Ending Big Pharma’s $2.6 Billion Coin Flip

When a Hungarian physician named Ignaz Semmelweis noticed something peculiar in the maternity ward  he was running at Vienna General Hospital, it kept him awake for weeks. 

Women were dying… but not from childbirth. 

They were dying because of doctors.

Of course, this was back in 1847, when maternity wards were populated by medical students delivering babies after performing autopsies. 

Back then, the death rate from childbed fever ran above 10%. Yet, in the ward staffed by midwives who never touched a cadaver, that mortality rate was under 2%. 

Semmelweis just couldn’t explain the gap, until a colleague named Jakob Kolletschka died after being accidentally nicked by a student’s scalpel during an autopsy and developing a severe blood infection.   

However, the autopsy results looked identical to the women that were dying on the maternity ward.

That’s when Semmelweis made the critical connection. 

Doctors were going directly from dissecting corpses to delivering babies, carrying what he called “cadaverous particles” on their unwashed hands. 

So, Ignaz implemented mandatory washing with a chlorinated lime solution. Within three months, the death rate on his ward dropped from 1-in-10 to 1-in-100.

What came next? Well, he started publishing the results, going on the lecture circuit, and sent letters to the most prominent obstetricians in Europe.

Despite the mathematical and clinical evidence, his contemporaries rejected his ideas. They scoffed at the notion that their hands could spread disease.

Tragically, his colleagues committed him to an asylum a few years later in 1865. The story that played out was actually far more nightmarish than that, given that he was tricked into visiting the institution. 

After realizing his situation, he tried to leave and was taken by guards, and died 14 days later from a gangrenous wound. 

In fact, his discovery wasn’t formally accepted by the medical profession for nearly 40 years after his death.

Today, we have a term for this kind of cognitive bias: the Semmelweis reflex — the automatic rejection of new knowledge because it contradicts what powerful institutions already believe.

It’s the oldest story in the history of medicine. 

And you know what? 

It’s happening again right now in the glass towers of Big Pharma. See, the evidence for a fundamentally better model of drug discovery is piling up, yet the industry stares at it the same way Semmelweis’s colleagues stared at his mortality charts.

The $2.6 Billion Coin Flip

Breaking down the traditional model of drug discovery is downright frightening for Big Pharma. 

Remember, bringing a single new drug to market today costs an average of $2.6 billion and takes between 10 and 15 years!

Those numbers come from peer-reviewed research across hundreds of drug development programs. 

But here’s the part that never makes the headlines… 

After all that time and money, roughly 9 out of 10 candidates still fail in clinical trials.

Yet, the leading cause of failure isn’t bad chemistry. 

It’s lack of efficacy — the drug simply doesn’t do what researchers hoped it would do in the human body, after working beautifully in cell cultures and animal models. 

Truth is, efficacy failures account for up to half of all clinical trial collapses, with unmanageable toxicity accounting for another 30%.

Go ahead and do the math with me…

Look, if the failure rate is 90% AND it costs roughly the same to run each candidate through the full development pipeline, then the true cost of every drug that makes it to your pharmacy shelf isn’t $2.6 billion.

Once you account for the graveyard of failed candidates behind every approval, it’s actually closer to $26 billion.

Is it me, or does that seem more like a chemistry roulette and less than a business model?

Still, Big Pharma has been running the same routine for 70 years.

And here’s why the failure rate never improved…

The traditional discovery process works backwards, if you think about it. 

First, researchers identify a disease, then screen hundreds of thousands of compounds hoping one binds to a relevant target, and run those survivors through years of animal studies. 

Up next come Phase II or Phase III trials, where they discover  — after hundreds of millions of dollars in sunk cost — that the biology doesn’t translate to humans the way it translated to mice.

The science wasn’t necessarily wrong, mind you. 

Rather, it was the method that was lacking. 

You can’t design a drug compound to fit a target you can’t see. 

And for most of the history of modern pharmacology, researchers couldn’t see the three-dimensional structure of the proteins they were trying to drug. 

Folks, they’ve been trying to pick a lock in the dark for nearly a century — and that dynamic is finally changing. 

AlphaFold was Just the Beginning

Back in 2021, DeepMind’s AlphaFold program solved a 50-year-old grand challenge in biology: protein structure prediction. 

For the first time, researchers could computationally model the three-dimensional folded shape of virtually any protein — the shape that determines how a drug molecule binds to it, blocks it, or activates it.

The lights came on.

But that wasn’t the only shift. 

Generative AI is now designing novel drug molecules from scratch — not screening libraries of existing compounds, but building new ones optimized from first principles for binding affinity, selectivity, and the ADMET properties (absorption, distribution, metabolism, excretion, toxicity) that kill 90% of candidates before they ever reach a patient.

Then it’ll run that simulation about a billion times so they come up with candidates that don’t fail, and then move on to the lab. 

Now we’re seeing those results starting to show up in the clinical data.

The difference is simply mind-blowing, too. 

AI-discovered drug candidates are hitting 80% to 90% success rates in Phase I clinical trials — the first human safety tests. 

Compare that to the historical average for the traditional routine, which is barely over 50%.

That gap isn’t merely incremental improvement, it’s an entirely new game being played on the same field. 

And that gap is exactly what Semmelweis showed his colleagues in 1848 — statistical proof so lopsided it should’ve ended the argument immediately.

But remember, his colleagues didn’t dispute the numbers, they rejected the implication.

Big Pharma has been doing the same thing for the better part of a decade, and they’re finally starting to recognize the potential. 

The Turning Point for AI-Drug Discovery

This is where it gets interesting for us. 

You see, there are 173 AI-originated drug programs right now in active clinical development. 

Roughly 94 of them are in Phase I, 56 are in Phase II, and 15 are in Phase III development — the pivotal trials that determine whether a drug gets approved or quietly shelved.

So far, none have reached the coveted green light by the FDA… not yet. 

I think that’s going to change very soon. 

Some analysts put the probability of the first FDA approval of an AI-discovered drug at roughly 60% within the next 24 months.

Right now, the most advanced program belongs to Insilico Medicine, whose lead compound rentosertib (INS018_055) is pursuing a pivotal Phase III trial for idiopathic pulmonary fibrosis — a progressive, fatal lung disease with limited treatment options. 

What makes rentosertib historically significant isn’t just its clinical position — it’s the first drug in the history of medicine in which both the disease target and the molecular compound were identified entirely by generative AI, with no human hypothesis guiding either step. 

Think about that… 

A computer chose what to attack, then designed the weapon. 

Now the candidate is in Phase III trials. 

We’ve hit a new era of drug discovery. 

The regulatory runway is being paved in real time, too…

Some of you might recall that the FDA launched its CDER AI Pilot Program to partner directly with companies on AI-assisted chemistry and clinical data analysis. 

Last December, the FDA qualified its first AI tool for use within clinical trials — the first formal regulatory recognition that AI-generated evidence is valid in the drug approval process. 

So far, there’ve been two ways to play this.

Either you find the pure-play clinical validation — companies with AI-discovered drugs already deep in trials, where a Phase III readout is the catalyst, or you go the platform route — the companies building the computational infrastructure that every drug maker will eventually need to license or acquire.

History says the platform layer captures the most durable value in a technological transition. 

After all, the railroad builders outlasted the gold rush miners every time.

But here’s the thing — there’s a third category of company I’ve been watching closely that doesn’t fit neatly into either bucket.

You see, it isn’t discovering drugs through target-based methods at all. 

Instead of asking what compound fits this protein, it’s asking a fundamentally different question: what does a diseased cell look like — and what would change that picture?

In fact, this small biotech player has imaged more than 50 trillion cells, and mapped the visual signature of hundreds of diseases at a resolution no human researcher could ever process. 

Of course, it also built an AI that finds patterns in its data which is invisible to any microscope — patterns that point directly toward the compounds that work, before a single clinical trial begins.

I suggest you take just a moment out of your day, and let me show you the most compelling AI opportunity most investors haven’t properly sized yet.