Is AI Actually Working in Biotech and Pharma? The Good, the Bad, and the Ugly
Is artificial intelligence (AI) in biotech and pharma just hype, or are we seeing true breakthroughs in drug discovery? Headlines tout billion-dollar deals, supercomputers for research, and high success rates for AI-designed molecules. But as some AI biotechs collapse, the risks and results are less clear.
You've seen the news: Lilly and Nvidia building a massive supercomputer for drug discovery, AI-discovered drugs moving through early trials at high success rates, and Big Pharma buying AI startups. But the other side of the coin has AI biotechs collapsing and shelving programs. The disconnect is striking. So what's really happening?
A new study on AI in pharma and biotech prompted me to break down what is going on in this space.
AI Spending Is Accelerating in Pharma
The AI pharmaceutical market is growing fast. It's worth $1.94 billion in 2025 and is projected to hit $16.49 billion by 2034. By the end of 2025, roughly 30% of new drugs will be discovered using AI in some way, compared to almost zero a decade ago.
The AI boom has created hype and companies are integrating AI into drug discovery. Biotech companies where AI is their core function show 75% AI integration for drug discovery. Traditional pharma companies lag behind but are catching up through partnerships and internal builds. Pfizer, AstraZeneca, J&J, Roche, and Novartis have all made AI central to their R&D strategy.
The value of the promise is straightforward: AI can cut drug discovery timelines from 5 years to 12-18 months and reduce early-stage costs by 40%. Since bringing a new drug to market costs around $1 billion and takes around 12 years, those efficiency gains matter.
Phase 1 and 2 success rates for AI-discovered drugs
AI-discovered molecules show strong Phase 1 success rates, around 80-90% compared to historical averages of 40-65% using traditional methods. This works because AI is good at predicting which molecules will be safe and have the right chemical properties before they're even made. Machine learning models trained on millions of compounds can flag potential toxicity and metabolism issues before synthesis.
Phase 2 is where things get tougher. AI-discovered molecules show Phase 2 success rates around 40%, similar to traditional approaches. This tells a crucial story: AI excels at designing molecules with the right properties but hasn't cracked efficacy prediction. Figuring out whether a drug will actually work in humans remains hard. Biology is messier than chemistry.
Sample sizes are still tiny. No AI-discovered drug has reached Phase 3 yet, but is expected by the end of 2025-2026. The next 12-24 months will show whether AI-designed drugs can actually clear Phase 2 and Phase 3.
Big Pharma is buying assets, not AI companies
Big Pharma's strategy remains the same: they're acquiring validated drugs, not AI platforms. Look at recent deals:
J&J paid $14.6B for Intra-Cellular Therapeutics, not an AI company, but a biotech with approved drugs
Merck spent $10B on Verona, $9.2B on Cidara, both traditional biotechs with clinical-stage programs
Novartis paid $12B for Avidity Biosciences, an RNA company, not AI-focused
Eli Lilly partnered with Nvidia to build computational infrastructure rather than buying an AI biotech
What they're actually doing: Big Pharma is building AI capabilities in-house through partnerships (Pfizer with Tempus, AstraZeneca with BenevolentAI, Lilly with Insilico) and infrastructure investments, then acquiring companies with proven late-stage molecules. This strategy makes sense because they get the validated science without as much biotech risk.
Who’s Winning and Losing with AI in Biotech?
Winners:
Big Pharma with AI infrastructure: Companies like Roche and Pfizer building AI capabilities while acquiring proven molecules.
AI companies with Big Pharma partnerships: Tempus (Pfizer), Insilico (Lilly partnership), companies that license tools to pharma.
Service companies: CROs and data providers selling AI tools and analytics to pharma without taking drug development risk.
Losers:
AI-native biotechs without partners: Recursion, most others that raised on hype but couldn't deliver commercial products fast enough.
Biotech companies ignoring AI: Traditional discovery shops are falling behind on speed and efficiency. They can succeed but they lack the new edge of AI drug discovery.
Anyone claiming AI will "solve" drug discovery: Phase 2 failure rates are proving otherwise, for now at least.
Why are AI biotech startups struggling?
Despite promising Phase 1 data, AI-focused biotechs are struggling. Some examples:
Recursion Pharmaceuticals: Shelved three AI-discovered drugs in 2025 during cost-cutting. Despite a Nvidia partnership, none of its AI compounds have reached market.
BenevolentAI: Delisted and merged with another company in 2025 after years of raised capital but no commercial products.
Exscientia: Merged with Recursion after struggling independently.
The funding environment dried up. Venture capital poured $18 billion into 200+ AI biotechs, but investor confidence is wavering. When interest rates are high, money flows away from speculative startups. AI adoption in large companies peaked in early 2025 and has already declined to 12%, reflecting broader disappointment with the technology.
The real issue isn't the AI, but the biotech fundamentals. Cost-cutting by Big Pharma, drug pricing uncertainty, and the fact that drugs still take 10-15 years to develop all squeeze AI biotechs. They burn cash fast but can't deliver drugs fast enough.
AI’s real constraints: Regulation and data quality
Regulatory bodies are still figuring this out. The FDA streamlined AI-driven approvals, but the "black box" problem remains. Many AI models can't explain their decisions. Recent gene therapy failures (Sarepta's fatal liver injuries, Intellia's clinical hold) have intensified scrutiny of novel approaches, including AI-discovered drugs.
Data quality is the fundamental problem. Healthcare data is sensitive, fragmented, and often biased. If AI algorithms train on unrepresentative datasets, treatments might work for some patient populations but not others.
The Larger Picture
People see hype in AI and a life sciences revolution. But the reality is not quite that. With AI, Big Pharma is looking to reduce risk, time, and cost of new asset development. Smaller biotechs are aiming for big moonshot tech that lets them get a foothold in a crowded market with dwindling capital.
AI isn't replacing drug discovery. It is speeding up specific steps. Target identification, molecule optimization, patient recruitment, and manufacturing are where AI delivers clear value. AI hasn't cracked the hard part: predicting which molecules will actually work in sick humans.
The bottleneck is simple biology. The 10-15 year drug development timeline exists because safety and efficacy validation takes time in humans. AI can compress early discovery from 5 years to 18 months, but it can't eliminate Phase 2 and Phase 3 trials. Anyone selling AI as a way to skip clinical trials is selling blatant hype.
The playbook remains the same
Big Pharma's playbook remains the same, the tools change. It is not just AI. When R&D became risky, they switched to big M&A to acquire relevant and close to market assets. When Chinese and Asian innovation results in fast R&D and cheap candidates, they buy them or license them. AI is just another tool. The big players will partner with smaller, more innovative biotechs, or straight up buy them when they need to.
It's a proven model with sure economic returns. Build or buy AI capabilities to improve pipeline efficiency, acquire validated late-stage molecules regardless of how they were discovered, and let venture capital fund the risky AI biotech experiments. When AI biotechs succeed, acquire them. When they fail, avoid the losses.
But AI can still evolve. More Phase 2 data will emerge over the next 12 months. If AI-discovered molecules match historical Phase 2 success rates (around 40%), the investment thesis weakens. If they exceed 50-60%, AI becomes genuinely transformational. If they underperform, the sector likely contracts further.
AI is a real tool with real benefits in specific applications. But it's not magic. The companies that survive will integrate AI into broader R&D strategies, not bet everything on AI solving biology. We'll keep watching Phase 2 and Phase 3 data, and see what happens. That's where the hype will make or break AI.
