
Annie Lamont
Key Views & Dialogues
Annie Lamont: $14B Managed, 70+ Exits, 15 IPOs, 7x Midas Investor
Healthcare AI has crossed O’Shea’s threshold for measurable impact, with biological and chemical models, laboratory robotics and simulations targeting drug development’s high failure rates. After a decade building licenses, networks and care infrastructure, Devoted Health tripled in size, halved operating expenses and significantly increased EBITDA, while China’s rising pharmaceutical research share leaves domestic research and regulatory capacity as risks to monitor.
View summary (7)
Molly O’Shea says AI has crossed from experimentation into measurable healthcare impact after a decade of watching it produce few additional drugs or therapeutics. “Everything has changed in the last two years,” led by biological and chemical models, laboratory robotics, and simulations that could raise success rates in the most expensive, failure-prone product-development industry.
Devoted Health is O’Shea’s strongest proof that AI compounds a hard-won operating moat rather than replacing one. The company spent a decade securing state licenses, provider networks, distribution and primary-care infrastructure; over the past year it tripled in size, halved operating expenses and significantly increased EBITDA. Her call is unusually categorical: it is “probably the best example of AI in healthcare in the world right now.”
Healthcare’s first AI dividend should be augmentation, with much of the value accruing invisibly to patients. Administrative work consumes 25-30% of healthcare costs, while O’Shea says roughly 30% of radiology images are read incorrectly; AI can reduce both burdens without eliminating doctors. Patients may never credit AI, but they should feel it when clinicians stop typing with their backs turned and miss fewer findings.
O’Shea frames the US-China life-sciences race as an existential industrial risk, not a routine competitive cycle. She says roughly 50% of pharmaceutical companies’ external research dollars now go to China, whereas this was not happening five years ago, while China may already match or exceed US university patent output. Her response combines AI investment with stronger university research and “a complete reorganization of our FDA and processes.”
AI has raised the required quality of founders faster than it has relaxed the need for engineers. O’Shea wants technically fluent, product-oriented entrepreneurs who can change course, recruit humbly and understand distribution; the best create conviction “within five minutes.” The career lesson she still carries is blunt: “Your bar for entrepreneurs is not high enough.”
Only about 10% of today’s AI valuation inflation is justified in O’Shea’s view. A handful of companies can address enormous markets spanning software and services, but many niche businesses valued at $10 billion or $20 billion may ultimately be worth only $1 billion or $2 billion and acquired on EBITDA. “Only 10% of the companies in the world are worth it, and 90% aren’t.”
Oak is pairing unusually flexible capital with company formation in markets where AI can execute, not merely analyze. Its funds grew from $500 million to $2 billion, checks range from $1 million to $100 million, and 20-40% of a fund can go to early-stage investments. Augur received $100 million at launch to build supply-chain agents, while Halluminate’s simulated financial environments went “from zero to a hundred in six months.”
We’re Reaching the Physical Limits of Chips
AI’s investable bottlenecks are shifting toward power, memory, cooling, and data transfer as Moore’s Law nears a physical plateau. Micron, SK hynix, and Samsung control 95% of memory, whose price rose 700% this year as hyperscalers absorbed supply. Photonics could relieve several constraints, with co-packaged optics expected beside GPUs within five years as memory pressure pushes specialized hardware beyond ever-larger models.
View summary (6)
- AI’s investable bottlenecks now center on power, memory, cooling, and data transfer. Adam identifies these constraints; Stephen says physical components are harder to improve overnight. Micron, SK hynix, and Samsung account for 95% of memory, whose price rose 700% this year as hyperscalers absorbed supply.
- Moore’s Law is reaching a physical plateau after decades of transistor scaling. Stephen puts the transistor’s dimension near 1 nanometer—roughly 100,000 times smaller than a human hair—forcing hardware solutions to become “significantly more inventive.”
- The apparent hardware renaissance reflects software’s current dependence on “brute force.” Adam says the software market is still immature: hardware is filling the pyramid’s base, while software that couples tightly to new architectures should create more value.
- Ever-larger models may be approaching their practical endpoint. Stephen cites open-source Kimi K3 as having more parameters than the human brain, while the chip needed to operate it is about a million times less energy-efficient; memory bottlenecks should push developers beyond the “scaling hypothesis.”
- Photonics is a leading candidate to relieve several infrastructure constraints simultaneously. Stephen flags optical-to-electrical conversion as a bottleneck, while Adam says light can improve power consumption, cooling, and data transfer. Optical links are expected to move from between clusters toward “co-packaged optics” beside GPUs within five years.
- Today’s LLM era may eventually resemble dial-up internet: transformative, but primitive in hindsight. Adam predicts that in 5–10 years “we will laugh at how old-fashioned AI was today.” He also favors hiring and investing in people who combine European academic and technical training with U.S. experience and business acumen.