Eli Lilly has brought its LillyPod supercomputer online, Roche has deployed NVIDIA-based AI computing infrastructure, and Bristol Myers Squibb has announced plans to expand its existing compute footprint — three concrete signals of how the AI spending boom, which has continued mostly unabated through 2026 so far, is reshaping what innovation looks like at the world's largest R&D spenders.

These moves surface in this year's ranking of the 100 most innovative companies of 2026, a list that spans NVIDIA and OpenAI on the technology side through Eli Lilly and SpaceX. The pattern behind the names is consistent: organizations that kept pouring capital into computing capacity during a period when many budgets tightened are the ones now moving up.

For R&D managers, the pharma examples carry the most immediate operational weight. Lilly's LillyPod supercomputer is not a branding exercise; it is dedicated infrastructure aimed at accelerating drug discovery workloads, from molecular simulation to the training of models that prioritize discovery pipelines. Roche's deployment of NVIDIA-based AI computing follows the same logic — the company is buying compute to shorten the distance between hypothesis and candidate.

Bristol Myers Squibb's announced expansion is the third data point, and in some ways the most telling. An expansion decision, rather than a completed deployment, signals that at least one major pharma R&D organization has reviewed its internal capacity forecasts and concluded that current infrastructure will not meet projected demand. Portfolio planners at competing organizations will likely face the same question this budget cycle: build, buy, or rent.

The spending boom itself has run mostly unabated so far in 2026, according to the ranking's analysis. That persistence matters for lab budgeting. Two years ago, AI infrastructure purchases could be treated as experimental line items. The 2026 cohort treats them as core R&D infrastructure — closer in kind to a new screening facility than to a software subscription, with corresponding capital commitments, depreciation schedules and facility requirements.

The presence of NVIDIA and OpenAI at the top of the list frames the supply side of this dynamic. NVIDIA supplies the hardware layer — the GPUs and accelerated-computing stacks — that Roche and others now deploy. OpenAI represents the model layer. What the rankings capture is the third layer: the organizations converting that computational supply into pipeline output, clinical candidates and, eventually, products.

Eli Lilly's position on the list alongside SpaceX also marks a broader shift in how innovation gets measured. SpaceX's reuse-driven launch economics and Lilly's compute-driven discovery programs share a trait R&D evaluators increasingly reward: repeated, measurable reductions in the cost of iteration. A supercomputer that lets discovery teams run more hypotheses per quarter does for pharma what a reflown booster does for payload economics.

The caveats deserve attention. Rankings of this kind aggregate many inputs, and the source material does not detail the specific metrics behind each placement, the funding sources behind the underlying assessments, or the sample of companies evaluated. The measured facts — LillyPod online, Roche's NVIDIA deployment, BMS's expansion plans — stand on their own. Where each company's infrastructure investment lands on its pipeline, in the form of candidates advanced or trials accelerated, remains a projection until the data appear.

What is measured, and what is not, also separates the pharma stories from the platform stories. NVIDIA's and OpenAI's placements rest on shipped products and deployed models. Lilly's and Roche's rest substantially on infrastructure now in place whose scientific returns will take years to quantify. BMS's rests on an announced intention. R&D leaders reading the list should price those differences accordingly when benchmarking their own organizations against it.

The practical takeaway for research organizations is straightforward. The 2026 rankings document that the largest R&D spenders have moved AI compute from the innovation periphery to the operational core, with named deployments at three major pharmaceutical companies inside a single cycle. Organizations that have not yet sized their own computational requirements against their discovery ambitions now have a cohort of peers — with supercomputers online and expansions announced — against which that gap is measurable. Expect the compute-to-pipeline conversion rate, rather than compute ownership alone, to become the differentiating metric in next year's rankings.