Rotating Data
Explore the map
Home/Deal Room/Cerebras IPO memo surfaces OpenAI inference backlog economics

Chips / compute · Evidence brief

Cerebras IPO memo surfaces OpenAI inference backlog economics

Cerebras Systems + OpenAI + AWS + WhiteFiber

Market reportingSEC summary2026-05-08
Deal structureIPO source trail / inference capacity backlog
GeographyUnited States + global cloud customers
Infrastructure layersChips + inference cloud + data centers
Disclosed scale$20B OpenAI take-or-pay agreement cited; 750 MW contracted base; option to expand toward ~2 GW

Why it matters

The strategic read

Frames Cerebras as a latency-sensitive inference infrastructure bet: proprietary wafer-scale chips, OpenAI coding-agent demand, AWS distribution, and colocation partners all tied to a large contracted backlog ahead of the IPO.

Source-supported terms

What was disclosed

  1. 01

    VanEck memo says Cerebras is seeking to raise roughly $3.5B by selling 28M shares at $115–125 in the most significant semiconductor IPO since Arm.

  2. 02

    Memo cites a $20B OpenAI take-or-pay agreement covering 750 MW of contracted capacity through 2028, with an option to expand toward ~2 GW for an additional ~$34B.

  3. 03

    Sigel points to Codex Spark usage as a public proxy for low-latency coding-agent demand, growing from roughly 600K WAUs in January 2026 to 4M+.

  4. 04

    AWS Bedrock integration, Meta Llama inference, Oracle, IBM, and WhiteFiber colocation are cited as additional distribution/capacity signals.

Disclosure boundary

What is known—and what is not

Investment memo/source trail, not a full public contract; treat the OpenAI backlog, MW terms, and valuation cases as cited claims requiring confirmation against IPO filings and definitive agreements.