rewrite this title AI Drug Discovery Stocks Continue to Disappoint – Nanalyze

We don’t need the Gartner Hype Cycle to tell us that investor expectations around artificial intelligence have reached ludicrous levels. Look no further than the world’s most valuable company with a market cap of $4.4 trillion: Nvidia (NVDA) represents roughly 8% of the entire S&P 500. (Let that sink in for a moment.) Much of the recent interest in the AI chip company’s hardware has been driven by the rise of generative AI (GenAI) following the public release of OpenAI’s ChatGPT at the end of 2022. 

Indeed, global venture capital investment in GenAI hit about $50 billion in the first half of 2025, already surpassing the $44.2 billion raised in 2024 and more than double the total for 2023. Most of the money so far this year went to OpenAI in a $40 billion venture round that valued the private AI company at $300 billion. A potential employee stock sale could jack up the price to $500 billion for a software firm that has surged to $12 billion in annual recurring revenue in just three years.

The AI frenzy is hitting just about every sector of the economy including healthcare. Since 2020, more than $30 billion in financing has been poured into AI-driven life sciences companies, according to Citeline, a pharmaceutical market research firm. In 2024 alone, AI healthcare startups raised a reported $10.5 billion. One of the biggest rounds went to a new AI drug discovery startup called Xaira Therapeutics, which landed a whopping $1 billion Series A round. Insilico Medicine, a well-known AI drug discovery startup, scored $100 million. Aside from these and hundreds of other private AI healthcare companies, we’ve been following a handful of AI drug discovery stocks, including Recursion (RXRX) and AbCellera (ABCL). 

How to Measure Success?

In today’s article we’re going to check in with these industry leaders and decide whether we are still interested in the AI drug discovery investment thesis based on the criteria we introduced in a video about the future of AI medicine last year. 

It really boils down to which company’s AI-powered drug discovery platform is the best. Since we’re MBAs and not computational biologists, we’re going to rely on other measures than the number of peer-reviewed papers a biotech has published. And, in this case, revenue growth for AI drug discovery firms is also not a reliable metric because it is inherently lumpy. By lumpy, we mean that these companies rely on royalties, licensing fees, milestone payments, grants, etc., mainly from partnerships with big pharma, rather than steady, predictable income.  

Certainly, this revenue is important, but the real money is in a commercial breakthrough – a drug that successfully completes all clinical trials, receives government approval (mainly from U.S. and European regulators), and becomes available at your local pharmacy for an astronomical amount of money. So far, no AI-powered drug discovery platform has made it across the finish line. 

However, there are certainly encouraging signs. Since 2015, AI-powered biotechs and their pharma overlords have introduced 75 molecules into clinic trials, according to one recent study. Of the two dozen that reached Phase I trials, the success rate was estimated at 80-90%. That is substantially higher than historical industry averages, which range from about 40% to maybe as high as 65%. In Phase II, 10 AI-discovered molecules completed trials with a success rate of 40%, which is in line with historical industry averages of 30-40%. Keep in mind that these are a very small sample size, and over time these percentages could drop. The other selling point about AI-driven drug development is that it is potentially faster and cheaper. For example,  Insilico Medicine says it has reduced preclinical and early clinical timelines from four years to under 18 months for certain drug candidates. 

Recursion Versus AbCellera

This brings us back to Recursion and AbCellera. Both companies offer a high-risk, high-reward profile, as we wrote a few years ago in our last full profile on the latter. More recently, however, we have reevaluated our approach as both companies shift away from being pure discovery platforms for hire and clinical-stage biotechs in their own right. That adds more risk and cost to a business model that already comes with a great degree of financial uncertainty.

AbCellera

Let’s start with AbCellera. We issued an alert earlier this year when shares fell about 16% after a disappointing 2024 earnings report. However, what really caught our attention was the company’s boast that it was transitioning from being a platform company to a clinical-stage biotech. We used to like AbCellera because it partnered with leading biopharma firms to churn out drug candidates. To wit: Its platform combines single-cell sequencing, microfluidics, and AI to discover antibodies in 70% less time than traditional methods, processing 100 million antibodies per day. The platform achieved a 99.7% success rate in identifying viable antibody candidates. The company’s main source of revenue is from royalty payments from those partnerships.

Now the focus is on its internal program and pipeline. That shift signficantly changes the risk-reward profile. Instead of sharing risk with big-pharma partners and earning research fees, milestones, and royalties across 100-plus partnered programs, AbCellera now bears more of the clinical, regulatory, and financing risk on its own early-stage assets like ABCL635 and ABCL575, where failure odds are inherently high and timelines long. 

Near-term revenue visibility remains limited and lumpy. For instance, AbCellera’s Q2-2025 revenue surged to $17.1 million, a 133% year-over-year increase, which has helped drive the stock up nearly 50% and to a market cap back over $1 billion this year. However, a big chunk of revenue came from a one-time $10 million licensing fee from its acquisition of a “humanized” rodent antibody platform from Trianni, a San Francisco biotech specializing in antibody-discovery technologies. The tech uses genetically modified rodents whose immune systems have been engineered to produce antibodies with human-like sequences instead of natural rodent antibodies. In turn, these human-like antibodies can be used to develop therapeutic drugs because they theoretically cause fewer immune reactions in patients. Again, theoretically, AbCellera can apply its AI tools to rapidly analyze, select, and optimize the best candidates from the rodent platform for development. 

Recursion

Recursion seems to be going down the same mouse rabbit hole as AbCellera in regard to internal versus external pipeline focus. This strategic pivot, underscored by the Q1-2025 decision to deprioritize three clinical programs – including its most advanced candidate, REC-994 – ratchets up the company’s risk profile. The original investment thesis for Recursion was built on its AI-driven drug discovery engine, Recursion OS, a platform capable of churning out a diversified portfolio of candidates to de-risk development through scale.

Recursion appears to be abandoning this diversified platform-as-a-service model and concentrating significant resources into fewer, earlier-stage programs. After the big merger with Exscientia, we expected the pipeline to expand, not contract, leaving Recursion with no candidates in Phase III trials. This puts even more pressure for its remaining lead candidates, like REC-617 or REC-1245, to deliver a decisive clinical win. And, despite the cost-saving measures from trimming the pipeline, the company’s cash runway is only vaguely some point in 2027. That seems a bit optimistic after Recursion burned through more than $370 million in the first half of this year. 

Like AbCellera, Recursion relies on its partnership programs for revenue, though more from milestone payments and research fees rather than royalties. In its most recent quarter, the company earned about $19 million based on contractual achievements, such as progressing drug candidates to specific R&D milestones or delivering proprietary datasets for drug discovery use. For example, Recursion earned a $7 million milestone payment from Sanofi during the Q2-2025 for advancing a clinical program. Previously, the company reported a one-time $30 million fee from Roche for licensing its neuroscience phenomaps dataset.

Recursion is projecting $100 million in potential payouts in 2026 (the keyword being potential), which leaves us wondering whether we should continue “liking” this stock. While we praised the Exscientia/Recursion merger as increasing cumulative pipeline breadth, the latest pipeline post merger has no later stage candidates and no apparent breadth. All the partnerships with major pharmaceutical companies have now been swept up into the below table which appears as a footnote to the company’s own internal pipeline.

So now the company gets to pick and choose what updates they provide investors relating to the above partnerships. Revenues will therefore become the only indicator that progress is being made for these various programs that are being worked on. Recursion calls these, “partnership catalysts,” and they all translate to pharma companies deciding to use the platform because it is becoming increasingly useful.

In order to expand their platform data, Recursion also partnered with a company we happen to be holding – Tempus AI (TEM) – which may be looking more like a pick-and-shovel play on AI-powered drug development.

A Pick-and-Shovel Play?

Once upon a time, we were more willing to embrace uncertainty around AI-powered ventures like drug discovery. But there are fewer excuses as other AI industries repeatedly demonstrate the stickiness and steady revenue of their platforms through subscription (and now usage-based) models. Even if this sort of model is not transferable to drug development, there are examples of companies using AI to generate serious alpha for healthcare applications. 

For instance, take Tempus AI, which has been riding the AI hype train better than most this year. In Q2-2025, the company recorded nearly 90% year-over-year revenue growth to about $315 million. Full-year 2025 revenue guidance is $1.26 billion, representing 82% annual growth. While about three-quarters of revenue is related to genomics, the company is leveraging its vast data wealth (350 petabytes and counting) from diagnostic testing to power its emerging healthcare services segment, including AI-powered drug discovery.

For instance, the company launched Tempus Loop in April 2025. The new platform leverages Tempus’s real-world data to identify patient subpopulations with similar clinical patterns, then uses systems biology to reveal novel target genes. The company has already deployed Loop with a major pharmaceutical partner, validating drug targets within one year – a significant acceleration from traditional five-year timelines. In addition, Tempus is developing what it claims will be the “largest multimodal foundation model in oncology” through a $200 million collaboration with AstraZeneca and Pathos AI. This AI model will extract biological insights, identify novel drug targets, and support therapeutic development using Tempus’s prodigious dataset. While Tempus is focused on cancer, isn’t curing cancer kind of what we think AI should be doing in the first place?

Conclusion

For those of us who believed that AI drug discovery companies like Recursion and AbCellera would revolutionize the industry by enabling multiple shots-on-goal quickly and cheaply, the recent pivots from platform to biotech are a disappointment. We’re not suggesting that success is impossible but both are now just as likely to add their names to the long list of biotechs facing uncertain prospects and high cash burn while waiting for that elusive breakthrough. As risk-averse investors, we find both companies have not shown enough proof of potential for their platforms. We don’t look favorably upon the risk-reward ratio of early-stage biotech companies, and that’s exactly what these two companies are looking like.