The a16z Show

Marc Andreessen × Ben Horowitz × Chris Dixon × Erik Torenberg

a16z is expanding the ACC Crypto Fund into a16z Frontier for “outer planets” beyond existing verticals as crypto’s boundaries blur with fintech and Wall Street.Marc Andreessen calls AI’s early-adopter gap with incumbents the widest of his career; employees adopting ahead of companies could fuel startup formation, while AI answers and Stack Overflow’s traffic drop point to an internet-economics reset.

Karan Singhal × Jay Rughani

GPT-6 Luna, OpenAI’s cheapest frontier API model, beat 5.6 Sol on HealthBench Professional at 34 times lower cost.Singhal sees a path to commoditized health intelligence, but says it has not yet affected the $5.3T US spend; savings require workflow integration.Epic, UCSF, and AdventHealth offer deployment evidence, including roughly 80% lower administrative workflow costs reported by AdventHealth.

Jad Naous × Martin Casado

AI coding has moved beyond autocomplete: Martin Casado built data-lake auto-promotion production code in 10 days versus roughly six months historically, a 10- to 20-fold gain Jad Naous says favors engineers already strong in systems.Grepr’s 90% telemetry-cost reduction offers a concrete commercial wedge, while anomaly-led diagnosis works internally and with a few customers; autonomous cluster management remains “not yet.”

Lisha Li × Nathan Lambert

Lambert estimates distillation may give Chinese labs a few months, but its value in RL-heavy post-training remains unclear.Closed labs integrate inference and lead coding, where revenue has gone from zero to $100 billion, while open-model economics fragment builders, inference firms and post-training companies.Open models may still win buyers seeking owned weights and the cheapest adequate model.

Martin Casado × Dhaval Adjodah × Owen He

RSI founders Dhaval Adjodah and Owen He report early signs of performance rising as resource use stays flat or falls, with the model itself as optimizee.They cite retraining GPT-2 in hours for about $100 on the same GPU and data, and target inference efficiency across the stack.Casado has not seen equivalent GPT-5 evidence, leaving scalability beyond fast-evaluation tasks unresolved.

Raghu Raghuram × Guido Appenzeller × Pat Gelsinger

AI is compressing logic design faster than silicon, packaging, and rack deployment, shifting value toward manufacturing, memory, power, cooling, and integration.With roughly 100 AI processor vendors, consolidation is likely as workloads evolve and winners require major capital, software, and customer backing, though agent-generated kernels may preserve some heterogeneity.Memory may be positioned for its first major reset in 30 years, while optical scale-up I/O around 2028–2029 and power shortages, including eight-year turbine lead times, define key catalysts and risks.

Raghu Raghuram × Matt Garman

AWS is underwriting durable AI demand with $220 billion of 2026 capex while reserving scarce accelerators for startups as well as frontier labs.Positive-ROI production workloads and single-digit maximum customer concentration temper demand risk, while AWS plans to buy 2 million NVIDIA GPUs over the next couple of years.Agentic workloads may strengthen AWS’s architectural edge, but power, memory, chips and enterprise evaluation remain constraints to monitor.

David George × Kevin Mandia

AI is shifting cyberattacks from resource-constrained “sniper shots” to scalable swarms, while Armadin continuously verifies exploitable network weaknesses.Since January 2026, Armadin has found more than 90 zero-days across Fortune 500 customer environments, while Blue aims to turn a five-minute lead into autonomous controls.With technical advantages potentially copied in six months, customer execution and the Red-to-Blue feedback loop remain the operating test.

Erik Torenberg × Alex Atallah × Amjad Masad

Stripe’s acquisition of OpenRouter preserves its brand, roadmap, and product autonomy while pairing payments with model-independent inference infrastructure.OpenRouter’s marketplace reduces vendor lock-in, combines models, and pressures inference prices downward as enterprises evaluate task-specific performance rather than feature lists.Specialized decision models could lower model debt and police agent actions, while multi-model fusion reportedly approaches frontier quality at 40–50% of the cost; deception risk and cache reuse remain unresolved.

Seema Amble × Elena Burger × Vladimir Keil

AI procurement startups are targeting the end-to-end work hidden behind an ERP’s “$8,000” result, where 30 meetings, 500 emails and supplier counteroffers determine cost and delivery.Lio’s path from human review to $10,000, $20,000, or $100,000 negotiations highlights the production bottleneck: 70% accuracy can leave humans doing 100% of the work, while complex multimillion-dollar cases still require specialists.