scripod.com

Cerebras IPO

Semi Doped

Shownote

Cerebras IPO is the only thing to talk about this week. IPO prices at $185/share. Pops nearly 70% right after. The first wafer-scale chip company to make it public — after a 40-year curse killed every prior attempt. A water-cooler-style convo on what Ce...

Highlights

This podcast discusses the recent Cerebras IPO, which saw its stock price surge nearly 70% on the first day. The conversation explores the unique technology behind Cerebras' wafer-scale chip, its engineering challenges, and its market position in the AI inference space.
00:00
Each Cerebras wafer consumes 23 kilowatts of power
00:20
Small steps lead to big changes
05:23
They route around defective cores to achieve ~900,000 working ones.
10:32
23 kW power draw requires vertical connectors across hundreds of points
18:12
Small models on a single wafer achieve unmatched token speeds
29:09
Cerebras succeeded after 40 years
34:48
Low latency is key for coding and trading.
47:17
The AI inference market is a 'Wild West'.

Chapters

Cold open: 23 kW per wafer
00:00
Cerebras IPO day at $185
00:15
What's a wafer-scale engine
02:39
Power, cooling, and thermal expansion
10:30
The 44 GB wall
18:12
The Trilogy Systems curse
26:35
Supercomputing → training → inference
32:11
The OpenAI deal and the Wild West
39:36

Transcript

Vikram Sekar: Each wafer consumes about 23 kilowatts of power. It's like, enormous. Like, if you think about a one-volt supply that is feeding these GPUs, you're talking about something. In the tens of thousands of amps of current that have to flow into a ...