AstraZeneca says AI is raising its odds where most drugs fail
The most expensive product in modern business is the one that never ships.
Nowhere is that more true than in a pharmaceutical lab. A company can spend more than a decade and a fortune chasing a single molecule, only to watch it collapse in a late-stage trial that confirms what everyone hoped was not true.
You pay for that even if you never take the drug. The cost shows up in the price of the medicine that does work, in the value of the pharma stocks sitting in your retirement account, and in the treatments that never reach your family because the money ran out before the science did.
For years the industry treated that failure rate as a fixed law of nature. The price of doing business. Something to be managed quietly, never actually solved.
That resignation is starting to crack. The companies spending the most on research are now claiming they can bend the odds that have defined the business for decades.
One of the largest drugmakers in the world is making that case out loud. AstraZeneca (AZN) Chief Executive Pascal Soriot used a June 5 appearance on CNBC's "Mad Money" to argue that artificial intelligence is already improving the company's odds, and that it can attack the exact point where most drugs die.
Why most new drugs never make it to your medicine cabinet
The numbers behind drug development are brutal, and they are why Soriot's claim is worth a closer look.
Bringing one approved medicine from discovery to market costs roughly $2.6 billion, according to the Tufts Center for the Study of Drug Development. The process routinely runs 10 to 15 years. And nine out of every 10 candidates that reach human trials never make it, according to a peer-reviewed analysis in the journal Acta Pharmaceutica Sinica B.
Soriot put the per-trial stakes in plain terms. A single trial can cost $300 million to $500 million, so even a small lift in the probability of success delivers an enormous productivity gain, he told CNBC.
When I look at where AI money has actually gone over the past two years, most of it chased compute, data centers, and chatbots. This is a different kind of bet. It targets a failure rate that has shrugged off everything the industry has thrown at it.
Here is what that failure math looks like, and why a few percentage points are worth billions.
- About $2.6 billion is spent to bring a single approved drug to market, according to the Tufts Center for the Study of Drug Development.
- The path from lab to pharmacy takes 10 to 15 years, according to that same body of research.
- Roughly 90% of drug candidates that enter clinical trials never win approval, according to Acta Pharmaceutica Sinica B.
- One trial alone can run $300 million to $500 million, according to Soriot's comments on CNBC.
What AstraZeneca's AI is actually trying to predict
Soriot's pitch is not vague futurism. AstraZeneca has built an AI agent that combines clinical and laboratory data to estimate the probability that a Phase 3 trial will succeed, supported by partnerships with Tempus AI (TEM) and Pathos, according to CNBC.
Phase 3 is the last and most expensive stage of testing. It is where a drug has already swallowed years of funding, and where a late collapse destroys the most value. Pointing a prediction engine at that stage is a direct strike at the costliest failure in the business.
Related: Doctor GPT? AI Gets Healthcare Questions Right 76% of the Time
"The value of AI in our industry is productivity improvement," Soriot said on CNBC. He described designing medicines faster, finding new targets, and stripping likely side effects out of a molecule before it ever reaches a patient.
The skepticism is fair. Investors have spent 2026 demanding evidence that AI spending turns into durable returns rather than louder headlines. Morgan Stanley flagged this tension when it warned of AI spending-bubble risk, covered in my TheStreet report. A pharma CEO putting a number on it is a rare, falsifiable claim.
Where AI in health care is already paying for itself
A prediction model is only as good as the data feeding it, and that data is born somewhere very specific. It starts in the exam room, the therapy session, and the specialist visit, long before any model sees it.
That is the part of the story trial data cannot tell. Gal Steinberg, chief executive and co-founder of Twofold Health, an AI scribe that turns clinician visits into structured notes, says the foundation gets built far earlier than most investors assume.
The point of care, he argues, is "where healthcare data is either made usable or made noisy," said Steinberg. If the original documentation is inconsistent or buried in free text, he said, every downstream model ends up predicting the future from a distorted version of the past.
In my reporting on AI rollouts, the projects that survive scrutiny tend to share one trait. They fix a problem someone can already measure in dollars and hours, which is why documentation tools are showing returns now while flashier bets still sit years out.
It is also the quiet risk inside Soriot's plan. A Phase 3 prediction agent inherits whatever flaws live in the records it learns from, which is why the unglamorous job of capturing a patient visit cleanly sits upstream of every headline model.
What the AI drug bet means for your portfolio
Soriot did not promise to erase the failure rate. He promised better odds, and in a business where a few points of probability are worth billions, that is the whole game.
He also did something most AI boosters avoid. He attached a number to the promise, and numbers can be checked.
Wall Street is leaning toward him for now. Bank of America analyst Sachin Jain reiterated a buy rating on AstraZeneca with a 16,500 pence (£165) price target on June 9, according to TipRanks, and JPMorgan has kept an overweight call on the stock.
For investors, the signal to watch is not the next splashy AI demo. It is whether AstraZeneca's trial success rate actually moves over the next several reporting cycles, and whether rivals can match it.
If those odds shift even slightly, the payoff is easy to feel. It lands in the drugs that reach your family sooner, and in the pharma names that stop burning billions on bets they could have called years earlier.
Related: Microsoft and Copilot just hit a jackpot in healthcare
The Arena Media Brands, LLC THESTREET is a registered trademark of TheStreet, Inc.
This story was originally published June 15, 2026 at 2:37 PM.