Cerebras Goes Public Tomorrow — Inside the $30B OpenAI Alliance

20× oversubscribed · IPO price $160 · $48B valuation — and Sam Altman's conflict is listed as an S-1 risk.

---

Cerebras (CBRS) lists on Nasdaq on May 14. The IPO price prints today (May 13 US time). At the top of the band — $160 — the market cap is about $48B (~₩68T), putting it on track to be the largest tech IPO of 2026.

Demand overwhelmed expectations. The price band lifted twice — $115–125 → $125–135 → $150–160. Shares offered went from 28M to 30M. Proceeds went from $3.5B to $4.8B. Subscription orders came in at more than 20× supply, with total bid value over $10B.

---

The OpenAI deal made this IPO

The way to read Cerebras is through its contract structure with OpenAI. This is not a routine supply deal.

The first contract, disclosed in January, supplies 750MW of compute built on Cerebras WSE over 2026–2028 — sized at over $10B. It cuts GPT-model inference latency by up to 15× to speed ChatGPT responses and handle concurrent load.

The expansion announced after April is bigger. Three more years of chip and server supply worth over $20B, up to $30B. OpenAI also invested about $1B into Cerebras data-center design and build. Not a customer relationship — a strategic partner and equity alliance.

For OpenAI the goal is one thing: reduce dependence on the Nvidia H100/Blackwell-centric supply chain. Cerebras being officially designated as a primary alternative chip supplier is the entire valuation case behind this IPO.

---

WSE — why a "dinner-plate chip" is faster than Nvidia

The tech is straightforward. Semiconductors normally cut a wafer into many individual chips. Cerebras does not cut. The entire wafer is the chip. Hence the name Wafer-Scale Engine (WSE). Physically it is roughly dinner-plate-size, which is where the nickname comes from.

When the chip isn't cut, there is no chip-to-chip data movement. That is the speed secret. A standard GPU cluster of hundreds of chips has data hopping between them — that hop is the bottleneck. WSE has no such hop. The company claims up to 20× faster inference throughput than an Nvidia H100 cluster on LLM workloads.

2025 revenue: $510M, up 76% YoY, with a swing to profitability.

---

Sam Altman is an investor. AMD is a shareholder. The cap table tells the story.

The Cerebras shareholder list reveals the positioning directly. Angel investors include Sam Altman (OpenAI CEO), Greg Brockman (OpenAI co-founder), Ilya Sutskever (former OpenAI chief scientist), Lip-Bu Tan (Intel CEO). AMD is a strategic investor. The shared interest in cultivating an Nvidia rival aligns AMD and Cerebras.

Benchmark bet early. Partner Eric Vishria sits on the board and re-upped in the February 2026 Series H. Early entry price isn't disclosed, but at the $160 top of the IPO band the multiple is very large. Tiger Global led the Series H; Coatue, Altimeter Capital, and Abu Dhabi Growth Fund (sovereign-affiliated) also joined.

OpenAI isn't officially a shareholder yet. But with the December 2025 $1B loan it also acquired warrants to buy 33M+ shares. If exercised, OpenAI becomes a major shareholder. Customer becomes equity ally; lender becomes potential shareholder.

Founder/CEO Andrew Feldman holds Class B shares (multi-vote), retaining control after listing. Only Class A shares are sold in the IPO; Feldman sells none of his own stake. Dual-class structure of the same kind Meta and Alphabet use.

---

G42 is out, but customer concentration is still high

The previous IPO failure was about G42. The UAE firm was a single customer accounting for 60–87% of total revenue, and CFIUS blocked listing on concerns that AI technology could leak to sanctioned countries. Cerebras voluntarily withdrew the listing in October 2025. After fully unwinding G42's equity and obtaining CFIUS approval, Cerebras could re-attempt the IPO. That said, G42's customer relationship is not fully unwound either — partial revenue dependence is named in the S-1 risk section.

But customer concentration didn't go away. Mohamed bin Zayed University of AI (MBZUAI) took G42's slot. As of 2025 MBZUAI accounts for ~62% of revenue. A separate UAE entity, not on the CFIUS blocklist. But the structural single-customer-dependence problem persists.

The OpenAI deal can change this. If the $30B long-term contract executes as planned, customer base diversifies. At the same time, a new concentration risk emerges — if OpenAI shakes, Cerebras shakes.

---

Sam Altman's conflict — listed as a formal S-1 risk

One uncomfortable question. Sam Altman is both an early personal investor in Cerebras and the CEO of OpenAI. When OpenAI signs a $10B compute contract with Cerebras, extends a $1B loan, and takes large warrants, Altman has interests on both sides of every decision.

Cerebras lists this as a formal risk in the S-1. There is no record so far of any legal issue or board sanction at OpenAI. But the mere existence of this structure is a risk investors should recognize. Amid 20× oversubscription, this question must not be buried.

---

Litmus-test day

Cerebras has long been called a test for the market's tolerance of AI-infrastructure valuations. If the price clears the top of the band and post-listing trading stays firm, it greenlights the queue of large AI IPOs — SpaceX, OpenAI, Anthropic. If it disappoints, the AI-bubble debate reignites.

20× oversubscription looks like the market has already given its answer. Tomorrow's first day of trading confirms it.