OpenAI Exercises Cerebras Warrants for $100, Acquiring 4.2% Stake With ~$2.3B Implied Value Days Before Ultrafast Launch
Summary
- • OpenAI exercised all vested Cerebras warrants in July, paying ~$100 cash for 10,033,508 Class N shares — a 4.22% stake with an implied value of ~$2.3 billion at Cerebras's Aug. 13 Class A price of ~$229.
- • The equity stems from a December 2025 master relationship agreement under which OpenAI committed to buy 750 megawatts of Cerebras inference capacity through 2028; a ~$1B working-capital loan in January 2026 triggered the first warrant tranche.
- • Two days after the equity filing, OpenAI previewed Ultrafast — a new service tier running GPT-5.6 Sol on Cerebras wafer-scale hardware at up to 750 tokens/second, up to 14× its Standard tier.
- • Every competitive benchmark for Ultrafast was run or characterized by Cerebras, and OpenAI published no price, model ID, or general-availability date for the service.
Details
Warrant Exercise: ~$100 for 10M+ Shares
OpenAI exercised every vested Cerebras warrant in July 2026, acquiring 10,033,508 Class N shares at $0.00001 each for approximately $100 total in cash.
Implied Value ~$2.3B; No Voting Rights
At Cerebras's Class A price of ~$229 on Aug. 13, the 4.22% stake (of 237,564,041 shares outstanding as of Aug. 5) carried an implied value of roughly $2.3 billion. Class N shares carry no votes and only convert to Class A when transferred — they are not liquid.
Full Warrant: 33.4M Shares; 23.4M Remaining
The full warrant covered 33,445,026 Class N shares when granted. OpenAI exercised the 10,033,508 vested shares; it retains warrants for another 23,411,518 shares tied to capacity, payment, or market-value milestones.
Cerebras Recorded $822.9M in Customer-Warrant Assets H1 2026
Cerebras valued the warrant at $82.02 per share at grant and recorded $822.9 million in customer-warrant assets during the first half of 2026 — reflecting the commercial value of the OpenAI relationship on its balance sheet.
December 2025 Master Agreement: 750 MW Through 2028, Option on 1.25 GW by 2030
The warrant was granted under a December 2025 master relationship agreement in which OpenAI committed to purchase 750 megawatts of Cerebras inference capacity in tranches through 2028, with an option on another 1.25 gigawatts by end of 2030.
~$1B Working-Capital Loan Triggered First Tranche
A secured working-capital loan of approximately $1 billion extended from OpenAI to Cerebras in January 2026 triggered the first warrant tranche — meaning OpenAI is also a significant creditor of the company it now partially owns.
Ultrafast: GPT-5.6 Sol at Up to 750 Tokens/Sec
OpenAI previewed Ultrafast on Aug. 13, two days after the equity filing. It is a service tier — not a new model — running the same GPT-5.6 Sol on Cerebras wafer-scale systems at a maximum of 750 output tokens per second, up to 14× the Standard processing tier.
Cerebras Wafer-Scale: 44 GB On-Chip SRAM, No Off-Chip Fetches
Each Cerebras wafer-sized chip holds 44 GB of SRAM, keeping model weights on-chip rather than shuttling them to off-chip storage between tokens — the architectural basis for the inference speed advantage.
Cerebras: Ultrafast 7× Faster Than Claude Fable 5 on HLE
In Cerebras-run tests, Ultrafast completed the 2,500-question Humanity's Last Exam in 11 hours 11 minutes (July 10), while Claude Fable 5 took 78 hours 27 minutes (July 13–15). Cerebras claimed nearly 7× faster with comparable accuracy but did not publish accuracy scores.
All Benchmarks Run by Cerebras; No Independent Verification
Every competitive speed comparison in the Ultrafast launch was run or characterized by Cerebras. The advertised ceiling of 'up to 14× Standard speed' disclosed no baseline workload or prompt details. OpenAI has published no Ultrafast price, model ID, or general-availability date.
OpenAI Is Customer, Creditor, and Now Equity Holder
The commercial structure is unusually entangled: OpenAI is simultaneously Cerebras's largest compute customer (750 MW committed), its working-capital lender (~$1B loan), and now a 4.22% equity holder — all governed by the December 2025 master agreement.
Source: SEC/equity filing (disclosed Aug. 12, 2026 after market close); Ultrafast preview (Aug. 13, 2026); Cerebras H1 2026 financials. Reported by Implicator.ai / TLDR AI.
What This Means
OpenAI's $100 warrant exercise for a ~$2.3B implied-value stake in Cerebras reveals one of the most structurally complex commercial relationships in AI: it is simultaneously customer (750 MW committed through 2028), creditor (~$1B working-capital loan), and now equity holder in the same chip company. The timing — equity filing on Aug. 12, Ultrafast preview on Aug. 13 — makes the vertical integration explicit: OpenAI is monetizing inference speed on hardware in which it holds a financial stake, with all benchmarks supplied by the hardware vendor. For the broader AI infrastructure market, this signals a new model of compute securing — preferred access plus equity upside in exchange for committed spend and a working-capital lifeline — that could pressure rivals to build similar bilateral arrangements with chip suppliers. The ~23.4M Cerebras warrants OpenAI still holds (tied to future milestones) mean the equity relationship is likely to deepen further as the compute deal matures.
Sentiment
Intrigued by the complex vertical integration and low-cost stake acquisition
“OpenAI quietly exercised warrants for 10,033,508 Cerebras shares in July, giving it a 4.2% economic stake now worth roughly $2.3 billion. Today, OpenAI launched GPT‑5.6 Ultrafast on Cerebras hardware. We found the exercise buried in yesterday’s 10‑Q.”
“OpenAI exercised warrants for a 4.2% Cerebras stake in July for ~$100, weeks before launching Ultrafast—whose speed claims all came from Cerebras.”
“OpenAI exercised warrants for a 4.2% stake in Cerebras. Total cash paid: about $100. Implied value: ~$2.3 billion. Enterprise focused though :(”
“Okay, this is a real commercialization signal for $CBRS—not just another speed demo. Cerebras previously disclosed a 750MW, >$20B OpenAI agreement; now OpenAI is wrapping that infrastructure into a customer-facing tier at up to 750 tokens/sec. But I wouldn’t confuse launch with revenue proof.”
Split
Limited discussion; ~80/20 intrigued vs. cautious on revenue proof vs. launch hype.