Jensen Huang Envisions AI Transforming Biology Into Engineering Discipline

In January 2025, at the Precision Medicine World Conference in Silicon Valley, NVIDIA CEO Jensen Huang delivered a statement that has since rippled through boardrooms, laboratories, and investor calls across the biotech and pharmaceutical industries: "For the very first time in human history, biology has the opportunity to be engineering — not science" [1].

It was not a throwaway line. Huang has been systematically building toward this vision for years, positioning NVIDIA's GPU computing empire as the backbone of a revolution he believes will rival the rise of the internet itself. But the claim — that biology, with all its stochastic chaos and emergent complexity, can be reduced to the kind of deterministic, iterative engineering that built the semiconductor industry — is one of the most audacious bets in modern technology.

The Vision: From Sporadic Discovery to Compounding Progress

The core of Huang's argument rests on a distinction between science and engineering. In science, breakthroughs are unpredictable — a researcher may spend decades pursuing a dead end. Engineering, by contrast, is systematic: each advance builds on the last, enabling exponential progress. "When something becomes engineering, not science, it becomes less sporadic and exponentially improving," Huang has stated. "It can compound on the benefits of the previous years. And every researcher's contributions compound on each other" [2].

This framing resonates with a broader consensus emerging among the most powerful figures in AI. In a landmark conversation, Anthropic CEO Dario Amodei predicted that "AI-enabled biology and medicine will allow us to compress the progress of 50–100 years into 5–10 years," while Google DeepMind CEO Demis Hassabis — whose AlphaFold earned him the 2024 Nobel Prize in Chemistry — emphasized that the goal is "not to replace scientists, but to let scientists armed with AI achieve breakthroughs once thought impossible" [3].

The three leaders share a conviction: that generative AI, foundation models, and accelerated computing are converging to create a new paradigm in biological research — one where proteins can be designed rather than merely discovered, where genetic mutations can be predicted computationally rather than identified through years of clinical observation, and where drug candidates can be generated and screened at a pace that was unthinkable a decade ago.

NVIDIA's Infrastructure Play

Huang is not merely evangelizing. NVIDIA has committed billions of dollars and years of platform development to making this vision operational.

The company's BioNeMo platform — an open development framework for AI-driven biology and drug discovery — now features more than a dozen generative AI models and cloud services for proteins, DNA, RNA, and small molecules [4]. Its Clara ecosystem encompasses AI tools for genomics (Parabricks), medical imaging (MONAI), medical devices (Holoscan), and drug discovery [5].

In January 2026, NVIDIA announced a $1 billion co-innovation lab with Eli Lilly, the pharmaceutical giant, to train large-scale biomedical foundation and frontier models for drug discovery. Lilly deployed what it described as the world's largest AI factory for drug discovery, powered by NVIDIA Blackwell-based DGX SuperPOD systems [6]. The five-year investment covers talent, infrastructure, and compute — a scale of commitment that signals how seriously the pharmaceutical industry is taking Huang's thesis.

NVIDIA also invested $50 million in Recursion Pharmaceuticals in 2023, leading to the creation of BioHive-2, described as the largest supercomputer in the pharmaceutical industry, packing 504 NVIDIA H100 GPUs. Within 30 days of the partnership's launch, the system had screened approximately 36 billion chemical compounds for predicted protein targets [7].

Evo 2: The "GPT Moment" for Biology?

Perhaps the most tangible demonstration of the biology-as-engineering thesis is Evo 2, a foundation model developed by the Arc Institute in collaboration with NVIDIA, Stanford, UC Berkeley, and UCSF. Released in February 2025, it is the largest AI model for biology ever built [8].

Trained on over 9.3 trillion nucleotides from more than 128,000 whole genomes spanning all domains of life, Evo 2 can process genetic sequences up to 1 million nucleotides long. It can identify disease-causing mutations — achieving 90% accuracy in BRCA1 mutation predictions for cancer risk assessment — and can design new genomes as long as those of simple bacteria [3][8].

The model was trained on NVIDIA DGX Cloud using more than 2,000 H100 GPUs over several months, and is now integrated into the BioNeMo framework and freely available through the Arc Institute's GitHub [8]. A user-friendly interface called Evo Designer accompanies it, lowering the barrier for researchers without deep machine learning expertise.

If AlphaFold was the "ImageNet moment" for structural biology — proving that AI could solve a 50-year grand challenge — Evo 2 represents something potentially more consequential: the ability to not just predict biological structures, but to generate new biological code.

The Clinical Evidence: Promising but Preliminary

Source: BioSpace / PitchBook

Data as of Mar 14, 2026CSV

The investment thesis depends on whether AI-designed molecules actually perform better in the clinic. The early data is encouraging — but far from conclusive.

AI-native biotechs have reported 80–90% Phase I success rates, compared to the industry average of 40–65%. In Phase II, the most challenging stage where most drugs fail, AI companies have achieved approximately 40% success rates versus the industry benchmark of 29% [9].

Insilico Medicine demonstrated that AI could compress drug candidate nomination from the typical 3–4 years down to just 12–18 months, achieving positive Phase IIa results for its idiopathic pulmonary fibrosis treatment ISM001-055 [10]. PitchBook has projected that if early trends hold, the overall probability of a drug successfully navigating clinical trials could increase from approximately 8% to 18% [9].

But the sample size remains vanishingly small. AI-focused biotechs have completed just 10 clinical trials to date [9]. As the American Chemical Society's Chemical & Engineering News noted, AI has not yet demonstrably improved the pharmaceutical industry's roughly 90% clinical failure rate [11]. And chemist Derek Lowe has pointed out that many "AI-discovered" drug targets were already well-known — not genuinely novel findings [11].

The Market Surge

Source: Precedence Research / Coherent Solutions

Data as of Mar 14, 2026CSV

Despite these caveats, capital is flowing into the space at an accelerating pace. The global market for AI in drug discovery was valued at approximately $3 billion in 2024 and is projected to reach $15.2 billion by 2030, reflecting a compound annual growth rate of roughly 31.7% [12]. More aggressive forecasts project the broader AI-in-pharmaceuticals market reaching $49.5 billion by 2034 [12].

Isomorphic Labs, DeepMind's drug discovery spin-off, secured a $600 million Series A in March 2025, with partnerships valued near $3 billion [3]. Drug discovery AI funding overall rebounded sharply to $3.8 billion in 2024 after a downturn [3].

NVIDIA itself has become the financial engine of this transformation. The company's full-year revenue rose 142% to $115.2 billion in fiscal 2025, and for the third quarter ended October 2025, it reported $57 billion in revenue — up 62% year-over-year [13]. While healthcare represents a fraction of this total, Huang has repeatedly identified it as one of the company's most important long-term growth vectors.

Healthcare AI Adoption: The Survey Data

NVIDIA's own 2026 survey of the healthcare and life sciences industry reveals how rapidly AI is moving from experimentation to deployment. 70% of organizations are now actively using AI, up from 63% in 2024, while 69% are deploying generative AI and large language models, up from 54% [14].

Among pharmaceutical and biotech companies specifically, 57% are now using AI for drug discovery, and 46% report achieving measurable return on investment from these efforts. Perhaps most tellingly, 85% of executives plan to increase their AI budgets this year, with nearly half budgeting increases exceeding 10% [14].

The adoption of agentic AI — systems that can autonomously execute multi-step research workflows — stands at 47%, signaling that the industry is already looking beyond simple prediction tools toward AI agents that can design, execute, and iterate on experiments with minimal human intervention [14].

The Skeptics' Case

Not everyone is convinced that biology can be engineered like software.

The fundamental challenge, critics argue, is that biological systems are irreducibly complex. Equating binding affinity with bioactivity, as many AI models implicitly do, ignores the elaborate physiological context in which drugs must operate [15]. A molecule that looks perfect computationally may fail in a living body for reasons no model currently captures.

Clinical trial duration, regulatory review timelines, and manufacturing scale-up remain bound by biology, patient enrollment logistics, and regulatory requirements — none of which AI can bypass [10]. Even if AI can compress early discovery by 30–40%, the longest and most expensive phases of drug development remain stubbornly analog.

Data quality presents another obstacle. A survey found that 68% of tech executives identify poor data quality and governance as the primary reason AI initiatives fail [15]. AI models trained on limited or biased datasets may generate confident but clinically meaningless predictions.

There is also the concern that AI applications in drug discovery remain fragmented — addressing isolated tasks like target identification or molecular generation rather than integrating insights across the entire drug development pipeline [15]. The "engineering discipline" Huang envisions would require seamless end-to-end integration that does not yet exist.

Fortune reported in September 2025 that the AI drug breakthrough "is taking a long time to arrive, for reasons that may have little to do with the technology's limits" — pointing to organizational inertia, regulatory conservatism, and the irreducible timescale of biology itself [16].

What's at Stake

The scale of the potential impact is difficult to overstate. On average, bringing a new drug to market takes 14.6 years and costs approximately $2.6 billion [10]. The global pharmaceutical industry generates over $1.5 trillion in annual revenue. If AI can meaningfully compress timelines and improve success rates — even modestly — the economic and human implications are enormous.

Novo Nordisk demonstrated the productivity gains possible when it reported reducing clinical documentation from over 10 weeks to just 10 minutes using AI [3]. While documentation is far simpler than drug design, the magnitude of the improvement illustrates the kind of efficiency gains that become possible when biological workflows are digitized.

Huang's vision is not that AI will replace biologists. It is that AI will give biologists the same tools that transformed electrical engineering, software development, and chip design — simulation environments, generative design tools, and rapid prototyping capabilities that turn open-ended exploration into iterative optimization.

Whether biology's intrinsic messiness — its mutations, its context-dependence, its emergent behaviors — will yield to this vision remains the central unanswered question. But the resources being deployed to test the hypothesis are unprecedented, and the early returns, while preliminary, are not discouraging.

As Hassabis put it: the goal is not to make biology simple. It is to make the tools powerful enough that complexity becomes tractable [3]. Jensen Huang is betting tens of billions that the tools are nearly ready.