability, pending equity-classification conditions |
| Recognised in the P&L | yes — $1.022B amortising to Oct 2031 | ZERO. No tranche has vested. |
AMD's Q2 2026 10-Q, Note 12, verbatim: "The warrants will be classified as liabilities until certain conditions for equity classification are met… As of June 27, 2026, no warrant shares had vested or become exercisable, and the warrants had no impact on the Condensed Consolidated Financial Statements." AMD additionally discloses $4.1B of maximum exposure on guarantees of commercial-partner data-centre lease obligations, terms to 15 years, "may be issued in exchange for warrants," booked as a credit derivative.
Three refinements to the original claim
1. The pattern has a pre-recognition phase, and that is when it is invisible. The original note
teaches you to find the contra-revenue in the filing. AMD proves there is an earlier stage where
the correct search returns nothing — no customer warrant asset, no contra-revenue line, no
segment attribution, because nothing has vested. Searching the filing and finding nothing is
therefore not evidence of absence. The new first step is: search the equity, commitments and
guarantees notes for warrants issued to named customers, whatever their current classification.
2. Liability classification is not a milder version of the same thing — it is a different recognition path with the same economics. Where the cost eventually lands (contra-revenue under ASC 606, versus expense, versus equity reclassification) is not disclosed in AMD's filing, and it determines whether the hit shows up in revenue growth and gross margin or merely in operating expense. Under ASC 606 a milestone-vested warrant issued to a customer is the textbook case for a reduction of revenue. Record the open question with a date and re-check at the vesting quarter — for AMD, plausibly Q4-2026 or Q1-2027, since Meta's first-gigawatt shipments begin 2H26.
3. 🔑 The new mechanism: price-target vesting makes the dilution reflexive. CBRS's warrants vest on purchases alone. AMD's vest on purchases and on share-price thresholds, reportedly rising to ~$600 for the final Meta tranche. That inverts the usual reading of dilution as a downside risk:
The dilution triggers on the thesis working. The 320M shares land precisely when the multiple is being justified, so a holder cannot own the upside without funding the giveaway — and the charge grows with the very success that would otherwise support the valuation.
Any DCF or multiple built on the bull case must use the fully-diluted count, because the bull case is the vesting condition. On AMD that moves FY2027 non-GAAP from 30.9x to 36.8x and FY2026 from 63x to 75x. Where the two are presented separately, they are not two scenarios — they are one scenario, half-counted.
Guard, added: when warrants to customers carry share-price vesting conditions, never quote a target price off the undiluted count. State the fully-diluted multiple as the primary figure.
History
- 2026-08-27 — established from CBRS, where the cost is already amortising through revenue.
- 2026-08-27 — amended the same day from AMD, the megacap control case: same structure, larger in absolute and relative size, nothing yet recognised, and a reflexive price-target vesting condition the CBRS instance does not have. The pair brackets the pattern's lifecycle — CBRS shows what it looks like after recognition, AMD shows what it looks like before.