Skip to main content
AF

Andrew Feldman

Cerebras CEO Andrew Feldman and Planet**ipo Reality Check**space-based Data Centers Timeline**ai Silicon Architecture Principle**post-ipo Value Capture
5episodes
4podcasts

Featured On 4 Podcasts

Top resources Andrew Feldman mentions

Books, tools, and gear cited across podcast appearances. Ranked by frequency.

SignalCast may earn commission on purchases via affiliate links on each resource page.

All Appearances

5 episodes

AI Summary

→ WHAT IT COVERS Cerebras CEO Andrew Feldman and Planet Labs CEO Will Marshall join Brad Gerstner at an All-In liquidity panel to discuss their IPO experiences, the convergence of AI and space infrastructure, next-generation silicon architecture, and why public market investors may capture more value than private ones in the current tech cycle. → KEY INSIGHTS - **IPO Reality Check:** Going public changes almost nothing operationally. Cerebras priced at $18.50, opened at $32, reached a $5–6B market cap, yet Feldman notes that vendor relationships, engineering progress, and sales pipelines remain identical the morning after listing. The primary tangible benefits are employee morale, balance sheet cash, and enterprise credibility with risk-averse customers. - **Space-Based Data Centers Timeline:** Planet Labs and Google's analysis shows space-based compute becomes cheaper than terrestrial data centers when launch costs reach $200–300 per kilogram. Current costs sit just above $1,000/kg, down 10x over a decade. Solar panels in sun-synchronous orbit generate five times more energy than ground-based panels with zero battery requirement, making the infrastructure model straightforward once launch economics close. - **AI Silicon Architecture Principle:** Building a chip that resembles a competitor's design yields approximately zero chance of outperforming them. Cerebras solved AI's core bottleneck — moving data between memory and compute — by building a dinner-plate-sized chip with on-chip SRAM directly adjacent to compute, delivering 15–18x speed advantage over GPUs for OpenAI workloads. Domain-specific architecture, not incremental GPU iteration, drives step-change performance gains. - **Post-IPO Value Capture:** Historical data consistently shows more absolute dollar value is created after IPO than before. Planet Labs stock moved from $5 to $50 — a 10x gain — entirely in public markets after a 2021 SPAC listing. LP pressure to distribute shares immediately post-lockup causes funds to forfeit the majority of returns, as demonstrated by Altimeter's MongoDB investment that went from $3–4B at distribution to $50B shortly after. - **Earth Data as AI's Missing Layer:** Current large language models are trained exclusively on internet text and lack real-world physical data. Planet Labs images the entire Earth daily across a 200-satellite fleet, creating a time-series dataset covering agriculture, energy, flooding, and security. Feeding this physical-world data into AI models unlocks what Marshall calls "planetary intelligence" — AI capable of answering real-world operational questions, not just text-based ones. → NOTABLE MOMENT Feldman recounted that when Cerebras brought employees who had worked nine-plus years to the NYSE floor alongside their families, he discovered engineers actually own ties — and that the event carried the emotional weight of a family milestone, particularly for children of immigrants whose parents had waited a decade for this moment. 💼 SPONSORS None detected 🏷️ IPO Strategy, AI Silicon, Space Infrastructure, Public Market Investing, Earth Observation

AI Summary

→ WHAT IT COVERS Cerebras CEO Andrew Feldman, following the company's landmark semiconductor IPO that priced at $185 and closed at $311, examines AI infrastructure demand outpacing supply, memory shortages persisting for years, why most corporate layoffs are COVID-hiring corrections rather than AI displacement, and why selling advanced chips to China represents a strategic mistake regardless of revenue opportunity. → KEY INSIGHTS - **AI Infrastructure vs. Bubble Logic:** Cerebras carries a $25 billion backlog because data center construction cannot match current demand — the opposite of historical bubbles like 1990s fiber optics, where supply preceded demand. Builders are not speculating on future customers; they are already behind on confirmed orders. Recognizing this distinction helps investors and operators avoid misreading infrastructure spending as speculative overreach. - **Memory Supply Constraint Timeline:** HBM memory, produced only by Samsung, Micron, and SK Hynix, faces shortages expected to last several years because fab capacity expands in $40 billion step-function increments requiring five-year build timelines. Micron currently earns 85% gross margins — software-level profitability on hardware — signaling extreme supply tightness. Companies architecting around HBM dependency, as Cerebras does using on-chip SRAM, avoid this bottleneck entirely. - **Layoff Misattribution to AI:** Feldman argues 90–95% of current corporate layoffs stem from COVID-era overhiring and productivity gains from pre-AI tooling, not generative AI displacement. Middle management roles built around information gathering and synthesis are shrinking due to software consolidation, not LLMs. Leaders should separate genuine AI workforce impact from restructuring that would have occurred regardless, avoiding both panic and complacency about actual AI-driven job transformation. - **Enterprise AI Adoption Blockers:** Legal and security teams represent the primary brake on enterprise AI adoption, not data quality or infrastructure readiness. Their incentive structure — zero credit for successes, full blame for failures — creates systematic risk aversion toward unproven technology. Organizations that want faster adoption should have senior leaders set explicit governance frameworks and risk tolerances, removing the decision burden from legal and security functions not designed to approve novel tools. - **China Chip Sales Strategic Error:** Selling leading-edge semiconductors to China benefits adversaries militarily and industrially, regardless of revenue upside. Feldman points to solar, lithium batteries, and automotive sectors as evidence of China's industrial policy effectiveness when given technology access. Credible chokepoints exist through TSMC and ASML dependencies, making export controls enforceable. US chip companies arguing for continued sales to preserve ecosystem influence underestimate the compounding industrial disadvantage created over time. - **Speed as Infinite-Value Differentiator:** Cerebras runs models like Kimi K2 at 6.7x faster throughput than competing GPU clouds, and Feldman frames inference speed as having no upper bound on value — analogous to how there is zero market for slow internet regardless of price. For agentic workflows and coding tasks specifically, a 3-minute versus 20-minute task completion gap compounds dramatically across workdays, making speed a structural competitive moat rather than a marginal performance feature. → NOTABLE MOMENT Feldman describes an 18-month period where Cerebras burned $8 million monthly without solving its core technical problem, returning to board meetings repeatedly with no progress. He frames this as some of his proudest work — each failure advanced the failure point from two seconds to one hour, until the problem was eventually solved by no competitor since. 💼 SPONSORS [{"name": "Base44", "url": "https://base44.com"}, {"name": "Corgi Insurance", "url": "https://corgi.com/20vc"}, {"name": "Turing", "url": "https://turing.com/20vc"}] 🏷️ AI Infrastructure, Semiconductor Supply Chain, Enterprise AI Adoption, US-China Tech Policy, Inference Speed, AI Workforce Impact

AI Summary

→ WHAT IT COVERS Cerebras founder and CEO Andrew Feldman discusses the company's path from a contrarian wafer-scale chip architecture to a $63 billion public company, covering the 2017–2019 technical breakthrough period, the G42 billion-dollar bridge deal, the $20 billion OpenAI agreement, and why inference speed becomes the defining competitive advantage once AI reaches daily utility. → KEY INSIGHTS - **Radical differentiation threshold:** Achieving 15–20x performance improvement over GPUs requires fundamentally different architecture, not incremental modification. Cerebras built a 46,000 square millimeter wafer-scale chip — the size of a dinner plate — versus competitors' postage-stamp chips. Hardware founders targeting radical gains should design from first principles rather than optimizing existing architectures. - **Market timing for hardware:** Speed advantages have zero commercial value until the underlying technology reaches daily utility. Cerebras was 15–20x faster than GPUs from 2019 onward but generated minimal sales until 2025, when AI models became useful enough for daily work. Hardware founders should plan financially for a 3–5 year gap between technical readiness and market readiness. - **Bridge customer strategy:** To cross the chasm between niche early adopters and mainstream enterprise customers, Cerebras secured a $1 billion order from sovereign partner G42. This single deal funded supply chain transformation, enabled large-scale cluster deployment for battle-testing, and built the operational capacity needed to fulfill the subsequent $20 billion OpenAI agreement. - **Accountability against the sunk-cost trap:** Founders should pre-define specific, falsifiable hypotheses about what conditions must be true to continue. Trusted former CEOs or seasoned operators serve as external accountability partners who can remind founders of their own stated exit criteria, preventing the sequential "one more test" rationalization that extends failing ventures indefinitely. - **AI coding productivity distribution:** Cerebras increased per-engineer token spend from near zero to $25,000–$30,000 monthly within eight months. Productivity gains are highly uneven: engineers who restructure their workflow around governing multiple parallel agents simultaneously — including dedicated QA agents — move from 10x to 100x output, while others see marginal gains. → NOTABLE MOMENT During the $20 billion OpenAI deal negotiation, Cerebras and OpenAI executed a term sheet the night before Thanksgiving and signed a full master agreement on December 24 — a four-and-a-half-week close on one of Silicon Valley's largest contracts, achieved by working seven days a week with multiple law firms simultaneously. 💼 SPONSORS None detected 🏷️ AI Hardware, Inference Speed, Semiconductor Architecture, IPO Strategy, Founder Psychology

AI Summary

→ WHAT IT COVERS Cerebras CEO Andrew Feldman explains how his company built a chip 58 times larger than any competitor, achieving inference speeds 15 times faster than leading GPUs. The episode covers wafer-scale engineering breakthroughs, inference economics, CUDA's declining relevance, open vs. closed source AI models, and semiconductor supply chain constraints. → KEY INSIGHTS - **Wafer-Scale Memory Architecture:** Cerebras achieves 15x faster inference than GPUs—and up to 1,000x faster on specific workloads—by using fast SRAM instead of slow HBM memory. The tradeoff is lower storage density per square millimeter, solved by building a chip covering an entire silicon wafer, roughly dinner-plate sized, stuffed with high-speed memory. - **Speed Premium Pricing:** Anthropic's 2x-faster inference tier sold out at 6x the standard price, demonstrating that enterprise buyers pay significant premiums for speed. Cerebras operates at 15x faster than that tier, suggesting substantial pricing power. Slow tokens cost less to produce on GPUs, but GPU cost-per-token rises sharply as speed requirements increase. - **Supply Chain Differentiation:** Cerebras avoids three major AI chip bottlenecks simultaneously: HBM memory shortages, TSMC's constrained CoWoS packaging process, and TSMC's oversubscribed 3nm node. By using 5nm fabrication and on-chip SRAM, Cerebras sidesteps constraints choking NVIDIA and other GPU vendors, leaving data center availability as the primary growth limiter. - **CUDA Moat Erosion:** CUDA has zero role in inference workloads—migrating a model from GPU to Cerebras requires roughly 10 configuration changes. In training, two of three leading frontier models (Gemini on TPUs, Claude on Trainium) now train without CUDA, representing a 70% market share loss for NVIDIA's software ecosystem compared to three years ago. - **Open vs. Closed Source Economics:** Open source models like Kimi K2 (1 trillion parameters) run on Cerebras today at a cost reflecting only compute and power—not training amortization. Closed source models outperform open source by roughly 4–5% on quality benchmarks but cost significantly more per token, creating a cost-versus-capability tradeoff enterprises must actively evaluate. → NOTABLE MOMENT Feldman reveals that despite solving a 75-year-old unsolvable engineering problem and building the world's fastest inference chip, Cerebras' primary growth constraint today is not manufacturing capacity or software—it is simply the availability of powered data center buildings, a limitation expected to persist for at least 15–18 months. 💼 SPONSORS [{"name": "VanEck", "url": "https://vaneck.com/raaxpod"}, {"name": "IBM", "url": "https://ibm.com"}, {"name": "Adobe Acrobat", "url": "https://adobe.com"}, {"name": "Public", "url": "https://public.com/market"}] 🏷️ AI Inference, Semiconductor Architecture, Chip Manufacturing, CUDA Alternatives, AI Economics

AI Summary

→ WHAT IT COVERS Cerebras CEO Andrew Feldman discusses the company's $1.1 billion Series G raise at $8.1 billion valuation, NVIDIA's competitive position, AI infrastructure bottlenecks, energy requirements for AI deployment, and the concentration of market value in seven technology companies. → KEY INSIGHTS - **Pre-IPO Capital Strategy:** Cerebras raised $1.1 billion from Fidelity and Tiger Global before going public to secure manufacturing capacity and data center expansion without IPO distraction. Getting Fidelity specifically signals Wall Street confidence and validates late-stage valuations for public market readiness. - **Chip Depreciation Reality:** Chip depreciation depends on performance improvement between generations, not arbitrary timelines. Current generation-over-generation gains deliver 2-2.5x actual performance when comparing apples-to-apples metrics like memory bandwidth, not just theoretical flops. System bottlenecks matter more than individual chip speed improvements for real-world applications. - **US Power Infrastructure Myth:** The US has sufficient power for AI expansion but in wrong locations. Abundant natural gas in West Texas and hydro in Upstate New York exist where people, buildings, and fiber optic infrastructure are absent. The challenge is geographic mismatch, not total capacity shortage. - **AI Talent Bottleneck:** Fundamental shortage of AI practitioners and data scientists limits industry growth more than hardware. Universities produce insufficient graduates while immigration policy restricts H-1B and J-1 visa pathways that historically brought top global talent. Companies must pay extraordinary compensation for irreplaceable expertise that no team size can replicate. - **Data Pipeline Investment Gap:** Unsexy infrastructure like data cleaning, tokenization, and pipeline management causes more AI project failures than actual AI technology. These roles receive minimal investment and recognition despite being critical success factors. Many billion-dollar AI initiatives fail on data preparation, not model performance. → NOTABLE MOMENT Feldman reveals that after 15 months burning $6-7 million monthly while unable to manufacture a single working wafer-scale chip, the founding team stood watching their first successful unit run for 30 minutes, having solved a 75-year problem that defeated IBM, Texas Instruments, and Gene Amdahl. 💼 SPONSORS [{"name": "Coda", "url": "https://coda.io/20vc"}, {"name": "Vanta", "url": "https://vanta.com/20vc"}] 🏷️ AI Infrastructure, Chip Architecture, Energy Requirements, Venture Capital, Talent Acquisition

Explore More

Frequently Asked Questions

What podcasts has Andrew Feldman appeared on?

Andrew Feldman has appeared on 4 podcasts we summarize, including 20VC (20 Minute VC), All-In with Chamath, Jason, Sacks & Friedberg, No Priors: Artificial Intelligence | Technology | Startups — 5 episodes in total. Every appearance is listed below with an AI-generated summary.

Does Andrew Feldman appear as a guest speaker on podcasts?

Yes. Andrew Feldman has been a guest on 4 shows we track, across 5 episodes. Browse each appearance below to read the key takeaways and listen to the original.

Where can I find summaries of Andrew Feldman's interviews?

Read AI-generated summaries of all 5 of Andrew Feldman's podcast appearances on SignalCast — each with key insights and a link to the full episode.

Never miss Andrew Feldman's insights

Subscribe to get AI-powered summaries of Andrew Feldman's podcast appearances delivered to your inbox weekly.

Start Free Today

No credit card required • Free tier available