During his graduate studies in Dr. Dean Li's laboratory at the University of Utah (currently the President of Merck Research Laboratories), Chris Gibson developed the technology behind Recursion and founded the company in 2013.

Initially, the core technology route of Recursion was "AI+phenotype screening", which analyzed cell images through machine vision and used deep learning algorithms to identify microscopic changes caused by drugs

Gibson's idea is simple. Since biological structure determines function, by building a large-scale biological image library and using cutting-edge computer vision technology to analyze its correlations, perhaps the key to interpreting life mechanisms can be found.

With the accumulation of massive cell images and compound data, Recursion's perspective has shifted from the initial "phenotype screening" to building large-scale models based on multimodal biopharmaceuticals.

Recursion has built one of the world's largest multimodal datasets, integrating gene, transcriptome, and multiple patient data sets.

Subsequently, Recursion iteratively developed multiple generations of basic models. These models have the ability to go beyond image recognition itself, and can autonomously learn and reveal deep biological laws from the cross analysis of cell images, compound structures, and clinical data.

With the expansion of models and data scale, the original computing power is no longer sufficient to meet the demand. At this moment, NVIDIA appeared.

In 2023, Nvidia announced a $50 million investment in Recursion to accelerate the development of breakthrough foundational models for AI in drug discovery. The stock price of Recursion rose by 78% in response.

In May 2024, the artificial intelligence supercomputer BioHive-2, jointly developed by both parties, was officially unveiled. It is driven by 63 DGX H100 systems and runs four times faster than the previous generation, becoming the fastest supercomputer independently owned and operated by global pharmaceutical companies at that time.

The launch of BioHive-2 provides Recursion with far more computing power support than ever before. It can not only significantly shorten the training cycle of models like Phenom-1, but also run multiple equally or even larger artificial intelligence projects in parallel.

At that time, Recursion chose a dual track approach: while collaborating with computing giants like Nvidia to continuously promote bio molecular generative AI models, it also provided research and development services for MNCs such as Roche and Sanofi; On the other hand, we are fully committed to advancing our clinical pipeline.

In 2023, Recursion acquired Cyclica and Valence for a total price of $87.5 million.

In August 2024, Recursion announced its merger with Exscientia, an AI drug development company dedicated to accelerating the discovery and optimization of small molecule drugs through AI technology. After the merger, Recursion is expected to announce approximately 10 clinical trials within the next 18 months.

In Gibson's view, the significance of this merger lies in the deep complementarity between the two parties. He stated that Exscientia's precise chemical tools and capabilities, including its new automated small molecule synthesis platform, will enhance our translational abilities in biological and chemical exploration. I am delighted to continue building the best example of the next generation of biotechnology companies. We still feel like we're just starting out. ”

Ideals are full, but reality is always fragile.

After the inclusion of Exscientia, Recursion significantly reduced its clinical pipeline in the first quarter of 2025, with at least one-third of the pipeline being cut.

The most important one among them is REC-994, which is a superoxide dismutase (SOD) inhibitor mainly targeting cerebral cavernous malformation (CCM), a rare disease for which there are currently no approved drugs.

REC-994 is one of the earliest projects approved by Recursion and was once the most advanced clinical pipeline. In the phase II study, it showed positive signals of lesion shrinkage and functional improvement. However, in subsequent long-term expansion studies, this therapeutic effect was not maintained, and there was no positive trend in the dose conversion group.

Another important pipeline that has been terminated is REC-2282, a small molecule pan histone deacetylase (HDAC) inhibitor that can penetrate the blood-brain barrier, aimed at treating progressive type 2 neurofibromatosis (NF2) mutant meningiomas. Due to limited tumor shrinkage and clinical efficacy observed in clinical research, Recursion has decided to discontinue development.

Despite being established for about 12 years, it is still deeply mired in losses, coupled with significant adjustments to the top pipeline, which is gradually exhausting the patience of capital towards Recursion.

On February 25th, Recursion announced its full year 2025 results, with revenue of $74.68 million and a net loss of $645 million. Since its listing in 2021, Recursion's stock price has been declining year by year, with less than one tenth compared to its peak.

In 2025, Gibson announced his resignation as CEO and assumed the position of Chairman of the Board. The successor to the CEO position is Najat Khan, who previously served as the Chief Data Science Officer and Senior Vice President of Research and Development Strategy, Product Portfolio, and Operations at Johnson&Johnson Innovative Pharmaceuticals.

The founder's exit from the front line has further deepened the market's concerns about Recursion.

At this point, Nvidia has chosen to clear its inventory. But does Nvidia's exit represent the ultimate failure of Recursion? The answer is negative.

Najat Khan emphasized during the latest performance conference call, "Our technology collaboration with Nvidia is still ongoing, with a focus on automation and supercomputing. This divestment is part of Nvidia's broader portfolio adjustment

According to the 13F document, Nvidia also liquidated its holdings in Applied Digital, Arm Holdings, Wenyuan Zhixing, and other stocks in the fourth quarter while selling Reucrsion.

ARK's decision to significantly increase its holdings at this time clearly sees an unusual opportunity.

On the one hand, the Recursion stock price has been at a historical low after years of decline, providing a window for ARK to increase its holdings at a lower cost; On the other hand, although Recursion has significantly streamlined its pipeline, the core projects it retains still have considerable potential and room for imagination.

In December 2025, Recursion released positive data for phase Ib/II trials of REC-4881. This is a variant MEK1/2 inhibitor under development for the treatment of familial adenomatous polyposis (FAP).

After 12 weeks of treatment, the median polyp burden of assessable patients decreased by 43%, with 75% of patients experiencing a decrease in polyp burden. More importantly, 82% of patients maintained a decrease in polyp burden after 12 weeks of discontinuation, a median decrease of 53% from baseline levels.

Morgan Stanley analyst Priyanka Grover predicts that the annual sales of REC-4881 are expected to exceed $1 billion, with a 60% probability of commercial success.

The significance of REC-4881 goes beyond clinical data itself, it is also the "touchstone" of the Recursion OS platform.

This drug was first introduced by Recursion from Takeda Pharmaceuticals in 2020. Recursion utilized its automated drug discovery platform to test over 200 potential molecules in Takeda Pharmaceuticals' compound library and compared them with the most effective potential therapies for cancer carrying specific mutations, ultimately discovering the potential of TAK-733 (REC-4881).

Initially, REC-4881 was primarily studied as a solid tumor treatment drug, but Recursion repositioned it as a candidate drug for FAP treatment. REC-4881 is currently the first MEK1/2 inhibitor to be clinically studied for this disease.

Recursion stated that communication with the FDA regarding the registration study of REC-4881 will be initiated in the first half of 2026, which is crucial for the potential design of the registration study and patient endpoint indicators. It is expected that more phase Ib/II clinical data will be released in the first half of 2027.

In addition to REC-4481, Recursion also released progress expectations for multiple pipelines. REC-1245 (RBM39) may disclose early safety data, and IND applications for REC-7735 (PI3K α H1047R) and REC-102 (ENPP1) may receive approval results in the second half of this year.

To cope with cash flow pressure and extend the capital turnover cycle, Recursion will take a series of "cost saving" measures in 2025, including streamlining its research and development pipeline and implementing approximately 20% of large-scale layoffs.

As of the end of 2025, Recursion's book value reached $754 million, an increase of nearly $150 million from the end of 2024. According to the current operational plan, Recursion expects to extend its cash turnover period until early 2028 without additional financing, thereby providing more ample financial support for the core project REC-4881 and a wider product line.

The case of Recursion provides a clear perspective for observing the evolution of the AI pharmaceutical industry.

The most prominent feature of this field currently is that the boundaries of business models are melting.

In the past, people used to divide AI pharmaceutical companies into three categories: one is a SaaS provider mainly providing software platform services, represented by Schr ö dinger; One is AI+CRO, which provides outsourcing services for research and development enterprises, such as Jingtai Technology; And AI Biotech for self built drug pipelines.

But now, this division is no longer sufficient to describe the true picture of the industry.

As a leading player in the SaaS model, Schr ö dinger's top 20 global pharmaceutical companies are all customers of Schr ö dinger, but they are no longer limited to selling software, but are now heavily expanding their self-developed pipeline. More and more AI pharmaceutical companies are inclined to build composite businesses, such as Yingsi Intelligent Recursion。

The integration of business models is only superficial, the watershed that truly determines fate is clinical validation. No matter how the business models are combined, the fundamental question ultimately comes back to whether the drugs discovered and designed by AI can prove themselves in human trials.

Yingsi Intelligence has taken a crucial step at this point. Its self-developed drug Rentosertib has become the world's first candidate drug discovered by AI and designed by generative AI, and has successfully completed phase IIa clinical trials on indications for idiopathic pulmonary fibrosis.

On the other side, Verge Genomics， The story of this once highly anticipated AI pharmaceutical star company is not so perfect. Its ALS drug VRG50635 failed in clinical trials and ultimately had to terminate research and development. It announced a strategic transformation, abandoning its self-developed pipeline and returning to the role of a technology service provider to provide AI platform support for large pharmaceutical companies. The founder of the company admitted that this move is to streamline the cost structure, ensure cash flow, and survive.

This case, one positive and one negative, reveals the true situation of the industry at present. The success of clinical validation is not an easy task, and the cost of failure is extremely heavy.

By the end of 2025, there will be over 350 global AI pharmaceutical companies and over a hundred domestic companies, but the vast majority are still in the start-up stage, with only a dozen or so truly entering the growth or maturity stage. Although capital is accelerating its layout in top enterprises, the entire industry is still deeply mired in high investment and long-term difficulties, with cash flow pressure following closely, and layoffs, contraction, and mergers and acquisitions becoming the norm.

The future industry competition will no longer be just a contest of technology and algorithms, but a comprehensive game of clinical implementation ability and survival resilience.

The back and forth between Nvidia and Cathie Wood does not directly prove the failure or success of Recursion. It reflects that the development of AI pharmaceuticals has approached its own "singularity moment" so far.

The market is no longer satisfied with just one story or one page of PPT, but is starting to focus on whether this concept can be implemented and validated. From fanaticism to rationality, from storytelling to product development, this is a necessary path for an industry to mature.

The story of Recursion continues. It may become a "martyr" in the field of AI pharmaceuticals, or a survivor of crossing the valley of death. But no matter what, every setback it is going through is drawing a clearer map for the entire industry.

The speed at which technology changes the world is often slower than optimists expect, but faster than pessimists expect. The singularity moment of AI pharmaceuticals may be at the next turning point.