Cerebras stock hits post-IPO low, tumbling 20% for the week on Nvidia pressure and lockup expiration

“Cerebras is being re-rated from a 'Nvidia-killer' platform play to a specialized hardware vendor, which cannot support its previous $95 billion valuation.”

Cerebras is suffering from a classic 'valuation-reality' collision. The $95 billion IPO valuation was never tethered to current revenue, but rather to the narrative that they were the sole 'Nvidia killer.' The shift to Nvidia for GPT-6.1 inference isn't just a technical blow; it signals that Cerebras’s Wafer-Scale Engine is a niche tool rather than a general-purpose replacement for H100/Blackwell infrastructure. With 8% of float unlocking and insiders aggressively offloading $240 million in stock, the supply-demand imbalance is acute. Until Cerebras proves their WSE-3 chips can scale beyond specialized inference tasks, the stock is likely to continue its reversion toward a valuation more consistent with a high-growth niche hardware provider.

“The market is pricing in inference loss to Nvidia, but Altman's statement and the intact $10B training deal suggest the bear case is already baked into the 59% post-IPO decline.”

The article conflates two separate pressures—a SemiAnalysis rumor about GPT-6.1 Sol inference workloads and lockup expiration—into a coherent bear thesis. But the rumor is unverified speculation from a research firm, not OpenAI confirmation. Altman's response explicitly states Cerebras remains 'a close partner' with 'deep engagement,' which directly contradicts the inference-loss narrative. The $10B training deal through 2028 remains intact and unmentioned in the selloff framing. Lockup expiration is mechanical and priced in; insider selling ($240M) is material but not disqualifying. The real risk: if inference truly becomes dominant by 2030 and Cerebras loses material OpenAI share to Nvidia, the $39B valuation (down from $95B) reflects that scenario already. The stock may be fairly valued here, not crushed.

“OpenAI's switch to Nvidia for a flagship inference workload plus repeated insider selling point to further downside for Cerebras beyond the recent 20% drop.”

The article shows Nvidia extending its grip on AI inference workloads, as OpenAI routes GPT-6.1 Sol's Ultrafast mode to NVDA GPUs rather than Cerebras ASICs. Combined with 8% insider share unlocks and $240 million in executive sales, this has halved Cerebras' market cap from $95 billion at IPO to $39 billion. McKinsey's forecast that inference will dominate data-center spend by decade-end amplifies the risk that Cerebras loses ground on its $10 billion OpenAI contract. Altman's vague reassurance does little to offset the concrete hardware decision or the ongoing lockup schedule through 2025.

“The sell-off is overly pessimistic until there is concrete evidence that OpenAI is abandoning Cerebras, because a single rumor doesn’t prove a fundamental revenue loss.”

This reads as a headline risk: Nvidia appears to win a single OpenAI workload, yet the evidence is rumor-based and OpenAI publicly hedged on the partnership. The stock drop seems driven as much by 8% unlocks and insider selling as by a durable loss of OpenAI revenue. The real question is OpenAI’s total compute mix through 2028 and beyond, and Cerebras’ ability to diversify beyond a single customer. If Nvidia scales as OpenAI’s preferred backbone, the shares could stabilize or rebound—provided Cerebras secures other customers and proves margin resilience in a cloud business that relies on capital intensity.