scripod.com

20VC: Cerebras CEO on the Future of Data Centres, Token Costs and Memory | We are Not in an Infra Bubble & Dario Got a Bad Deal with Elon for Compute | Should US Companies Sell to China & Why Most Layoffs are AI Washed with Andrew Feldman

Andrew Feldman, CEO of Cerebras Systems, discusses the company's record-breaking IPO and the state of the AI industry. He argues that the current AI infrastructure buildout is not a bubble but is instead lagging behind surging demand, driven by the recent practical utility of AI models. He also addresses key industry challenges, including memory shortages, enterprise adoption barriers, and geopolitical considerations in chip manufacturing.
Feldman contends that the AI infrastructure buildout is behind demand, not ahead of it, dismissing bubble comparisons. He highlights Sam Altman's foresight in forecasting capital expenditure and notes that memory shortages, particularly in HBM, will persist for years, benefiting Cerebras's SRAM-based architecture. He argues that most layoffs are 'AI-washed,' stemming from COVID-era over-hiring rather than AI, and predicts that more productive engineers will lead to more hiring. The biggest inhibitors to enterprise AI adoption are lawyers and security teams, not technology. Feldman also argues against selling leading-edge chips to China due to military competition and emphasizes the need to onshore chip manufacturing in the U.S. He reflects on the personal cost of building a hardware company, including creating 800 millionaires through the IPO, and the importance of a supportive partner and board during challenging times.
02:37
02:37
Jensen Huang predicts $3-4 trillion in AI infrastructure spending by 2030
06:00
06:00
AI infrastructure buildout is behind demand, not ahead of it
08:00
08:00
Believing exponential growth and acting on it is a superpower.
09:05
09:05
Micron achieves 80-85% gross margins on HBM
15:59
15:59
Cerebras is 15x faster due to architecture
16:40
16:40
Full-stack ownership limits hardware sales to one customer
19:30
19:30
Speed is critical in AI.
27:15
27:15
Data centers must be good neighbors by paying their own way.
30:16
30:16
Most layoffs are AI-washed.
33:44
33:44
The biggest barrier to enterprise AI adoption is data structure and cleanliness.
34:49
34:49
Lawyers and security teams are the biggest inhibitors of enterprise AI adoption
35:38
35:38
Leaders weigh productivity gains against risks
39:22
39:22
Chip industry choke points make it manageable
45:55
45:55
IPO timing was driven by effort and luck.
51:01
51:01
Money changes entrepreneurs and investors.
53:28
53:28
The personal cost of building a hardware company