The AI race in pharma has already been decided at the top.
In October 2025, Eli Lilly announced a one-billion-dollar joint lab with Nvidia to build LillyPod, what the company describes as the most powerful supercomputer owned and operated by any pharmaceutical company in the world. Johnson & Johnson signed a parallel partnership with Nvidia. Novo Nordisk built a deep relationship with Anthropic and Amazon Web Services. Pfizer has been operating its PACT collaboration with AWS since 2021. Roche has standing AI research collaborations with Google. The four largest AI infrastructure providers in the world now have direct, named, multi-year relationships with the largest pharma companies in the world.
Eli Lilly went from fourteenth to first in CB Insights' pharma AI readiness ranking in two years, made thirteen separate AI investments, and is now functionally an AI-native company sitting inside a pharma wrapper. Industry-level pharma AI spend reached four billion dollars in 2025 and is projected to reach twenty-five point seven billion dollars by 2029.
That is the table the AI race is being played at.
Mid-size pharma is not at that table. The question is whether mid-size pharma can stay in the game at all.
What Mid-Size Pharma Is Actually Competing Against
Look at the asymmetry directly.
Johnson & Johnson's 2025 R&D budget was approximately fourteen point sixty-six billion dollars. AstraZeneca's was fourteen point twenty-three billion. Eli Lilly's was thirteen point thirty-four billion. Novartis spent eleven point two billion. Bristol Myers Squibb spent nine point ninety-five billion. AbbVie spent nine point one billion.
The single one-billion-dollar Lilly-Nvidia AI lab is larger than the entire annual R&D budget of every mid-size specialty pharma in Europe, the Middle East, and most of Asia.
This is the size of bet that has already been placed. It is not a forecast. It is a deployed commitment.
A mid-size pharma with, for example, one and a half billion dollars in annual revenue, eight percent of revenue allocated to R&D, and a strong digital transformation appetite is looking at an absolute ceiling of around one hundred and twenty million dollars in annual R&D, of which AI might receive somewhere between five and fifteen million dollars in dedicated infrastructure and tooling spend. The same amount Eli Lilly spends on its AI program every working day.
The asymmetry is not marginal. It is generational.
The Four Structural Gaps
The capital gap is the loudest, but it is not the only one. Mid-size pharma faces four structural gaps at the same time.
The capital gap is direct. Building AI infrastructure that competes with LillyPod or with the J&J-Nvidia stack requires billions, not millions. Mid-size pharma cannot fund it.
The talent gap is more painful than it looks. A GlobalData survey published in November 2024 found that forty-nine percent of pharma professionals named lack of specific skills and talent as the single biggest obstacle to digital transformation in the industry. Big pharma is now bidding for the same AI talent as the hyperscalers, the AI-native biotech startups, and the financial services sector. Mid-size pharma cannot win that bidding war. The senior data scientist who would have joined a mid-cap pharma in 2018 now has six better-paid offers, three of which come with equity in an AI-native company.
The data gap is structural. Big pharma sits on decades of proprietary clinical trial data, real-world evidence, commercial CRM data, and pharmacovigilance archives. Lilly, Novartis, Roche, Pfizer, and J&J each have trillions of structured data points to train proprietary models on. Mid-size pharma has, for the most part, smaller datasets, fewer therapeutic areas, and less infrastructure to make the data they do have usable for AI. A model trained on smaller, narrower, less curated data underperforms. The performance gap compounds over time.
The regulatory overhead gap is rarely named, but it bites mid-size pharma hardest. The EU AI Act, the FDA-EMA joint AI principles, GxP validation requirements for AI in regulated workflows, and the conformity assessment regime through notified bodies all apply equally to a five-billion-dollar specialty pharma and to a hundred-billion-dollar global. The fixed cost of building a compliant, auditable AI operating model is roughly the same. Spread over a smaller revenue base, that cost is brutal.
These four gaps compound. They are not independent failures that can be fixed one at a time. They are reinforcing.
The Build, Buy, or Partner Dilemma
Faced with these gaps, mid-size pharma has three options on paper. In practice, two of them are closed.
Building in-house AI infrastructure is closed. The capital, talent, and data are not there. Trying anyway produces multi-year programs that consume budget without delivering production AI, and they get cancelled in the third year when a new CFO arrives and asks where the return is.
Full-stack buying from enterprise vendors is functionally closed. The IQVIA-class platforms, the Veeva-class CRMs, the ArisGlobal-class pharmacovigilance systems, and the major medical writing and content ops platforms are all priced for enterprise scale, with multi-year contract terms, dedicated implementation teams, and customization that mid-size pharma neither needs nor can fully absorb. Specialist commentary on the mid-market data infrastructure landscape has openly named this gap: healthcare data tooling was built for companies with enterprise budgets and dedicated data science teams to extract value from it, and mid-market pharma is structurally underserved on price, contract terms, and fit.
That leaves partner. And partner is the only path that actually works at mid-size, but only if the partnering is unusually disciplined.
The Escape Path That Actually Works
Three operating moves separate the mid-size pharma that survives the AI race from the one that fades into the background.
First, narrow the AI bet. Big pharma is investing across drug discovery, clinical trials, manufacturing, commercial, medical affairs, pharmacovigilance, and corporate functions in parallel. Mid-size pharma cannot do that. The companies that survive will pick two or three functions where their domain depth gives them a defensible AI advantage, and they will go deep on those. Pharmacovigilance, field force enablement, medical content ops, and HCP engagement are the most natural starting points because the volume math forces AI and the deployment surface is contained. Drug discovery AI is, for most mid-size pharma, a trap. It is the headline area where they will be most outspent by Lilly and most outcompeted by AI-native biotech.
Second, partner with AI-native vendors aggressively, not with hyperscalers directly. The hyperscaler partnerships make sense for Lilly and J&J at billion-dollar scale because they are buying compute, foundational model access, and joint engineering at a level mid-size pharma cannot consume. The right partners for mid-size pharma are the AI-native vendors that have already built specialized stacks for one or two pharma functions, and that price for the mid-market. The right contract terms are short, modular, and exit-friendly, because the vendor landscape is consolidating fast and the wrong long-term commitment is more dangerous than no commitment.
Third, treat regional and specialty data as a strategic asset. Mid-size pharma typically operates in a smaller number of therapeutic areas, in a smaller number of markets, with deeper specialty depth than the global majors can match. A regional specialty pharma in the EMEA or APAC region knows its KOLs, its prescribers, its payers, and its patient pathways at a level that big pharma's centralized commercial machines cannot replicate. AI applied narrowly to that proprietary regional knowledge is a real edge. AI applied to the same generic enterprise use cases that the top twenty are already automating is not.
Fourth, accept that the operating model has to change. Mid-size pharma cannot build a separate AI organization sitting next to the traditional functions. The headcount is not there. AI capability has to be embedded into the medical, commercial, and PV functions directly, with a small central enablement team rather than a parallel structure. The pharma companies running this model are deploying AI faster, with less governance overhead, and with better adoption than the ones trying to imitate the big pharma operating model at one-tenth the scale.
Fifth, move while big pharma is still slow. The largest pharma companies are extraordinarily resourced, but they are also extraordinarily bureaucratic. A mid-size pharma that picks the right narrow battles can deploy production AI in six to twelve months on the same use cases where big pharma is still in committee a year later. That is the only window in which size disadvantage flips into speed advantage. Once the big pharma operating models harden, the window closes.
The Default Outcome
Without these moves, the default outcome for mid-size pharma is straightforward and uncomfortable.
Big pharma deploys AI across drug discovery, clinical trials, commercial, and medical affairs at a scale and pace that mid-size pharma cannot match. Cost per HCP touchpoint falls. Speed to launch shortens. Pharmacovigilance becomes structurally cheaper. Commercial productivity rises. The cost-of-goods curve flattens for the top twenty while remaining stubbornly high for everyone below.
At the same time, AI-native biotech and specialty companies attack the same therapeutic areas mid-size pharma occupies, with much smaller teams, lower fixed costs, and AI built into the foundation rather than retrofitted onto legacy infrastructure.
Mid-size pharma gets squeezed from above and below at the same time. Margins compress. M&A activity rises. The companies that retain independence five years from now will be the ones that made deliberate, narrow, well-partnered, regionally-anchored AI bets early enough to be operationally distinctive by the time the squeeze tightened.
The companies that did not will be acquired, divested, or quietly absorbed.
The Real Question
The AI race in pharma was never going to be won by the company with the most data scientists. It was always going to be won by the company that deployed AI fastest into the workflows that actually move the business.
Mid-size pharma is genuinely outmatched on capital, talent, and data. It is also genuinely advantaged on focus, speed, and regional depth, in any boardroom that has the discipline to pick the right battles.
The default path is losing. The escape path is real. It is narrow, specific, and visible. Almost no mid-size pharma is on it yet.
The pharma companies that get on it in the next twelve months will be the ones still standing when the dust settles.
The rest will be footnotes in someone else's annual report.
