Nvidia Pushes Agentic AI Into Pharma as Chinese Startups Retreat From Open Source

Nvidia used its annual GTC conference to declare that agentic AI—systems capable of autonomous decision-making toward specific goals—has reached an inflection point in healthcare and life sciences, with pharmaceutical giants now racing to deploy the technology for computational drug discovery at production scale.

"This is the first time in history that every one of these companies need compute—lots and lots of it," Nvidia CEO Jensen Huang said during his opening keynote, framing the moment as a major platform shift building on generative AI tools like ChatGPT and reasoning models such as OpenAI's o1.

Kimberly Powell, Nvidia's vice president of healthcare, described the current moment as "the transformer moment for biology and drug discovery," signaling the company's belief that the same neural network architecture that revolutionized language processing is now poised to reshape pharmaceutical research.

The conference spotlighted enterprise deployments moving beyond experimentation. Persistent Systems announced it is deploying Nvidia's full AI stack—including the BioNeMo platform for digital biology and NIM microservices—to enable biopharma clients to compress drug discovery timelines from months to days through its GenMolVS solution for generative molecules and virtual screening.

"Healthcare and Life Sciences organizations need to discover new therapies faster, but traditional R&D is too slow and labor-intensive," said Ganesh Nathella, executive vice president at Persistent Systems. "By combining our GenMolVS solution with NVIDIA's full-stack AI platform, we enable BioPharma clients to use generative molecules and virtual screening in production."

Nvidia also announced NemoClaw, an enterprise-optimized version of the open-source OpenClaw autonomous agent that adds privacy and security controls for organizational deployment at scale. The move extends Nvidia's strategy of offering both open frameworks and commercial-grade implementations.

(The GTC announcements come as Nvidia expands its physical AI initiatives through partnerships with robotics leaders including ABB, FANUC, and Universal Robots, integrating its Isaac and GR00T frameworks into Hugging Face's LeRobot open-source platform to connect 2 million robotics developers with 13 million AI builders.)

The healthcare AI push unfolds against a shifting global landscape. Chinese AI startup MiniMax, previously a standard-bearer for open-source frontier models, released its proprietary M2.7 model with self-evolution capabilities, joining compatriot z.ai in abandoning the open-source approach that made Chinese firms appealing to cost-conscious enterprises. Rumors suggest Alibaba's Qwen team is also pivoting to proprietary development following senior leadership departures.

Meanwhile, Spanish startup Multiverse Computing is pursuing the opposite strategy, launching compressed versions of models from OpenAI, Meta, DeepSeek, and Mistral AI through its CompactifAI app and API portal. The company claims its HyperNova 60B model, derived from OpenAI's gpt-oss-120b, delivers faster responses at lower cost—particularly valuable for agentic coding workflows where AI autonomously completes multi-step programming tasks.

The divergence reflects competing visions for AI's commercial future: Chinese firms increasingly betting on proprietary models to capture value, while efficiency-focused players like Multiverse see opportunity in making powerful models accessible through compression and local deployment for privacy-sensitive industries.