Cerebras Systems shares opened Tuesday at $291.50, carrying a market capitalization of approximately $61.8 billion on the outstanding share count from the May 14 IPO, with the company's S-1-disclosed OpenAI commitment of 750 megawatts through 2028 — valued at more than $20 billion — operating as the structural ceiling on the rally [1]. The paper's Monday standard on the day-eight past-IPO tape at $185 framed the concentration question; the May 23 brief on the day-seven 10% slide tracked the post-IPO peak.

The mechanics. CBRS priced at $185 on May 13; opened at $350 on May 14 in a 90% first-day pop; reached an intraday peak of $385; closed Thursday at $311.07; closed Friday May 15 down 10% at approximately $280; and has traded between $270 and $295 for the week ending Friday May 22 [2]. At Tuesday's $291.50 open, the implied market cap of $61.8 billion is fifteen times the company's $510 million 2025 revenue — a 121x sales multiple. Cerebras's GAAP net income for 2025 was $237.8 million, following a large prior-year loss. The valuation is among the most aggressive AI-IPO multiples since the Snowflake debut.

The OpenAI concentration is structural. Per the S-1, OpenAI has committed to 750 megawatts of Cerebras-deployed compute capacity across the 2026-2028 deployment window, with payments valued by independent estimates at more than $20 billion over the contract term [3]. The first 250 MW must be operational by end of 2026 — a capacity scale-up Reddit and engineering forums have flagged as ambitious given Cerebras currently operates at roughly 35-50 MW [4]. The deal is in two earlier disclosed buckets: an initial 250 MW agreement announced January 2026 valued at over $10 billion, then a $20 billion total revised in May.

AWS deploys Cerebras CS-3 systems via Amazon Bedrock as the second-customer channel — a partnership announced in March 2026 that gives Cerebras a hyperscaler-channel deployment route distinct from the on-premise model OpenAI is buying [3]. The AWS revenue scale is significantly smaller than OpenAI's commitment but provides the only public-record diversification leg. UAE-based G42's Condor Galaxy partnership produces additional dependency along the sovereign-AI dimension; the $7.1 billion multi-year contract disclosed in the S-1 adds capacity and reach but does not break the OpenAI concentration.

The Anthropic comparison surfaces the absolute scale of Cerebras's valuation question. Anthropic's $900 billion private mark — most recently rumored in the venture-capital cycle around the May SpaceX-compute disclosure — is about fifteen times Cerebras's $61.8 billion market cap. The two companies, both pure-play AI infrastructure plays, sit at fundamentally different scales of compute supply. The structural question for Cerebras's revaluation is whether customer diversification can come at sufficient pace to justify a 96x sales multiple while OpenAI's concentration ceiling holds.

The S-1's customer-concentration disclosures are explicit. As of the end of 2025, Cerebras's $24.6 billion contracted backlog was disproportionately attributable to OpenAI, with only about 15% expected to be recognized as revenue through 2027 [5]. The remainder is contingent on data-center capacity, power availability, and customer acceptance — three operational variables Cerebras has limited control over. Add to that the manufacturing concentration through Taiwan Semiconductor, whose leading-edge capacity is under pressure from multiple major customers, and the operational risk profile is meaningful.

The competitive context is what makes Anthropic-related-news Tuesday's tape-mover. The May 18 Anthropic acquisition of Stainless creates concentration in developer tooling; the May 20 SpaceX disclosure of Anthropic's $45 billion compute commitment to Colossus I and II creates concentration in compute supply. Both moves operate against an Cerebras concentration bet by signaling that the AI infrastructure stack is consolidating into Anthropic-Microsoft-SpaceX rather than diversifying. Cerebras's pure-play position relative to that consolidation is what investors are reading; what they price is a different question.

Sovereign-AI customer announcements are what could move the read. The G42 Condor Galaxy partnership is one such commit; Saudi Arabia's HUMAIN, the UAE's MGX, and Singapore's AI initiatives are candidates for similar agreements. Whether Cerebras lands a Saudi or UAE sovereign-AI announcement before Q2 earnings would meaningfully shift the concentration read. The S-1 lists "sovereign AI" exposure as a strategic priority; the public-record cadence has been quarterly.

The Tuesday operational watch is whether AWS Bedrock issues a usage update or whether Cerebras files an S-1/A amendment disclosing new sovereign-AI commits. Either lands the rerating cycle in motion. Without either, the 121x sales multiple sits on top of a structural ceiling that the public AI sell-side has not yet been comfortable lifting.

The encyclical resonance lands here too. Pope Leo XIV's May 19 Magnifica humanitas warning about "concentration of power" reads against the Cerebras structure: the IPO is a public listing of a company whose contracted revenue is more than 80% one customer in 2027. The Vatican is not naming firms. Cerebras and OpenAI's mutual concentration is one of the structures the warning addresses.

-- THEO KAPLAN, San Francisco