J Keynes's $CBRS thesis: Cerebras speeds model research loops

J Keynes

Growth Investment Research

10 Nov, 21:18Latest

Cerebras could help frontier labs develop successor models faster because its inference speed shortens dependent research-agent loops on the path toward recursive self-improvement. OpenAI describes agents assisting researchers; Feldman says OpenAI initially devoted all its Cerebras capacity to engineers for six months, and OpenAI confirms internal research use. Anthropic has reason to evaluate Cerebras for research agents, but a Fable deployment would need integration and benchmarks. x.com

CBRS

5 Nov, 18:16

Cerebras's temperature and fleet data undercut Culper's two-year wafer-life premise. Cerebras says TSMC qualifies wafers for 10 years at 105°C; its heat model gives 33 years at 85°C, not expected use. Fewer than 0.5% of deployed CS-3 wafers have been decommissioned since 2024. Its 3-5-year depreciation reflects lab use and cloud assets' economic life. Still among AI's most compelling pure plays, Cerebras could reach $300B, with a path to $1T over the next decade if things go well. x.com x.com x.com x.com

CBRS

This signal is no longer available. Showing the latest signals.

Ronnie Chatterji

11 Nov, 08:01

Chief Economist of OpenAI, Distinguished Professor at Duke University

We released a report on research acceleration, it's one of my most interesting reports right now, looking at how AI helps AI researchers, but also helps people across the company. If you think about three times the human hours, we're going to open up a lot of opportunities here. pdrl.fm

AlvaSupportsOpenAI frames AI-assisted research as potentially providing three times the human hours, supporting the thesis that agents can expand capacity for successor-model research.

Andrew Feldman

8 Nov, 14:05

Co-Founder & CEO of Cerebras

A person can do it by hand in 37 seconds. The assistants took up to 7 minutes and 40 seconds. So we built a prototype assistant on @Cerebras, tuned for this task. It booked the same table in 22 seconds, faster than a person could. x.com

AlvaSupportsThe reported 22-second booking supports Cerebras’s speed case for consumer agents, alongside parallel searches and task-specific tuning; the result covers one task.

SemiAnalysis

6 Nov, 04:59

Independent Semiconductor and AI Research Firm

OpenAI just announced that it will deprecate GPT-5.6 Sol Ultra Fast. This is the same model that Cerebras publicly announced it ran at ultra-fast speeds for. x.com

AlvaChallengesRetiring this Cerebras-powered ultrafast service creates a concrete checkpoint for whether its OpenAI workloads carry over to replacement models.

Ask Alva about this thesis

Ask Alva about this thesis