Source: Stripe
On April 30, 2026, OpenAI CEO Sam Altman took the stage at Stripe's annual conference for an in-depth fireside chat with Stripe CEO Patrick Collison.
The two have known each other for nearly two decades, and their conversation covered the inflection point in AI development, OpenAI's management philosophy, changes in the startup ecosystem, and the profound impact of AI on science and humanity's future.
During the conversation, Altman dropped a series of heavyweight viewpoints:
-
We are indeed in the midst of some kind of takeoff. AI is developing very fast; it's a little different every week.
-
OpenAI has undergone three evolutions: from a research lab, to a product company, and now to a large-scale token factory.
-
The revenge of the "Ideas People" is here: I am now willing to invest in people who deeply understand user needs and have product insight but can't write code at all.
-
What excites me most about AI is not products, nor business models, but the potential to accelerate scientific discovery.
We Are in the Midst of Takeoff
When does the Singularity begin?
Patrick Collison offered an interesting perspective in his opening: treat this year as the first year of the Singularity.
In response, Sam Altman said, "We are indeed in the midst of some kind of takeoff." From the second half of last year to the beginning of this year, AI model capabilities broke through a certain tipping point—especially in the field of code generation.
"It's a little different every week, things are moving very fast."
A Shift in Perception
Currently, Codex (OpenAI's programming model branch) is having its "moment."
While the most dedicated users are still programmers, a large number of users without programming backgrounds are also flocking to it, trying to use it to handle all their daily computer-based work.
Altman believes people will undergo a more widespread cognitive shift: realizing how much of their time is wasted on computer drudgery.
Switching messaging apps, copying and pasting content, handling repetitive emails that could obviously be automated—these fragmented tasks are quietly eroding people's focus and work experience. When most people truly realize AI can do this "grunt work" for them, the feeling will be revolutionary.
Who's Really Using AI Well?
It Only Counts if the CEO Gets Hands-On
After observing many enterprise clients, Altman concluded: The companies most successfully applying AI often share a common characteristic—the CEO personally gets hands-on.
Not symbolically announcing "we will embrace AI," but actually building automated workflows themselves, then demanding the team follow. He gave the example of Shopify's CEO: he was one of the earliest hands-on CEOs Altman had seen, directly driving the entire company to integrate AI into all processes.
OpenAI is now trying a new experiment: sending an engineer to directly accompany a company's CEO, helping them automate as many workflows as possible.
If you can make a company's leadership truly feel the power of AI, that feeling will permeate the entire organization like a fractal.
OpenAI's Three Evolutions
Sam Altman candidly shared the management evolution inside OpenAI, which is essentially a microcosm of AI's industrialization path.
Three Stages of Transformation
The first stage, a pure research lab, aiming to figure out how to build AGI when everyone thought it was crazy.
The second stage, while continuing research, needed to learn how to be a product company.
The third stage, which is now being entered: on top of the previous two, building a large-scale token factory. Altman compared it to a new type of utility, like electricity; the world needs massive, cheap, accessible intelligence.
Vision of Low-Margin Infrastructure
Addressing doubts about "will AI giants monopolize everything," Altman used Stripe as a reference point: Stripe and its customers are highly aligned; the more Stripe earns, the better its customers do, a positive infrastructure relationship.
Altman hopes OpenAI can ultimately play a similar role: an intelligence infrastructure provider, even if perpetually low-margin, as long as it's big enough, fast enough, and deeply tied to the success of the world's distributed economy.
He also acknowledged that AI's switching costs are inherently low, making high margins difficult to sustain. The recent large-scale migration of users from competitors to Codex shows that friction for switching will become smaller in the AI era.
Compute Investment: The Most Expensive Infrastructure in History
Regarding massive compute investment, Altman stated: "This will be the most expensive infrastructure project in human history."
Efficiency gains per GPU have exceeded his expectations, but demand growth is even faster. As for how much compute needs to be built? "I don't have a good answer... in some sense, demand is almost unlimited."
OpenAI's Management Philosophy
OpenAI gathers some of the smartest and most unique individuals in the world. Altman revealed his secret is extreme conviction gambling:
-
Concentrate Resources: When training GPT-3, OpenAI threw almost all its compute resources into one project. At the time, people at DeepMind warned this would create a toxic competitive culture; OpenAI's response was: We have conviction, this is the right direction.
-
Shared Vision: Altman believes that even if team members have personal conflicts or clash, their shared belief in "Scale" allows them to sit together and solve problems.
"Directly Managing Hundreds" Communication Style
Asked about any unusual management habits, Altman mentioned: He communicates directly with hundreds of people in the company daily via Slack—not through assistants, but directly himself, with one or two short messages at a time.
This decentralized approach sometimes brings him very valuable information.
The New Paradigm for Startups
"Ideas People" Revenge
Altman developed a deep-seated bias during his YC days: disdain for entrepreneurs who "only have an idea and need a programmer to implement it," thinking it was as absurd as saying "I have a great song idea, just need someone who plays guitar to make it."
But now, "the revenge of the Ideas People has arrived."
Those who deeply understand user needs and have product insight but can't code at all can now quickly build products using AI tools. Altman said he is now very willing to invest in such people.
How to Invest Before the Singularity?
AGI might arrive in three to five years. Is the traditional ten-year investment horizon for venture capital still reasonable?
Altman's answer: Do everything on this timescale with a 'suspension of disbelief.' You can't do nothing just because "the Singularity is coming in three years, we can't see through it." You still have to act as if life will continue.
OpenAI has signed twenty-year power and land agreements but only has a clear grasp of its product roadmap for the next two years.—Make long-term infrastructure investments while staying clear about the near term; that's his answer.
AI is Reshaping Scientific Discovery
What excites Altman most about AI is not products, nor business models, but the potential to accelerate scientific discovery.
He believes this will be this technology's most profound contribution to improving human quality of life.
Tackling Complex Diseases
Through collaboration with the Arc Institute, OpenAI is supporting the use of large biological foundation models like Evo 2 to research complex diseases involving multiple genes, such as cancer and Alzheimer's.
AI is shortening research cycles that originally took ten years down to one year.
Leaps in Energy and Materials
He specifically pointed out a severely underestimated field: materials science.
AI is exceptionally good at finding optimal solutions in vast combinatorial spaces, which will bring breakthroughs in catalyst development, energy efficiency improvements, etc. He expects very rapid progress here, changing all our lives in profound ways.
👉 On energy, Altman boldly predicted: Driven by the demands of AI compute, the first profitable fusion reactor could appear within five years.
Democratization: Sam Altman's Final Stand
At the end of the conversation, Altman discussed the most controversial decision in OpenAI's history: Iterative Deployment.
He recalled that many safety experts advocated locking AI in an "ivory tower," controlled by a select few, before distributing the results to the world.
"That idea made me very uncomfortable." Altman stated: Avoiding concentration of power, making this technology truly belong to the world, is extremely important.
"People will use AI in all sorts of ways, but I believe most people are good, and most will use tools to do amazing things. I think my most important contribution is pushing for this technology to be a democratized technology, accessible and buildable by everyone."
