The algorithmic inversion of pharma 

The century-old model is cracking

For more than a hundred years, pharma innovation followed a reassuringly stable pattern. A chemist had a hunch, a lab tested it, a molecule emerged after years of trial and error, and a company spent billions navigating it through clinical trials and regulatory approvals. Intellectual property sat squarely with whoever discovered the molecule. Manufacturing, distribution, and commercialisation were tightly bundled around that core act of discovery.

That model is now under pressure from an unlikely direction: Silicon Valley. The same forces that upended media, retail, and transportation — artificial intelligence, platform economics, and data network effects — are now converging on the most complex and consequential industry of all.

Artificial intelligence — spanning protein structure prediction, generative chemistry, reinforcement learning, and biological simulation — is rewriting the economics of drug discovery. When the equation becomes data + models + compute = molecules, the centre of gravity shifts toward whoever controls the computational infrastructure. And that terrain belongs overwhelmingly to technology companies.

Think of what happened in consumer electronics. Apple captures the lion’s share of smartphone value through design, software, and ecosystem control. Foxconn builds the devices but keeps thinner margins. In apparel, leading brands own the design and the brand; contract factories do the stitching. The value migrated upstream, toward whoever controlled the intellectual blueprint, while the physical making of things became interchangeable. Could pharma be heading down the same path — with tech companies owning the molecular ‘design’ and pharma firms reduced to manufacturing contractors? The industry is already more modular than most people realise: clinical trials are outsourced to contract research organisations, manufacturing to CDMOs, even regulatory writing to specialist agencies. If discovery itself becomes modular, the last bastion of pharma’s differentiation falls.

The AlphaFold moment — and what came after

The shift from theoretical possibility to tangible reality has a name: AlphaFold.

ame: AlphaFold. In 2020, Google DeepMind’s AlphaFold 2 cracked the protein structure prediction problem — a grand challenge that had stumped biologists for half a century. Since then, more than three million researchers have used the tool, and it has been cited in over forty thousand academic papers. Demis Hassabis and John Jumper won the 2024 Nobel Prize in Chemistry for the work.

But AlphaFold 2 was just the opening act. Its successor, AlphaFold 3, launched in May 2024, goes further: it predicts not just protein shapes but the full molecular dance between proteins, DNA, RNA, drug-like ligands, and ions — outperforming the best physics-based methods on standard benchmarks. In February 2026, DeepMind’s drug-focused spinoff, Isomorphic Labs, unveiled an even more powerful proprietary engine called IsoDDE. The company has been described as building ‘an AlphaFold 4’ in all but name, and it comes backed by partnerships with Eli Lilly and Novartis worth a combined three billion dollars. Isomorphic is preparing its first AI-designed oncology drugs for human clinical trials.

And DeepMind is far from alone. An open-source model called Boltz-2, built by MIT researchers and Recursion Pharmaceuticals, can predict how tightly potential drugs bind to their protein targets. EvolutionaryScale’s ESM3 generates entirely novel proteins that don’t exist in nature. Baidu and ByteDance have launched comparable platforms from China. The race to build a ‘foundation model for biology’ is well and truly on.

What makes this particularly potent is the compounding nature of the advantage. Each clinical outcome feeds back into the model, improving the next prediction. Each patient dataset enriches the training corpus. Each failed molecule teaches the system what to avoid. These data network effects — the same flywheel dynamics that made Google and Meta so dominant in their markets — create self reinforcing moats that late entrants will struggle to replicate. The companies that build these loops earliest will be hardest to catch.

The rise of the dark lab

If AI is the brain of the new drug discovery, robotic laboratories are its hands.

A new breed of ‘dark labs’ — fully automated facilities that operate around the clock with minimal human intervention — is emerging as a critical piece of the puzzle. Recursion Pharmaceuticals runs one of the most advanced: a platform that combines robotics, highcontent cellular imaging, and a staggering sixty-five-petabyte proprietary dataset in a continuous design–make–test–learn loop. When Recursion merged with Exscientia in 2024, it created an end-to-end system linking AI-driven molecular design with automated precision chemistry.

Insilico Medicine has gone a step further, deploying a humanoid robot in its AI-powered laboratory — designed to observe human scientists, learn their techniques, and eventually replicate them autonomously. Meanwhile, XtalPi, founded by quantum physicists at MIT, operates automated labs in Shenzhen, Shanghai, and Cambridge, Massachusetts, linking physics-based simulation with machine learning and robotic wet-lab work in a single closed loop.

These are early prototypes of what the industry calls ‘selfdriving laboratories’: systems that autonomously propose molecules, synthesise them, test them, and feed the results straight back into the generative model. The human scientist doesn’t disappear — but their role shifts from bench worker to systems architect. A 2025 industry review noted that while these autonomous platforms have dramatically accelerated the design–make–test–learn cycle, none has yet independently discovered a validated drug candidate. The technology is proven for acceleration; the question of whether it can improve clinical success rates remains the defining test for the field.

Enter Quantum: The next computational frontier

If AI is already disrupting drug discovery, quantum computing could supercharge it.

Here’s the core insight: drug-target interactions are fundamentally quantum-mechanical phenomena. When a drug molecule approaches a protein pocket, the electrons in both systems interact through quantum effects — superposition, tunnelling, entanglement — that govern whether the drug binds effectively. Classical computers can only approximate these interactions, and those approximations introduce errors, particularly for complex cases like metalloenzymes and chemical transition states. Quantum computers, by representing molecular states natively using qubits, can in principle simulate these interactions with far greater fidelity. In essence, you’d be using quantum mechanics to simulate quantum mechanics — an idea first articulated by the physicist Richard Feynman in the early 1980s.

McKinsey estimates that quantum computing could unlock between two hundred billion and five hundred billion dollars of value in life sciences by 2035. The world’s biggest pharma and technology companies are already placing bets. AstraZeneca is working with IonQ and NVIDIA on quantum-accelerated workflows for small-molecule drug synthesis. Merck and Amgen are collaborating with QuEra to predict the biological activity of drug candidates. Pasqal and Qubit Pharmaceuticals are applying neutral-atom quantum computing to model how drugs bind to proteins. 

Fault-tolerant quantum hardware is still years away, but hybrid quantum-classical pipelines are already producing useful results for specific chemistry problems. Researchers have demonstrated quantum workflows for covalent bond simulation and free energy calculations in prodrug activation — real pharma problems, not just academic exercises. When the hardware catches up, quantum simulation will likely become the foundational layer of the entire TechBio stack — feeding highfidelity molecular data into the AI models that sit above it. Nations that lead in both AI and quantum computing will hold the keys to the next generation of medicines.

Mapping the TechBio stack

Taken together, these developments are assembling into a coherent architecture — what we might call the TechBio stack. Think of it as the pharma equivalent of a modern tech platform, organised in layers, each reinforcing the ones above and below it. Whoever controls the critical middle layers controls the value.

At the base sits compute infrastructure: hyperscale cloud, GPU clusters, and emerging quantum accelerators. Above that, a biological data layer — population-scale genomics, proteomics, wearable telemetry, longitudinal health records — creates a powerful data moat. The foundation model layer is where molecules are actually generated: protein folding predictors like AlphaFold 3, small molecule designers, and antibody engineering systems. A digital twin and simulation layer compresses timelines by testing candidates in silico before they enter a lab. A clinical platform layer optimises trial design through AI, wearable monitoring, and decentralised recruitment. And at the top, a patient ecosystem closes the loop — continuous monitoring feeds real-world outcomes back into the models, making the whole system self-improving.

An important milestone: in 2025, the first drug with both its target and molecule designed entirely by AI completed Phase IIa clinical trials. No AI-discovered drug has yet received FDA approval, but the pipeline is maturing fast. The World Economic Forum projects that by 2025, thirty per cent of drugs in early development will have AI involvement. The gap between ‘AI-assisted’ and ‘AI-native’ drug discovery is narrowing with each quarterly earnings call from the major TechBio companies.

Who owns the molecule?

This brings us to perhaps the most consequential question of all: where does the intellectual property sit?

Traditionally, pharma patents protect specific chemical entities, synthesis methods, and therapeutic uses. In a computational world, the real source of competitive advantage may shift to the model architectures trained on proprietary datasets, the algorithmic design pathways, and the synthetic data pipelines that simulate efficacy. The shift is subtle but momentous: you don’t just own a drug — you own the machine that can generate an infinite number of optimised variants. And that machine gets better with every molecule it designs, every trial it observes, every patient outcome it ingests. 

Can the generative capacity of an AI model be patented? Does algorithmic design count as inventorship? Who owns a molecule that a model designed autonomously from globally aggregated data? These aren’t abstract thought experiments. They are live questions that will reshape patent law and international IP treaties in the coming decade.

The geopolitics of molecular sovereignty

If you think this is purely an industry story, think again. The geopolitical implications are enormous

Historically, pharma power was shaped by who had the labs, the regulatory expertise, and the manufacturing plants. In a compute-driven world, power shifts to whoever controls advanced semiconductors, hyperscale cloud, population-scale biological datasets, and AI talent. Biological sovereignty becomes inseparable from digital sovereignty

The US leads in AI infrastructure, venture capital, and semiconductor design. Its tech giants have the scale to integrate computing, health data, and platform ecosystems into a unified stack. NVIDIA’s BioNeMo platform, Microsoft’s protein design tools, and Google’s AlphaFold ecosystem are already assembling the pieces. China offers a different model: state coordination, massive population data, and vertically integrated AI ambitions that could produce sovereign biological foundation models insulated from Western systems. Baidu, ByteDance, and Chinese research institutions have already launched protein prediction platforms of their own. The world may bifurcate into parallel biological model ecosystems — Western and Chinese — with limited interoperability, much like rival internet ecosystems today. 

Europe retains regulatory muscle through the EMA and data protection frameworks, potentially shaping global norms on algorithmic transparency and AI governance in medicine. But without competitive compute infrastructure, it risks becoming a rule-writer without innovation ownership. India occupies a unique position. It is the ‘pharmacy of the world’ in generics, has massive population dataset potential, and deep IT talent — but currently lacks hyperscale compute and foundation model leadership. The strategic choice is stark: remain the manufacturing backbone of global therapeutics, or climb the stack into algorithmic drug design leadership. The decision will shape India’s position in the global health architecture for decades.

The big-picture question is whether we’re heading toward consolidation or fragmentation. In one future, a small number of tech firms own the biological foundation models and license drug designs globally — efficient, but dangerously centralised. In another, multiple sovereign AI-biology ecosystems emerge, each protecting domestic data and compute — more equitable, but potentially slower. History suggests a messy middle: partial consolidation with strategic fragmentation at critical chokepoints, much like today’s semiconductor supply chains.

Pharma is not dead yet

Before declaring the triumph of TechBio, it’s worth acknowledging what pharma companies still bring to the table. Regulatory expertise is a genuine moat. Navigating FDA and EMA approvals requires decades of institutional knowledge, safety infrastructure, and pharmacovigilance systems that no tech company has built overnight. The sheer complexity of running a global Phase III clinical trial — with thousands of patients across dozens of countries, each with different regulatory requirements — remains a capability that takes years to acquire. Manufacturing is another barrier: biologics production, cell therapy scaling, and cold chain logistics are capital-intensive, process-sensitive operations that resist easy commoditisation. And commercial infrastructure — physician networks, insurance negotiations, market access — remains firmly in pharma’s hands.

The most likely near-term future is co-evolution, not outright replacement. Three models will probably coexist. First, tech discovers and licenses molecules to pharma for commercialisation — already happening today. Second, joint ventures with shared IP and pooled risk, which is likely to be the dominant model in the near term. Third, and further out, fully integrated TechBio companies that handle everything from discovery through to patient delivery. In January 2025, the US FDA issued its first draft guidance on AI in regulatory decision-making for drug products — a signal that the regulatory ecosystem is beginning to adapt to a world it didn’t build.

The equity question

There’s a darker possibility embedded in this transformation. If foundation models are trained primarily on Western population data, the therapies they design may carry systematic biases — working well for some demographics and poorly for others. If access to personalised AI-designed medicine is gated behind platform subscriptions, new forms of health inequality could emerge. The concentration of biological IP within a handful of corporations raises hard questions about who benefits from the algorithmic future of medicine — and who gets left behind.

On the other hand, AI could dramatically reduce R&D costs and direct attention toward neglected tropical diseases and rare conditions that have long been commercially unattractive. AlphaFold-powered research has already been used to identify existing drugs that could be repurposed to treat Chagas disease, a tropical parasitic illness that infects up to seven million people annually. The outcome will depend on governance: whether public AI infrastructure — think government-funded biological foundation models trained on public datasets — emerges as a counterweight to corporate concentration, and whether regulators mandate dataset diversity, transparency, and equitable access.

The Endgame: Therapy as software 

The most radical scenario looks like this: AI designs a molecule personalised to your genome. It’s synthesised on demand at a hospital-based automated unit. Your response data feeds back into the model. The therapy is updated like a software patch. The IP sits in the algorithm, not the molecule. The drug isn’t a product — it’s a service.

Speculative? Yes. But early precursors already exist in mRNA platforms and gene-editing pipelines, where the therapeutic modality is already programmable. Moderna’s mRNA platform, which designed a COVID vaccine in just two days from the viral sequence, is a glimpse of this future. And when quantum computing matures enough to simulate molecular interactions without classical approximations, generative AI could design compounds invisible to today’s methods. The convergence of quantum simulation and generative AI may be the ultimate compute substrate for programmable biology. That’s not science fiction — it’s engineering with a timeline.

The decisive question

The platform inversion of pharma is not inevitable. But it is structurally plausible — and the signposts are multiplying.

As drug discovery becomes computational, intellectual property migrates upstream toward whoever controls the models and the data. Platform economics could reshape health innovation as profoundly as they reshaped retail, media, and communications. The geopolitical stakes are immense: control of biological generative systems — and the quantum and classical computing infrastructure underlying them — could determine national resilience, economic competitiveness, and public health autonomy for the rest of this century.

The question is no longer whether AI will design drugs. It is who will own the systems that design them — and on what terms. Will the molecular blueprints of future medicines be open or proprietary? Distributed or concentrated? Governed for public health or optimised for shareholder return? In those answers lies the future balance of power in global health. The time to shape it is now.

References 

• Jumper, J. et al. ‘Highly accurate protein structure prediction with AlphaFold.’ Nature, 596, 583–589 (2021). z Abramson, J. et al. ‘Accurate structure prediction of biomolecular interactions with AlphaFold 3.’ Nature, 630, 493–500 (2024). 

• Google DeepMind. ‘AlphaFold: Five Years of Impact.’ DeepMind Blog, November 2025. 

• Isomorphic Labs. ‘IsoDDE: Drug Discovery Engine.’ Technical Report, February 2026. Reported in Nature, d41586-026- 00365-7. 

• Fang, Z. et al. ‘AlphaFold 3: an unprecedented opportunity for fundamental research and drug development.’ Precision Clinical Medicine, 8(3), 2025. 

• Wohlwend, J. et al. ‘Boltz-2: Bridging Structure Prediction and Molecular Property Estimation.’ MIT / Recursion Pharmaceuticals, 2025. 

• Hayes, T. et al. ‘Simulating 500 million years of evolution with a language model.’ EvolutionaryScale / ESM3, bioRxiv (2024). 

• Singh, R. ‘AI in drug discovery: 2025 in review.’ Drug Target Review, February 2026. 

• Huang, Y. et al. ‘Leading AIdriven drug discovery platforms: 2025 landscape and global outlook.’ ScienceDirect, November 2025. 

• Soorya, A. et al. ‘A hybrid quantum computing pipeline for real world drug discovery.’ Scientific Reports, 14, 16942 (2024). 

• Bonde, N.J. et al. ‘Quantummachine-assisted drug discovery.’ npj Drug Discovery, 3, 2 (2026). McKinsey & Company. ‘The quantum revolution in pharma: faster, smarter, and more precise.’ Quantum Technology Monitor, June 2025. 

• World Economic Forum. ‘How quantum computing is changing molecular drug development.’ WEF Agenda, January 2025. 

• US FDA. ‘Considerations for the Use of Artificial Intelligence to Support Regulatory DecisionMaking for Drug and Biological Products.’ Draft Guidance, January 2025. 

• The Nobel Committee for Chemistry. The Nobel Prize in Chemistry 2024: Computational Protein Design and Protein Structure Prediction. Royal Swedish Academy of Sciences, 2024.