For a decade, computational biology companies told a compelling story:
AI would discover drugs faster, cheaper, and more reliably than traditional biotech. The platform — the models, screening engines, and simulations — would be valuable on its own. Partnerships would fund the science. Risk would be lower. The upside would be higher.

It was neat, logical, and appealing.

And now, the market has delivered its verdict:
The standalone AI drug discovery platform is breaking down.

Two companies illustrate this better than anyone: Schrödinger and Recursion. They took opposite approaches. Both ran into the same underlying problem.

Schrödinger started as a profitable software company. Great margins, steady revenue, clean story. Then they decided to build their own drugs. If they could discover molecules with their platform, they could keep the value instead of licensing it away.

But running a real pipeline is expensive. By 2025, Schrödinger stopped pushing programs like SGR-1505 and SGR-3515 into trials. They cut roughly $70 million in R&D costs and retreated back to software.

Their model-first approach collided with three hard realities:

Schrödinger survived — but only by abandoning the part of the business with the most upside.

Recursion took the opposite approach.
If Schrödinger tried to be efficient, Recursion tried to be enormous.

In 2024, Recursion acquired Exscientia and became the largest AI-driven drug discovery company in the world. Their bet was simple:

But scale didn’t solve the core problem: biology is the bottleneck, not compute.

By early 2025, the company cut deep:

These weren’t strategic pivots. They were survival moves.
The platform did its job. The biology didn’t cooperate.
But the market punished the platform anyway.

Recursion’s runway shortened faster than its discoveries matured.

Schrödinger retreated.
Recursion expanded.
Different strategies — same underlying trap.

The platform trap looks like this:

The core issue is simple:
Platforms don’t matter unless they generate drugs that work, and whether drugs work depends on biology — not on how good your models are.

This is the uncomfortable truth the market is now pricing in.

The industry is reorganizing into two camps:

1. The toolmakers
Companies like Schrödinger that sell software, models, or services.
They accept slower growth, lower multiples, and fewer existential risks.

2. The drugmakers
Companies that own the programs and take the biological risk seriously.
They stop marketing their “platform” and focus on the pipeline.

There is no sustainable middle ground.

The promise of “AI-first biotech” was that models would transform drug development into an information problem. But drug development remains a biological problem — messy, nonlinear, stochastic, and expensive.

Platforms don’t get rewarded.
Drugs do.

The collapse of the platform narrative is not a failure of the science.
It’s a correction toward commercial reality.

Models are getting better.
Experiments are getting cheaper.
Biological datasets are getting deeper.

But until a drug works in a patient, the platform is just a cost center.

The next generation of computational biology companies will look less like software startups and more like integrated biological systems:

In other words:
The platform era is ending.
The translation era is beginning.