The Computable Molecule

Nicholas Mitsakos

Artificial Intelligence Is Reordering Life Sciences

The New Force

  Artificial intelligence is no longer a tool that the life sciences industry is adopting. It is a force that is relocating where value is created and who captures it. A tool speeds up an existing process. A force changes the structure of the industry that uses the process. Three costs are collapsing at once: drug discovery, company independence, and the ability to reach the patient. Each of these costs was, for forty years, a moat protecting the incumbents who could afford to pay it. AI is draining all three moats. There is a fourth dimension. The capacity to discover and manufacture medicine has become a strategic infrastructure. It now sits in the same category as energy, semiconductors, and compute. The reordering across the full value chain of discovery, development, capital, and distribution is impacting life sciences and creating new, significant opportunities.

Discovery and Distribution

  For four decades, the industry was predictable. Biotech discovers. Pharma distributes. A small company with scientific talent and venture capital would find a molecule, prove it works in early trials, and then hand the asset, or itself, to a large pharmaceutical company with the balance sheet to finance late-stage trials and the global sales force to place the drug in physicians’ hands. The biotech was the laboratory. Pharma was the factory, sales, and distribution. This paradigm held because each side controlled something the other could not replicate. The biotech controlled the science. The incumbent controlled the capital and the channel. Neither could function without the other, so value was shared. But the larger share went to whoever owned distribution. This is shifting. Acquirers still hold enormous capital and a genuine, accelerating need to refill pipelines. What is changing is the leverage, the terms, and who captures the value. The company that once had no choice but to sell early is discovering that it now has options that did not exist in a pre-AI era.

Computable Discovery

  The economics of discovery are brutal. A new drug costs well over a billion dollars and a decade of work to bring to market, and the overwhelming majority of candidates fail somewhere along the way. Most of that cost was money spent on molecules that would never become medicines. AI has the potential to change the odds of success dramatically. Models trained on chemical and biological data now predict a candidate’s toxicity, binding affinity, and developability before a single molecule is synthesized in a laboratory. The expensive, slow, physical work of trial and error moves into silico, where iteration is nearly free. Programs that build on computationally predicted protein structures, more than half of the new programs, can now pursue targets that had no experimental structure at all. Promise to Evidence In late 2023, roughly two dozen AI-originated molecules had reached the clinic. By early 2026, the number had approached 200. Drug candidates designed with these methods are clearing Phase I safety trials at rates between 80 and 90%, compared with a historical industry average closer to 50%. Discovery-to-clinic timelines that once ran four to five years have, in the strongest cases, been compressed to under eighteen months. The first fully AI-designed molecule produced both safety and efficacy data in humans in 2025. Fifteen to twenty AI-originated candidates are entering pivotal trials this year, and independent analysts put the odds of first full regulatory approval by 2027 at roughly 60%. The aggregate effect is a structural improvement in capital efficiency. The annual value of generative AI to the pharmaceutical sector could be between $60 and $110 billion. More than eighty percent of large drug developers now deploy these methods in some form. This is no longer experimental. It is becoming the default infrastructure of discovery. AI has not made drug development easy. It has made it more tractable, capital-efficient, and predictable.

In Vivo, Veritas

  Discovery is being industrialized. The clinical trial is not. “In Vivo, veritas” simply means that only by testing in humans and understanding organic performance can we truly know the truth. This remains the fundamental bottleneck in drug discovery. Perhaps it can be surmounted, but this is an even greater challenge. Phase I tells you a molecule is safe. It does not tell you the molecule works. The eighty-to-ninety-percent figures that dominate the headlines measure safety, not efficacy, and safety was never where most candidates died. The expensive failures occur in Phases II and III, where a drug must demonstrate meaningful clinical benefit in large, randomized, years-long studies. AI’s early advantage in hit rates means very little until it survives that test. We are receiving the first serious evidence on whether the early advantages carry over into the trials that actually determine approval. The mechanism designs molecules with better drug-like properties. But the data is incomplete, and it is foolish to extrapolate continuous progress in a straight line from early Phase I success to approval. That has still not been determined, but it remains promising. AI is reshaping the trial upstream. Advancements include patient selection based on molecular and biomarker profiles, site identification, and enrollment modeling. Additionally, synthetic control arms and digital twins reduce the number of patients a study requires, and adaptive designs adjust in the trial. Each of these compresses the cost and duration of the single most expensive activity in the industry. Efficacy in humans via standard trials remains the last analog step in a chain that is otherwise going digital. It is too optimistic to predict that this will ultimately become digital. Drug regulations and regulators will likely impose too much friction on the digitization of human data. It may be possible, but it will be a very lengthy timeline.

New Capital Models

  Cheaper discovery means new capital models and structures. Cheaper clinical validation means capital creates more leverage and the potential for independence. Intense capital expenditure programs to reach final drug approval typically require a capital partner via a licensing arrangement with a large pharmaceutical company or selling the company outright. Now, growth capital, typically available for technology companies, especially as they transformed from capital-intensive hardware to software “eating the world,” can be available for AI-based biotech companies that are essentially growing via a high-margin software model as well. Capital is conserved along with independence. The computable molecule can reduce the risk of drug development and solve many issues via software. The financing structure has bifurcated. Early-stage capital has grown scarce and demanding. Investors now write milestone-based tranches rather than large upfront commitments, releasing capital only as risk is retired. At the same time, capital has concentrated in later, more mature rounds backing assets that have already cleared meaningful clinical hurdles. The market is paying the premium since progress is easy to verify and less speculative. The companies that capture the new economics are those that treat AI as a core competency and build proprietary data assets and predictive capability. Value creation is shifting to data ownership. The competitive advantage is not using AI. The new AI-based competitive advantage is the proprietary, real-world experimental data that makes one company’s model better than another’s, and that data compounds. AI creates a biotechnology data flywheel, extracting value from proprietary data, using it to make more effective products, and using the data from those products to build better models. Drug discovery platforms have received premium valuations, but this is morphing into quality data, the development of AI-based biological vectors, and the creation of a credible path to commercial success. Capital is now applied more efficiently to greater risk-return opportunities. AI is creating more irrefutable science and potentially delivering greater value.

Distribution

  The pharmaceutical sales force was the critical infrastructure component that enabled a new pharmaceutical to reach patients. For most of the modern era, a biotech had no way to reach patients at a commercial scale without an incumbent’s distribution machine. This is changing selectively, but the paradigm is clearly shifting, and the industry structure is being affected. Direct-to-patient distribution By 2025, approximately 90% of leading pharmaceutical companies are either planning or have in place a direct-to-patient program. In some of the largest consumer-facing therapeutic categories, an increasing percentage of new prescriptions are now filled through the manufacturer’s own channel. Examples include over 25% of new metabolic therapeutic drugs sold directly to consumers from the drug manufacturers. A majority of clinicians now rank the manufacturer’s direct channel as the most clinically acceptable route when a payer denies coverage, ahead of independent telehealth and retail membership. The forces behind this are structural. Self-administered therapies in consumer-aware categories such as metabolic disease, dermatology, migraine, and fertility do not require intensive in-person monitoring. This makes a mostly digital delivery model viable. Patients expect convenience. Payers create friction. A direct channel improves access, lifts adherence, and gives the manufacturer a direct relationship with the patient and the data that relationship generates. Recent policy has accelerated the shift. Government-facilitated direct-pricing portals now advertise steep discounts on approved therapies directly to patients, normalizing the idea that the drug company and the patient transact without the traditional intermediaries. At the same time, regulators have moved aggressively to shut down unauthorized compounded-drug telehealth channels, and manufacturers have used litigation to redirect those channels into their own authorized distribution. The direct channel is being both legitimized and consolidated. The incumbents understand that owning the channel directly to the patient captures and sustains substantial value through the patient relationship, data accumulated, and economic efficiency. Direct-to-patient distribution has not replaced the sales force. It has created an alternative pathway to the patient that did not exist at a commercial scale five years ago. It is a credible option that has permanently altered the power dynamic between innovators and distributors.

The Patent Cliff

  The coming patent cliff is a crisis for the incumbents. It may also be an opportunity. By 2030, $200 billion-$300 billion of patented and branded pharmaceuticals will lose exclusivity. The cliff increases over the next decade, where roughly 60% to 65% of the largest pharmaceuticals sold, representing more than 30% of total revenue, will lose patent protection. Internal pipelines cannot replace that fast enough. The only strategic and competitive response is to acquire additional IP. This has been the business model for large pharmaceutical companies for over 20 years. Essentially, large pharmaceutical companies are selling and distribution organizations that derive significant marginal profitability from every patented medication they sell through their infrastructure. Now those patents are falling away, and the inventory is not replenished to at least equal a sustainable replacement rate. Revenues and values will fall unless something changes more fundamentally. This is the crisis. Industry M&A more than doubled in 2025, exceeding $100 billion, and dry powder available for future transactions is approximately $1 trillion. In 2026, more than $20 billion in acquisitions is expected. The pace is likely to accelerate. A typical deal targets Phase II and Phase III drug candidates with a lower risk profile and is usually structured with contingent payments that share risk between buyer and seller. AI-related computable molecules reduce approval risk, require less capital, and are shifting competitive dynamics. Biotech companies that can plausibly reach the patient with lower overall capital requirements have added strength and leverage, can remain independent longer, and can command higher valuations.

Value Shifts

  Put the four shifts together, and value in life sciences is migrating toward whoever controls both the science and the patient relationship. For forty years, the incumbent captured the largest share of the drug market because it controlled distribution. The biotech that can now own both discovery and, in some cases, drug delivery. It captures the discovery premium and the distribution margin, owns the patient data that improves the next program, and remains independent longer, and exits at a higher value.

What Does Not Change

  Every genuine transformation arrives wrapped in claims that do not survive reality. Biology is hard. AI has not repealed the laws of physiology. Most candidates still fail. What changes with AI is where and how cheaply they fail. It kills weak molecules early, in silico, before they consume the capital that late-stage failure destroys. It de-risks the inexpensive failures, not the expensive ones. The body remains the final, unforgiving validator, and it has no interest in the elegance of the model that produced the drug. The model is not a moat. Predictive capability is increasingly available to everyone. A company whose only advantage is access to a good model has no advantage at all. The defensible position is that proprietary data generated from real experiments accumulates with every program and is unavailable to anyone who did not run those experiments. Companies that confuse buying AI with building AI competency will gain no advantage. Headlines mislead. Phase I success is not approval. Faster discovery is not a marketed product. A pipeline of AI-originated candidates is a portfolio of hypotheses, not a portfolio of medicines. It is still essential to discount the early-stage figures heavily and wait for the efficacy data. None of this diminishes the transformational impact of the computable model AI generates. Skepticism and effective execution remain critical, but the life sciences will be affected going forward, and competitive leverage will shift.

Strategy and Security

  The capacity to design, test, and manufacture medicines is now part of the national infrastructure, and it should be analyzed in the same way as energy and semiconductors are. The same AI that compresses drug discovery compresses biological design generally. That capability is dual-use by nature. The data, models, compute, and manufacturing base that produce next-generation medicines are the same assets that determine a nation’s biosecurity posture. A country that cannot discover and produce its own therapeutics is dependent in the way it is dependent when it cannot produce its own chips or its own energy. The vulnerability is identical; only the molecule differs. Supply chains are reorganizing. Policy pressure to reshore drug substance and biomanufacturing away from concentrated foreign sources is turning what looked like a compliance burden into a strategic tailwind for domestic capacity. The firms positioned to serve a reshored, AI-native discovery and manufacturing base are beneficiaries of a regulatory and geopolitical wave. Strategic, economic, and national-security decisions in this industry are now primary concerns. Biotech, based only on return on investment, is solving a fraction of the problem. The asset is medicine, the margin, the data, and the sovereign capability, combined.

Disruption and Transformation

  Life sciences are experiencing disruption. They are undergoing a structural reorganization, relocating value, renegotiating the terms between innovators and distributors, and making viable a new class of independent, vertically integrated companies that could not have existed a decade ago. Discovery has become computable. The distribution moat is cracking, and the incumbents are scrambling to reclaim a channel they once took for granted. The patent cliff has turned acquirers into urgent buyers and well-positioned innovators into sellers who can name their terms. The whole apparatus now sits inside the strategic perimeter of the nations that host it. The companies that treat AI as a core competency will build compounding advantages. The executives and investors who understand the reordering earliest and who back the right companies to navigate it with discipline and clarity will define the next generation of the industry. The molecule has become computable. The architecture of value creation in life sciences has changed.