By Matthew Tuttle, Tuttle Capital Management 

 

A generation of start-ups have failed to live up to the hype. Executives are now betting that more powerful tools will crack the complexities of human biology

From The Financial Times

 

The Big Picture

Jensen Huang (Nvidia) and Marc Andreessen have both said the biggest long-term impact of AI will be in healthcare. The logic is obvious:

Drug discovery costs ~$2B per drug, with >90% clinical failure rates.Aging populations mean medical costs are ballooning globally.The process is data-heavy, slow, and expensive — exactly the kind of problem AI should disrupt.

But the FT article shows the other side: AI hasn’t delivered yet. First-generation AI drug discovery companies like BenevolentAI, Exscientia, and Recursion have struggled to move compounds beyond early trials. Many “AI-discovered” drugs have flopped. A decade in, not a single AI-born drug has been FDA-approved.

Why? Because drug discovery is where “bits meet atoms.” Biology is messier than chess or text prediction — incomplete datasets, noise, and fundamental gaps in knowledge about how the human body works.

 

Why the Next Wave Could Be Different

Two watershed advances reset the clock:

AlphaFold2 (2021): Protein-folding breakthrough from DeepMind that showed algorithms can generalize across biology.Generative AI (post-2022): Tools that can design novel molecules instead of just pattern-matching existing ones.

Pair that with:

Explosion of new biological data (labs like Recursion and Insitro building “AI science factories”).Compute scale from Big Tech (Alphabet’s Isomorphic Labs, Microsoft partnerships).Better capital discipline — companies focusing on hard targets (cancer, immunology) instead of “me too” drugs.

The promise is still intact: once AI produces a single blockbuster approval, the floodgates open.

 

Winners (ratings 1–10)Big Tech with money + computeAlphabet / Isomorphic Labs (GOOGL) — 8.5/10 Backed by DeepMind, aiming to build a generalizable drug discovery engine. Has compute, cash, and patience.Microsoft (MSFT) — 8/10 Aggressive in healthcare AI partnerships (e.g., Nebius deal, Nuance acquisition). Offers Azure compute + enterprise trust.Nvidia (NVDA) — 9/10 Picks-and-shovels winner: every lab and pharma company training models for drug design needs GPU/accelerator clusters.Data-rich next-gen biotechsRecursion (RXRX) — 7.5/10 Building enormous proprietary cell image datasets; early-stage drugs but long road to approvals.Insitro (private) — 7.5/10 “Cell factories” and massive data creation; led by Daphne Koller, but still pre-proof.Relay Therapeutics (RLAY) — 7.5/10 Targeting hard diseases; differentiated vs. “me too” approaches, but binary drug risk remains.Tools & infrastructureThermo Fisher (TMO), Danaher (DHR) — 8/10 Picks-and-shovels in lab automation, data collection, sequencing. If AI labs scale, they sell the gear.ASML (ASML), NVIDIA (NVDA), AMD (AMD) — 8–9/10 Compute & silicon backbone of healthcare AI.Losers / HeadwindsFirst-gen AI drug discovery start-ups: BenevolentAI, Exscientia — overpromised, underdelivered. Many merged, de-listed, or retrenched.Investors chasing “AI-me too” drugs: Compounds without differentiation are value traps.Small AI biotechs without cash or proprietary data: They’ll be squeezed out by Big Tech with compute + balance sheets.Investor TakeawayShort term (1–3 years): Still a hype-to-disappointment cycle. Binary risk (trial failures) + long timelines = landmines for pure-play biotechs.Medium term (5–10 years): Healthcare is likely the biggest eventual AI prize — one real FDA-approved drug from AI will reset the sector.Best trade today: Own the picks-and-shovels (NVDA, TMO, DHR, MSFT, GOOGL). Keep speculative exposure small in data-rich biotechs like RXRX and RLAY.

AI hasn’t yet cracked drug discovery — but the combination of AlphaFold, generative AI, and Big Tech’s compute makes healthcare the biggest eventual prize. The near-term trade is picks-and-shovels (NVDA, TMO, MSFT, GOOGL). The long-term asymmetry comes when the first AI-born blockbuster drug hits the FDA.

 

AI in Healthcare: Scenario Map for InvestorsBase Case (Most Likely – 5–10 Years)

AI accelerates drug discovery efficiency (target ID, molecule design, preclinical work), but timelines remain long and failure rates high. FDA approvals lag.

Winners:

NVDA (9/10): GPU backbone for all healthcare AI training.MSFT, GOOGL (8–8.5/10): Enterprise trust + compute scale + partnerships.Thermo Fisher (TMO), Danaher (DHR) (8/10): Lab automation, sequencing — steady demand from AI-driven research.

Speculative:

RXRX (7.5/10): Largest proprietary cell-image dataset; promising but binary risk.TEM (7/10): Building a “clinical data moat” from oncology and diagnostics. Could monetize as platform for AI models. Still pre-proof.

Losers:

First-gen AI biotechs that overpromised and underdelivered (BenevolentAI, Exscientia).Bull Case (Best Case – 3–5 Years)

Generative AI + AlphaFold-level breakthroughs → first FDA-approved “AI-born” drug by late decade. Investors re-rate healthcare AI as the next gold rush.

Winners:

TEM (8.5/10): If it proves its clinical dataset leads to new oncology diagnostics or drugs, re-rate could be massive.RXRX (8.5/10): Early proof-of-concept + partnerships could make it the category-defining public biotech.GOOGL / Isomorphic Labs (9/10): DeepMind + compute = only player with patience and resources to build a true “drug discovery engine.”

Losers:

Traditional CROs/outsourcers (IQV, CRL) could lose pricing power if AI replaces brute-force trial design.Bear Case (Worst Case – 5+ Years)

Human biology remains too complex; data too noisy. AI helps on the margins (molecule screening) but not at scale. Drug failures continue at ~90%. Funding dries up.

Winners:

Picks-and-shovels still win (NVDA, TMO, DHR): labs still need GPUs and sequencing tools.

Losers:

RXRX (5/10): burns cash without a late-stage pipeline; risk of dilution or consolidation.TEM (5.5/10): clinical data platform doesn’t translate to breakthrough therapies; may pivot to niche diagnostics.Small AI biotechs without data moats or cash → wiped out.

 

Investor TakeawaysCore exposure: NVDA, MSFT, GOOGL, TMO/DHR — structural winners regardless of drug discovery outcomes.Speculative basket: TEM and RXRX for asymmetric upside — but size carefully, binary risk is real.Watch for catalyst: The first FDA-approved AI-designed drug would be the “ChatGPT moment” for healthcare AI, driving an overnight re-rate.

 

The base case is that AI makes drug discovery more efficient but approvals remain elusive for years. The bull case is asymmetric — if TEM or RXRX deliver even one FDA approval, the sector re-rates overnight. Until then, NVDA, GOOGL, MSFT, and the lab picks-and-shovels are the safest way to play AI in healthcare.

 

Taken from today’s Daily HEAT. Gratis subscription: https://theheatformula.beehiiv.com/subscribe

 

 

The post WHY IS AI STRUGGLING TO DISCOVER NEW DRUGS? appeared first on Sick Economics.