Cerebras is built for AI inference. Why Shay just bought in.

3 Просмотры

Спасибо! Поделитесь с друзьями!

Вам не понравилось видео. Спасибо за то что поделились своим мнением!

Издатель Oct 1, 2026

Cerebras (CBRS) builds wafer-scale processors for AI inference, and OpenAI now runs an ultra-fast service tier for its most capable model on Cerebras hardware. Chief Market Strategist Shay Boloor opened a speculative starter position in Cerebras in his family portfolio. He walks through why inference speed and power efficiency make the case, and why the valuation keeps the position small.

The numbers:
• OpenAI agreement: 750 megawatts of inference capacity and more than $20B of potential value
• About $25B of remaining performance obligations (RPO), the majority from OpenAI
• Cerebras says the OpenAI configuration delivers frontier intelligence at about 14x the speed of standard processing and around 750 output tokens per second
• CS4: management says up to 2x the performance of CS3 and up to 30x more tokens per second per user than GPU-based alternatives in certain workloads
• For comparison, NVIDIA is growing its top line 100% while trading at 18x earnings

Shay's read: the AI stack is becoming specialized. Training, prefill, reasoning, decode and agent orchestration do not have to run on the same processor, and AWS already pairs Cerebras with Trainium in a disaggregated inference architecture. Cerebras does not have to replace NVIDIA for the thesis to work. It has to become a preferred engine for latency-sensitive inference, where a coding agent multiplies every second of delay across its whole workflow. With power scarce, Shay cares more about revenue and gross profit per megawatt than any standalone tokens per second benchmark. The risks he names are converting RPO into revenue, unproven gross margins, concentration in OpenAI and a valuation that already prices in growth. If inference becomes as important as he expects, Cerebras is built around the constraints that matter most: latency, throughput, power efficiency and useful AI work per megawatt.

Chapters:
0:00 The starter position
1:41 Inference and decode
3:54 The OpenAI deal
6:12 Tokens per megawatt
9:30 Agentic AI demand
11:06 The valuation risk

About FE Earnings Edge:
FE Earnings Edge is Futurum Equities' rapid-reaction earnings series with Chief Market Strategist Shay Boloor (@StockSavvyShay), breaking down the prints that move the AI, semiconductor, space, energy, security, and robotics names the firm covers.

Disclaimer: This content is for informational and educational purposes only and should not be considered financial advice. Always do your own research.