Tiny Spoon

⚡ Why this matters

• First publicly-listed Nvidia challenger.
• Validates Ben Thompson's Inference Shift call.
• Specialty silicon is now a real public-market category.

🔍 What happened

• May 14, 2026. Cerebras Systems IPO Nasdaq debut.
• $185 pricing above raised range. 30M shares. $5.55B raised.
• Opened $385 (+108%). Closed +68% at $311.07.
• $56.4B fully-diluted at IPO. $95B at peak.
• G42 dependence reduced from 85% to 24%.
• 2025 revenue $510M (+76% YoY). Net income swing to +$237.8M.
• Customers: OpenAI, G42, MBZUAI, AWS.
• Wafer Scale Engine 3.

💬 Smart takes

• Ben Thompson: compute is splitting into answer-inference (speed wins) and agentic-inference (memory wins).
• Skeptic: G42 dependence reduction is a thumb on the scale. Underlying revenue concentration is still real. Cerebras's whole-wafer approach has yield challenges that drive costs. Long-term viability depends on whether the answer-inference market is big enough to support specialty silicon at premium.

🧭 Where this goes

1. Groq pursues an IPO or strategic acquisition within 12 months.
2. Watch the next big Cerebras customer signing. That's the read on whether Nvidia's premium starts to compress.
3. AWS, Azure, GCP roll out Cerebras-as-a-service offerings by Q3.
4. Specialty-silicon procurement category opens for enterprises by Q4.

🎯 Implication

• For execs tracking AI vendor risk: compute supplier concentration matters. The post-Cerebras-IPO multi-vendor compute story is now real.
• For CFOs: model your compute spend with specialty-silicon scenarios. Cerebras pricing pressure on Nvidia is real but slow.