20VC: Cerebras CEO on Why Raise $1BN and Delay the IPO | NVIDIA Showing Signs They Are Worried About Growth | Concentration of Value in Mag7: Will the AI Train Come to a Halt | Can the US Supply the Energy for AI with Andrew Feldman
Timestamps are as accurate as they can be but may be slightly off. We encourage you to listen to the full context.
Podcast Summary
In this insightful episode, Andrew Feldman, co-founder and CEO of Cerebras, joins Harry Stebbings to discuss the company's record-breaking $1.1 billion Series G funding round at an $8.1 billion valuation. (00:47) The conversation explores the explosive growth in AI demand that has caught even industry experts off guard, with customers requesting orders that vary by factors of 35 million queries per second. (07:53) Feldman explains how Cerebras solved a 75-year-old chip manufacturing problem by creating wafer-scale processors that overcome traditional SRAM limitations through sheer silicon real estate. (20:46) The discussion covers everything from NVIDIA's market dominance and the infrastructure challenges facing AI adoption, to the geopolitical implications of the US-China AI race and the fundamental changes AI will bring to education and work.
- Main Theme: The episode focuses on navigating unprecedented AI market growth, the technical and business challenges of competing with NVIDIA, and the infrastructure investments needed to support the AI revolution while maintaining competitive positioning in a rapidly evolving landscape.
Speakers
Andrew Feldman
Co-founder and CEO of Cerebras Systems, Andrew Feldman leads the company building the world's fastest AI inference and training chips. Before Cerebras, he was a successful entrepreneur who previously competed against Cisco for 15 years in the networking space. Under his leadership, Cerebras has achieved breakthrough wafer-scale chip manufacturing that solved a 75-year-old industry problem, with the company recently raising a $1.1 billion Series G round at an $8.1 billion valuation from premier investors including Fidelity and Tiger Global.
Harry Stebbings
Host of 20VC podcast and managing partner at 20VC, Harry Stebbings interviews leading entrepreneurs and investors. Known for his engaging interview style and ability to extract actionable insights from complex technical topics, Stebbings has built one of the most respected venture capital podcasts in the industry.
Key Takeaways
Plan More Frequently in Uncertain Environments
When facing unprecedented market conditions, traditional long-term planning becomes less effective than adopting flexible planning frameworks. (08:42) Feldman emphasizes that in rapidly moving environments, companies need "good planning changing rules rather than good planning." The key is to plan more frequently, maintain shorter-term views, and take strategic options on the future rather than making rigid multi-year commitments. This approach allows companies to adapt quickly when market conditions shift dramatically, as they inevitably will in emerging technologies. For professionals, this means developing comfort with ambiguity and building systems that can pivot quickly rather than betting everything on a single long-term vision.
Focus on System-Level Performance, Not Component Performance
Many companies make the mistake of optimizing individual components rather than overall system performance. (17:20) Feldman explains that having faster chips doesn't matter if memory bandwidth can't keep up - "It doesn't matter how many flops your chip has. If you can't get data onto and off of the chip, those are wasted." This principle applies beyond hardware to business strategy: optimizing one department or metric in isolation can create bottlenecks elsewhere. Successful professionals and companies must think holistically about how different components work together to deliver end-to-end value rather than pursuing point solutions that may actually harm overall performance.
Invest in Unsexy but Critical Infrastructure
While everyone focuses on glamorous AI breakthroughs, the most valuable opportunities often lie in boring, foundational work. (55:02) Feldman points out that "nobody puts on their LinkedIn data pipeline expert, and yet these are some extraordinarily valuable cats." Data cleaning, pipeline management, and tokenization are where many AI projects actually fail, not because of algorithmic issues. This represents a massive investment opportunity that's being overlooked. For career development, becoming exceptionally skilled in these foundational areas can provide enormous leverage and job security, as these skills become increasingly critical as AI adoption scales.
Use Balance Sheet Strategically When Technology Advantages Wane
Large companies facing competitive threats often shift from competing on technology to using financial resources for market control. (11:49) Feldman observes that when companies worry about their technical prowess, they "use your balance sheet more and your technology less" through acquisitions and strategic investments like NVIDIA's reported $100 billion commitment to OpenAI. Recognizing this pattern helps both entrepreneurs understand when they have windows of opportunity against larger competitors, and helps investors identify when market leaders may be vulnerable. For professionals, understanding whether your company is competing on innovation versus financial muscle helps inform career and strategic decisions.
Pay Extraordinary People Whatever It Takes
In rapidly scaling markets, the value created by exceptional talent far exceeds their compensation costs. (34:39) Feldman states definitively: "No company ever went bankrupt by paying extraordinary people too much." When someone can add $50 billion in enterprise value, paying them even $1 billion is economically rational. The real risk is "paying mediocre people too much." This principle applies beyond just executive compensation - identifying and investing in truly exceptional talent, whether as a manager, entrepreneur, or investor, provides asymmetric returns. For individual career development, this suggests focusing on becoming genuinely exceptional in valuable skills rather than seeking incremental improvements across broad areas.
Statistics & Facts
- Andrew Feldman revealed that 75-80% of Cerebras's 2024 revenues came from UAE customers, with orders so large they consumed the company's entire manufacturing capacity. (49:39) This concentration highlights how aggressive early adopters in certain regions are driving AI infrastructure investment.
- Feldman noted that customers are requesting between 5 and 40 million queries per second, representing uncertainty by a factor of 35 million in demand forecasting. (07:53) This massive range demonstrates the unprecedented unpredictability in AI market sizing.
- NVIDIA's gross margins on high-end chips reach 78-85%, according to Feldman, which explains why companies like AWS are motivated to build their own training chips to escape these extraordinary margin levels. (43:28)
