The Twenty Minute VCCerebras CEO, Andrew Feldman on Why Raise $1BN and Delay the IPO & Why NVIDIA’s Worried About Growth
At a glance
WHAT IT’S REALLY ABOUT
Cerebras CEO: Beating NVIDIA, Billion-Dollar Bets, And AI’s Future
- Andrew Feldman, CEO of Cerebras, explains why the company raised a $1B pre-IPO round, emphasizing capital to scale manufacturing, data centers, and ambitious AI hardware R&D while delaying going public. He argues that AI demand is vastly underestimated, with customers unable to forecast usage even within an order of magnitude, making capacity ‘options on the future.’
- Feldman offers a critical view of NVIDIA’s dominance, predicting limits to its growth and describing classic incumbent behaviors like buying business with investments and using pre-announcements to slow competitors. He dives into chip economics—depreciation, memory bottlenecks, wafer-scale design, and margins—framing Cerebras as faster on both training and inference, but acknowledging software and adoption frictions.
- Beyond chips, he explores systemic constraints in AI: talent shortages, fabrication and data center bottlenecks, power-location mismatches, and underinvestment in unglamorous areas like data pipelines. He also reflects on geopolitical dynamics (US–China, Gulf states), immigration, and how AI will diffuse slowly but deeply into productivity, education, and white-collar work.
- Throughout, Feldman underscores the value of extraordinary talent, the risks of mispriced concentration in Mag 7 stocks, and the reality that building AI hardware is a long-horizon, brutally hard, experience-driven game where most newcomers and naïve capital will be wiped out.
IDEAS WORTH REMEMBERING
5 ideasUse late-stage capital to buy strategic flexibility, not just runway.
Cerebras’ $1B raise, led by Fidelity, is about scaling manufacturing, adding data centers, and funding non-incremental R&D while retaining the option to IPO later; securing blue-chip public investors pre-IPO also sends a strong signaling effect to future public markets.
Treat AI infrastructure commitments as options on an uncertain future.
Customers requesting anywhere from 5–40M queries per second show that demand forecasts are off by orders of magnitude; large ‘up to’ multi-year deals are less firm orders and more capacity options in a rapidly shifting environment.
Depreciation of AI chips hinges on solution-level speedups, not marketing specs.
Feldman stresses that useful chip life depends on how much faster complete systems (including memory bandwidth and power efficiency) become; modest 2–2.5x generational gains mean H100s and even A100s still hold value, extending their economic life.
Memory, not raw FLOPS, is the key bottleneck in AI hardware.
GPU architectures are constrained by slow high-capacity HBM, so Cerebras built wafer-scale chips packed with fast SRAM to keep models on-chip; the move sounds obvious but had been technically impossible for 75 years and required solving a fundamental manufacturing problem.
Expect incumbent behavior from NVIDIA: buying demand and pre-announcing roadmaps.
Feldman interprets NVIDIA’s massive strategic investments (e.g., in OpenAI) and aggressive B200/B300/Rubin pre-announcements as classic signs of a giant using its balance sheet and marketing to defend growth and delay customers from adopting rival technologies.
WORDS WORTH SAVING
5 quotesThere is unbelievable demand and nobody knows where it will go in the future.
— Andrew Feldman
If NVIDIA keeps growing at the rate they're currently growing, 11 years from now, everybody on Earth works for them.
— Andrew Feldman
The question of depreciation is how much faster are future generations than the current generation? That's the actual question on depreciation.
— Andrew Feldman
No company ever went bankrupt by paying extraordinary people too much. If you want to go bankrupt, pay mediocre people too much.
— Andrew Feldman
We have just solved the problem that for 75 years the smartest people in our industry had been unable to solve. And we have done it.
— Andrew Feldman
High quality AI-generated summary created from speaker-labeled transcript.
Get more out of YouTube videos.
High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.
Add to Chrome