Martin Shkreli on AI, Pharma, and What Actually Matters

What It Covers

Martin Shkreli joins the a16z podcast to analyze the AI model wars between OpenAI and Anthropic, the case for photonic computing as NVIDIA's long-term successor, why peptide biohacking is scientifically unsound, and where pharma entrepreneurs should focus to generate both impact and returns.

Key Questions Answered

• •OpenAI Revenue Gap: OpenAI's current enterprise revenue sits around $30B, but Shkreli estimates aggressive monetization matching Anthropic's pricing model could push that figure to roughly $200B. Anthropic charges 5–7x above stated seat prices, billing customers far beyond quoted rates, suggesting significant untapped pricing power exists across the AI platform market.
• •Photonic Computing Opportunity: Light performs matrix multiplications — the core operation in AI workloads — at no energy cost, making optical computing a potential 1,000x to 1,000,000x improvement over silicon in flops-per-watt. Only a handful of startups exist in the space versus thousands of AI agent companies, representing a structurally undercrowded $5–10T hardware opportunity.
• •Deep-Domain Software Moats: Vibe-coding cannot replicate enterprise software requiring 1,800 data relationships, validated bond pricing APIs, and six-month vendor contracts. Traders at firms like Citadel demand accuracy and accountability over convenience. Investors should treat the recent software stock selloff as a buying opportunity in verticals where domain expertise creates durable differentiation.
• •Pharma Value Creation Formula: The highest-return pharma opportunities remain rare diseases and severe cancers, not lifestyle drugs. A single approved therapy for conditions like Duchenne muscular dystrophy can command $1M per patient annually because it restores productivity and eliminates ongoing care costs. Neuralink exemplifies this model — solving paralysis generates insurance-reimbursable value at scale.
• •Peptide Biohacking Flaw: BPC-157, the flagship peptide in self-administration stacks, has a half-life measured in seconds after injection, leaving no pharmacological window to produce therapeutic effects. Drug half-life is a foundational requirement for efficacy. The trend persists because placebo response is real and regulations on unscheduled peptides remain minimal, not because the compounds work.