Another traditional industry has officially announced that it has "fallen in love with AI".
On March 29 local time, US pharmaceutical giant Eli Lilly and Hong Kong-listed AI pharmaceutical company Insilico Medicine announced a strategic collaboration—an upfront payment of US$115 million, with subsequent milestone payments bringing the potential total value to US$2.75 billion , plus future sales tiered royalties.
This number silenced the entire industry for a second—has the "GPT moment" for AI-driven drug development arrived?
01
AI-driven drug development: From "story" to "real money"
Before today, AI-driven drug discovery was more like a story that was repeatedly told but never truly resolved. Startups raised funds, major companies entered the field, and academia endorsed it… but every time someone asked, “Have you actually gotten any drugs that patients can actually use?” the answer was always ambiguous.
Eli Lilly's payment represents a clear shift in stance.
In fact, Eli Lilly itself is also accelerating its "AI transformation".
At the JP Morgan Healthcare Conference in March, Eli Lilly announced a $1 billion joint innovation AI lab with NVIDIA, dedicated to addressing long-standing challenges in the pharmaceutical industry. That same month, NVIDIA also partnered with Novo Nordisk to accelerate drug discovery using the Gefion sovereign AI supercomputer.
That's not all. In February, Takeda Pharmaceutical signed a collaboration worth over $1.7 billion with AI company Iambic Therapeutics, aiming to use AI to find new drugs for cancer and other diseases. On March 27, Quotient Sciences and Intrepid Labs just announced a multi-year collaboration to introduce the machine learning model ANDROMEDA into early-stage drug development.
This is not an isolated case; it's a collective gamble. At the beginning of 2026, the pharmaceutical industry saw a flurry of major AI platform deals, with Eli Lilly, Sanofi, Novo Nordisk, Bayer, and almost every top pharmaceutical company rushing to sign contracts .
Industry analysts' figures are even more telling: the AI drug discovery market will be worth approximately $2.9 billion in 2025, and is projected to reach $5.1 billion in 2026, exceeding $13.4 billion by 2035 .
However, the influx of hot money does not mean the problems have disappeared.
02
How can AI be used to design "new molecules"?
Insilico Medicine is not a new company.
Founded by Chinese scientists and headquartered in Hong Kong, this company is one of the very few that has truly advanced AI-generated drug molecules to the clinical stage. Its core technology is to directly "design" entirely new molecular structures using generative AI, rather than simply screening existing compound libraries —a fundamental difference in its technological approach.
Traditional drug discovery typically follows this process: identifying a target → repeatedly screening millions of known compounds → finding candidate molecules → lengthy optimization. This process takes an average of more than ten years and costs billions of dollars.
InSilicon's approach is somewhat like "drawing a key from scratch"—directly telling the AI "what this lock looks like, and you design the key to unlock it." Its end-to-end platform, Pharma.AI, covers three core stages: target discovery, molecular generation, and clinical outcome prediction. The company claims to have compressed the drug discovery cycle for some drugs to within 18 months .
The CEO of InSilicon Intelligence put it bluntly: "In terms of AI, only Eli Lilly is stronger than us; there are no other companies."
That sounds arrogant. But the fact that Eli Lilly is willing to pay that price is itself an endorsement.
What Eli Lilly needs is not just an AI tool, but a "production line" that can continuously generate drug candidate molecules.
03
$2.75 billion deal
However, to understand this deal, you must first understand its structure.
The $115 million upfront payment is the actual money that has been deposited into the account today. But the remaining $2.6 billion is "milestone payments"—these will only be paid out in stages once Insilico's AI model actually produces validated targets, candidate molecules successfully enter human trials, and even completes clinical trials .
This can be seen as a structure that uses "potential value ceiling" to express strategic intent and "upfront investment" to test actual capabilities. For Eli Lilly, the balance sheet will not immediately be under pressure from $2.75 billion; for Insilico, each milestone payment is a public report card.
Industry experts agree on this transaction structure: it limits Eli Lilly's direct risk while giving Insilico the strongest business incentive—you don't get paid if you don't deliver.
But that's where the problem lies.
Drugs discovered by AI still ultimately need to undergo human clinical trials. The failure rate of clinical trials is frustratingly high across the pharmaceutical industry— approximately 90% of candidate drugs fail to survive Phase II trials . Whether AI-generated molecules can truly break this curse remains to be seen, as there is currently insufficient data to answer this question.
One industry observer's assessment is relatively fair: "The prediction for 2026 is that validation and disappointment will each account for about half. The field has moved from the speculative stage to the early clinical validation stage, but the gap between promises and performance remains significant."
Without any drugs reaching the market or regulatory approval, the entire field of AI drug discovery is still in a very long "proof-of-concept" phase .
No amount of large contracts can change this outcome.
04
AI Experiments in Traditional Industries
When Eli Lilly decided to incorporate AI into its core strategy, rather than just its lab budget, the significance went beyond that of a business contract.
Looking back, one of the most obvious trends in the AI industry over the past two years has been "vertical penetration"—large-scale model capabilities are being deployed to various professional fields, from code generation and legal documents to molecular design. The pharmaceutical industry, known as the "last bastion," has always been characterized by conservatism, stringent regulations, and extremely long R&D cycles, and is theoretically one of the slowest areas for AI penetration.
But the signal is clear now: even the most conservative money has started to circulate .
Eli Lilly's choice also has a deeper industrial logic. Smegglutide propelled Novo Nordisk to the pinnacle of global pharmaceuticals, while Eli Lilly's telposide achieved great success in the weight loss market. This "GLP-1 war" taught all pharmaceutical companies one thing: whoever finds the next "target" first will win the next decade .
AI is currently the most likely tool to accelerate this "finding" process.
InSilicon's $2.75 billion valuation today is less of a deal and more of an entry ticket—it's announcing to the entire industry that AI drug discovery has evolved from "research curiosity" to "commercial reality."
In the next two to three years, if the candidate molecules discovered by AI can demonstrate statistically significant advantages in clinical trials, then this "AI-driven drug-making revolution" will truly begin. If a large number of molecules entering clinical trials continue to fail with traditional probabilities, the industry will enter a period of cooling down, the transaction frenzy will subside, and Insilico will face a real test.
Nobody knows the outcome.
But Eli Lilly's $115 million check today is the most expensive vote ever cast.
This article is from the WeChat public account "GeekPark" (ID: geekpark) , author: Hualinwuwang, editor: Jingyu, published with authorization from 36Kr.
