Greg Brockman on Navier-Stokes, Astra's breakthrough in computer use, ChatGPT health at 300M weekly users, and OpenAI's startup-mode refocus

Greg Brockman on Navier-Stokes, Astra, and OpenAI's refocus

Greg Brockman, OpenAI's co-founder and president, covers a lot of ground here: a claimed breakthrough in pure mathematics, a computer-use milestone that generated significant weekend buzz, health as OpenAI's most consequential consumer vertical, and a frank admission that the company spent part of last year operating like a large incumbent before snapping back into startup discipline.

Navier-Stokes

OpenAI says one of its models has solved the Navier-Stokes problem, finding a counter-example that demonstrates the equations break down under certain conditions. Navier-Stokes is one of the seven Millennium Prize problems, math's canonical list of deep unsolved questions. Brockman is careful to separate the specific result from the broader signal: the fluid-dynamics applications matter, but what matters more to him is that the model generated genuinely new knowledge rather than retrieving or recombining existing answers. Whether that distinction holds up to peer scrutiny is an open question the transcript doesn't settle.

“Today, we announced that our model had solved the Navier-Stokes problem... We're now well over a billion users every week. We've had this like almost vertical wall of agentic adoption since we launched ChatGPT work... 300,000,000 people every single week with health queries.”

Astra and computer use

Brockman says Astra represents a genuine threshold in computer use, not an incremental improvement on what came before. The weekend reception appeared to validate that claim, with users building 3D models of physical spaces, designing mechanical parts, and iterating on furniture and renovation concepts inside Blender at a scale OpenAI hadn't seen before.

The lineage runs through Operator, the earlier cloud-based computer-use product Brockman describes as simply below threshold — slow, inaccurate, and painful enough that most users couldn't extract real value. Astra is the same long-term bet, executed after the team spent this year working through a long list of capability gaps. The core argument is that every application on a computer is designed for humans, so an AI that can operate computers the way humans do unlocks everything simultaneously, rather than requiring connectors painstakingly coded for each integration.

ChatGPT at scale

OpenAI is now past 1 billion weekly active users across ChatGPT, with 300 million people using it weekly for health queries alone. Brockman says market share is climbing again after a period where the company was investing heavily in education, health, and other verticals that moved quietly. The agentic adoption curve since ChatGPT Work launched has been, in his words, "almost a vertical wall."

The discovery problem sits alongside that growth. With chat and agentic work currently presented as two separate text boxes, users don't naturally understand why one is more powerful than the other or when to switch. Brockman says the answer is to let the model itself surface what it can do — proactively telling users to add a connector, authorize an action, or approach a request differently. He frames this as a large unsolved onboarding opportunity, not a product gap OpenAI has closed.

On marketing, the argument is that esoteric capability demonstrations don't transfer. Showing someone a candle-holder sketch that becomes a 3D render that becomes a physical object works because the viewer immediately wants that specific outcome. Showing a grand scientific capability doesn't produce the same "I want to try that" response.

Health

The three-pillar structure Brockman describes: a consumer product already at 300 million weekly health users, a clinician-facing tool with direct citations to medical literature, and an enterprise offering sold directly to hospitals, including an integration with Epic. He argues the compounding value comes when all three scale together — shared data, continuity across providers, and the ability to connect symptoms that specialists working in silos would never link.

Brockman's example is personal: his wife spent five years seeing multiple specialists, each treating a separate symptom, until an allergist finally recognized a single genetic condition underlying all of them. He frames that diagnostic failure as a systemic problem that an AI with cross-domain memory and persistent context is structurally better placed to solve than a fragmented specialist network.

Startup refocus

The most pointed observation comes from outside Brockman's direct answers: that OpenAI spent much of last year shipping like a hyperscaler — tolerating low hit rates across many experimental products — and has spent this year operating like a startup again, with the full organization aligned on a narrower set of priorities. Brockman doesn't push back on that framing. He credits the shift for the momentum behind Astra and says the company is "laser focused" in a way that brings individual efforts together into something coherent.

The test is whether that focus holds at billion-user scale, with health, agentic work, images, voice, and coding all being positioned as a single converging platform rather than a portfolio of separate bets.