Latest Interviews
Showing 1–13 of 13 interview transcripts.
Clear all filters- Dwarkesh Patel1h 17m
Dylan Patel – Two labs will soon control most of the world's workforce
Projections indicate that global AI infrastructure spending will surpass $2 trillion by 2025 and reach $7 to $10 trillion annually by 2030, driven primarily by OpenAI and Anthropic which are expected to control up to 80% of incremental compute capacity. This aggressive capital accumulation is forcing hyperscalers to become major borrowers and pushing interest rates higher, which risks a broader economic crowding-out effect and potential sovereign debt crises in developing nations. Concurrently, a widening geopolitical divide is emerging as the US maintains a 70% dominance in deployment while China attempts a delayed domestic scaling effort, potentially creating a significant gap in effective AI capabilities by the decade's end.
- Y Combinator44 min
He Built the World's #1 Open-Source Coding Agent
Jay V, Gary Tan, Mark Mandelmann, Seth Vargoevic, Frank, Dax, Dylan Patel, Soyal, PG, Jay Poldinger
OpenCode, a sixteen-year-old platform founded by Jay Poldinger and Frank, has grown to 13 million monthly active users and processes 7 trillion daily tokens by serving as a neutral model harness that bypasses vendor lock-in. The company's strategy of supporting over 70 open-source models, which became critical after Anthropic restricted open usage, generated $160,000 in monthly subscription revenue and an estimated $33 million in inference revenue as of June. With significant adoption in China, Brazil, and the United States, OpenCode validates its product-market fit through enterprise organizations that proactively seek security agreements to integrate its agent loops into internal workflows.
- RAISE Summit19 min
Inside the AI Inference Cluster: Measuring What Matters | Mansour Karam, Aria Networks | RAISE 2026
The AI infrastructure sector is pivoting from homogeneous general-purpose systems to specialized architectures that optimize distinct model training and serving economics, with high-throughput inference offering potential cost savings of up to 100x. This shift demands deep networking solutions capable of managing dynamic, unpredictable traffic patterns and microsecond-scale latency requirements for complex agent workflows. Leading this transition, ARIA proposes a multi-layered AI agent architecture that integrates full-stack telemetry to dynamically coordinate hardware, software, and network operations across immediate physical reactions and long-term strategic planning.
- RAISE Summit18 min
Fast by Design: The Infrastructure Powering the Agentic Era | Rodrigo Liang | RAISE Summit 2026
Rodrigo Liang, Monti Saroya, Dylan Patel
SambaNova has secured the first close of a $1 billion funding round at an $11 billion valuation, establishing itself as JP Morgan's preferred inference provider for multi-trillion parameter models. The company differentiates its heterogeneous architecture by delivering up to 10x faster decode performance in full precision, avoiding the accuracy trade-offs of quantization while integrating with NVIDIA and Intel infrastructure. By leveraging a distribution network of 94 subsidiaries, SambaNova targets enterprise adoption through hybrid air-gapped and cloud solutions that prioritize data sovereignty and model choice over direct competition with general intelligence labs.
SemiAnalysis, Altimeter, Nebius, Glean.. 12 Hot Takes From Biggest Names in AI
Dylan Patel, Qasar Younis, Apoorv Agrawal, Arvind Jain, Ariel Cohen, CJ Desai, Gil Feig, Nikhil Benesch, Barak Kaufman, Max Junestrand, Marc Boroditsky, Laura Diorio, Kasser, Mark
RAISE Paris marked a decisive industry shift from speculative hype to enterprise-grade cost reconciliation, as buyers now demand clear ROI and physical AI adoption outpaces volatile large language model growth. Key figures including Applied Intuition's Kasser and analysts from Altimeter Research warned of an impending market bust driven by unsustainable spending, while companies like Navan and TurboPuffer demonstrated new economic models focused on profitability and reduced inference costs. The event concluded with a consensus that success requires resilient multi-model strategies, robust data layers, and a global expansion mindset to navigate rising hardware prices and geopolitical energy constraints.
- RAISE Summit18 min
Cheap Tokens, Expensive Mistakes: The Real Economics of AI at Scale | RAISE Summit 2026
Liran Zvibel, Dylan Patel, Karen Kwok
Major AI operators are pivoting from pure pricing power to structural cost efficiency, leveraging advanced KV cache offloading and optimized storage architectures to slash inference expenses. Recent benchmarks reveal that NAND-based memory solutions and strategic hardware combinations, such as high-capacity AMD GPUs, can outperform standard setups by up to 20 times in multi-turn agentic workflows while delivering 7x higher throughput. This shift allows emerging providers and frontier labs to significantly expand gross margins, with companies like Anthropic and OpenAI now projecting substantial operating profits driven by these infrastructure innovations rather than model weights alone.
- Sequoia Capital1h 10m
Why Hardware-Software Co-Design Is AI's Real 100x: Dylan Patel of SemiAnalysis
Dylan Patel, Shaun Maguire, Sonya Huang
Semi-Analysis, founded by Dylan Patel after a period of homelessness and a stint in quantitative trading, has grown to a 90-person team generating nearly $100 million in revenue by blending engineering expertise with hedge fund economics. The firm leverages Patel's annual attendance at over 40 global semiconductor conferences to gather proprietary supply chain data and launched InferenceX, a living benchmarking platform supported by over $50 million in compute donations from major tech firms. Looking ahead, the company projects critical shifts in the industry driven by power grid limitations, the obsolescence of single-architecture hardware, and a strategic pivot toward software-hardware co-design to navigate the escalating demands of artificial intelligence.
- Dwarkesh Patel2h 31m
Dylan Patel — The single biggest bottleneck to scaling AI compute
The Big Four hyperscalers have forecasted a combined $600 billion in capital expenditure, yet only about 20 gigawatts of incremental compute capacity is expected to come online in the US this year due to long-lead infrastructure projects. While OpenAI aggressively secured long-term capacity, Anthropic now faces a critical 4-gigawatt gap that forces reliance on expensive spot markets, highlighting a broader industry struggle against semiconductor supply bottlenecks and memory bandwidth constraints. Ultimately, EUV tool production limits and labor shortages constrain global AI scaling, positioning US allies with advanced manufacturing capabilities to maintain a significant lead over China for the foreseeable future.
- Dwarkesh Patel1h 29m
Satya Nadella – How Microsoft thinks about AGI
Satya Nadella, Dylan Patel, Dwarkesh
Microsoft is executing a massive infrastructure shift toward a 50-year horizon, highlighted by the 10x capacity boost of its Fairwater 2 data center and a move to support autonomous agents through tiered subscriptions and sovereign cloud compliance. CEO Satya Nadella warns against the "winner's curse" for pure model providers, instead positioning Microsoft to profit from a fungible fleet strategy and an "Agent HQ" ecosystem that orchestrates diverse AI tools across enterprises. With capital expenditures projected to triple to $500 billion globally, the company aims to balance massive hardware investments with software-driven efficiency to compress decades of economic growth into the next two decades.
- RAISE Summit18 min
Together AI & SemiAnalysis: In Conversation Together AI's Vision For the Future of AI Infrastructure
Vipul Vaid-Prakash, Dylan Patel
Founded in late 2022, Together AI challenges hyperscaler dominance by reengineering the cloud stack to achieve 75% maximum FLOP utilization on H100 GPUs, significantly outpacing the 50% industry standard. Led by chief scientist Tri Dao and the ClusterMax initiative, the firm drives down inference costs for massive models like DeepSeek from $8 to 30 cents per million tokens through architectural innovations and Reinforcement Learning frameworks. These technical advancements enable new MoE-based workloads and position the company to define emerging optimization surfaces before new hardware generations are fully adopted.
- a16z1h 39m
Dylan Patel on the AI Chip Race - NVIDIA, Intel & the US Government vs. China
Dylan Patel, Erik Torenberg, Sarah Wang, Guido Appenzeller
NVIDIA and Intel have established a transformative $5 billion strategic partnership to jointly develop custom data center and PC products, a move that significantly boosted NVIDIA's stock and acknowledged the industry's shift from CPU to GPU dominance. This alliance contrasts sharply with China's aggressive pursuit of semiconductor self-sufficiency, where Huawei navigates US export bans and high-bandwidth memory bottlenecks through domestic innovation and alleged smuggling channels. Simultaneously, hyperscalers like Oracle and AWS are capitalizing on surging demand with massive infrastructure deals and repurposed capacity, while the market faces technical complexities in deploying next-generation Blackwell architectures and optimizing hardware for specific inference workloads.
- a16z1h 6m
Dylan Patel on GPT-5’s Router Moment, GPUs vs TPUs, Monetization
Dylan Patel, Erin Price-Wright, Guido Appenzeller, Erik Torenberg
The event details OpenAI's strategic pivot in GPT-5 toward optimizing inference latency and cost rather than raw model size, introducing dynamic routing to monetize free users while shifting the industry focus to cost-performance Pareto frontiers. Competitors face significant hurdles displacing Nvidia due to its entrenched ecosystem, prompting hyperscalers to accelerate custom silicon development despite performance gaps and global power grid constraints driving infrastructure expansion strategies. Strategic recommendations urge major tech leaders to adopt usage-based pricing, restructure internal product execution, and invest heavily in physical infrastructure to secure long-term market dominance.
- Dwarkesh Patel2h 11m
@Asianometry & Dylan Patel — How the semiconductor industry actually works
Dylan Patel, Jon Y, Jane Street, Stripe, Xi, Liang Mong Song, China, Huawei, Taiwan, US, John Y
Semiconductor experts discuss the escalating geopolitical race where China's ability to rapidly build gigawatt-scale data centers and leverage domestic chip manufacturing could allow it to surpass Western AI capabilities by next year. The dialogue highlights critical bottlenecks in power infrastructure and supply chains, noting that while export controls have inadvertently spurred Chinese innovation in 7nm and 5nm processes, the US and its allies face significant grid limitations and capital requirements ranging from $50 billion to $100 billion to meet future cluster demands. Ultimately, the speakers analyze a market driven by a "Pascal's Wager" among tech CEOs who are betting massive capital on transformative models like GPT-5 to justify current debt-financed infrastructure despite looming risks such as a potential Taiwan crisis and delayed revenue generation.