CH Health Tech Advisory

21 July 2026 · 2 min read

The Pharma Compute Race: Three NVIDIA Deals in Nine Months

Bristol Myers Squibb just claimed the most powerful AI supercomputer in life sciences—the third pharma company to do so in nine months—and I think the real signal isn't the superlative, it's the demand curve behind it.

Bristol Myers Squibb just claimed the most powerful AI supercomputer in life sciences. It is the third pharma company to claim that title in nine months.

Eli Lilly and Company announced a $1B NVIDIA co-innovation lab at JPM in January. Roche followed with a hybrid-cloud AI factory running 2,176 Blackwell GPUs. Yesterday BMS announced its second NVIDIA DGX SuperPOD, built on eight Vera Rubin NVL72 systems, a more than tenfold increase in its AI compute capacity. I wrote about both deals when they broke, and commented (rather skeptically) on this on Swiss news.

The superlative is marketing. The demand curve behind it is the signal.

Greg Meyers, Chief Digital and Technology Officer at BMS, gave the most honest line of the announcement: the company consumed all the compute it had. Three years ago the first cluster ran single-purpose tools like protein structure prediction. Today it trains foundation models on proprietary data and powers agentic workflows across oncology, hematology, cardiovascular, immunology, and neuroscience.

Compute consumption is revealed preference. Any company can publish an AI strategy. Very few outgrow their infrastructure because scientists keep submitting jobs. When your CDTO justifies a capital investment by pointing at a full queue and an electric bill, the internal demand is real.

Chief Research Officer Robert Plenge put numbers on it: early screening capacity moves from roughly 10 candidates to dozens, and AI has already cut the time to produce medicines for clinical testing by 20 to 30 percent, with 50 percent possible in the coming years.

Three operating models are now visible. Lilly co-locates with NVIDIA in a shared lab. Roche splits its footprint across hybrid cloud. BMS owns its infrastructure outright and trains on proprietary data behind its own walls. For an industry built on data exclusivity, watch how quickly the third model becomes the default.

The title will change hands again before year end. The pharma compute race is on.