Sally Epstein looks at why the world’s biggest AI companies are all pivoting towards biopharmaceuticals. Epstein, an entrepreneur in residence at Cambridge Innovation Capital, argues that drug discovery – if done right – could be where the AI industry has the biggest societal impact (and can therefore justify companies’ enormous current valuations). How these new market entrants will interact with (or replace) traditional pharma, however, remains unclear, as does how they plan to solve the clinical trial bottleneck.
In 2026, the teams at the leading frontier AI labs, Anthropic, OpenAI, and Google DeepMind, all reached the same conclusion: they must do science.
While Silicon Valley traditionally excels at software but not at clinical trials, the commercial reality driving this pivot is that science unlocks entirely new commercial opportunities that a focus on software alone cannot reach.
To be clear, while this logic applies equally to novel materials, catalysts, and battery chemistry, the vast majority of current capital flow is concentrated in biology. This is where blockbuster deals are being struck, and where the economic thesis of frontier AI will ultimately be proven.
The value ceiling problem
Most enterprise LLM deployments today, such as email drafting or data entry automation, suffer from an ROI ceiling. When an AI replaces or augments a human workflow, its perceived economic value is capped by the historical cost of that human labour or the legacy software that it displaces. Replace a call centre seat, and you save a salary, payroll taxes, and onboarding overhead. Every corporate business case currently justifying AI investment relies on this defensive, cost-cutting pricing model.
However, the frontier labs are sitting on eye-watering valuations that do not make sense under the labour-replacement paradigm we’re seeing. Anthropic’s and OpenAI’s multibillion-dollar fundraising and astronomical valuations will soon face a reckoning if AI remains confined to squeezing nominal efficiencies from corporate budgets.
The venture model only works if these companies capture a market that isn’t bound by “what a human used to cost”. Scientific discovery is one of the few markets large enough to shoulder the weight of the valuations.
A new drug molecule creates entirely new economic value rather than rebottling existing value at a lower price point. Crucially, they unlock pricing models, such as licensing agreements or IP-based revenue sharing, that scale not with API tokens, but with the multi-billion-dollar societal value of the discoveries themselves.
Market archetypes
Throughout 2026, the major frontier labs have each carved out their own unique layers of the emerging biotech stack.
Anthropic completed a $400 million acquisition of Coefficient Bio in April, bringing in elite talent from Genentech’s Prescient Design Unit to build biology-specific foundation models from the fundamental architecture level up, integrated into Claude’s system.
Google DeepMind’s dedicated life sciences arm, Isomorphic Labs, remains the most pertinent player in the space. Building on the momentum of Demis Hassabis and John Jumper sharing the 2024 Nobel Prize in Chemistry, Isomorphic secured a massive $2.1 billion investment and over $3 billion in joint development deals signed with Eli Lilly and Novartis. Isomorphic is operating as a peer to Big Pharma, boosting its own discovery pipeline.
In January, NVIDIA and Eli Lilly announced a $1 billion co-innovation initiative. The partnership focuses on linking agentic “wet labs” (automated physical labs) with computational models in the world’s first industrial-scale, AI-native facility.
OpenAI responded to the competitive pressure by launching GPT-Rosalind, a domain-specific reasoning model engineered for molecular biology, positioning itself as the premier infrastructure and tooling provider to existing biotech ecosystems.
The unresolved bottleneck
Despite the massive influx of capital, it’s important to remain realistic about what AI can, but also cannot, fix in drug development and discovery.
Historically, molecule discovery has never been the bottleneck. Instead, the downstream clinical trial pipeline required to bring a therapeutic asset to market consumes the majority of the time and capital. According to research, 90% of clinical-stage drugs fail during human trials.
AI has yet to prove it can genuinely mitigate these late-stage failure rates. Currently, generative AI-designed molecules in drug trials remain largely stuck in early-stage safety studies. Until companies like Isomorphic Labs steer an AI-designed compound successfully through Phase 3 (human) trials, the broader investment thesis remains unproven.
What will companies look like?
If this whole-sector pivot lands successfully, four distinct corporate structures will emerge.
The first is that they become AI tools providers to pharma. Labs license models like GPT-Rosalind or AlphaFold to pharma companies, taking a straightforward software margin.
Or, they become the discovery layer themselves. The lab designs the molecule, partners with a pharma giant for clinical development, and retains significant royalty exposure. This transforms AI labs into intellectual property powerhouses. We’re already seeing this play out with Isomorphic Labs’s work with Eli Lilly and Novartis.
Companies might also vertically integrate. For example, Labs build out their own automated wet labs, own their proprietary datasets end-to-end, and bypass traditional pharma only until regulations force them to do so.
The fourth is that companies will look at the entire biotech stack and identify that the true opportunity lies in building target identification, hit discovery, lead optimisation, and ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity), and pulling it together into a single service that a pharma company can buy. These companies will become suppliers to pharma companies rather than directly competing with them.
Looking to the future
The frontier AI labs have relentlessly searched for a market capable of scaling beyond the boundaries of human white-collar labour costs, and they have found it in deep science.
The deal velocity, capital volume, and coordination witnessed across the industry over the last six months reflect a fundamental departure from the speculative “AI for drug discovery” hype cycles of the past decade.
The defining question now is how much of the traditional pharmaceutical value chain these AI labs will try to conquer.
