OpenAI for Startups is OpenAI’s programme designed to help early-stage companies build and scale products using AI. Rather than just offering model access, it focuses on removing practical barriers for founders by combining technical support, resources, and credits to accelerate product development.
In practice, that means pairing access to OpenAI’s models with people, infrastructure, and hands-on support designed to help teams move faster from prototype to production.
Startups in the programme can receive OpenAI API credits, higher rate limits, and access to hands-on guidance from OpenAI’s technical teams. Those backed by participating VC partners can unlock additional benefits, including enhanced support and invitations to founder-focused events.
I spoke with Mark Minera, Head of Startups at OpenAI, and Romain Huet, Head of Developer Experience, at Slush in Helsinki, to learn more about their support for startups and the biggest trends.
Mark Minera, who leads startups at OpenAI, heads a global team working closely with companies building on top of the OpenAI platform. VC partnerships are an extension of startup support.
The team also collaborates extensively with venture capital funds through a dedicated VC partnerships function, focused on providing the right resources and hands-on support to the startups they back
OpenAI also hosts events and VC summits in places like London, San Francisco, where it brings together its leadership, product teams, and startup teams to share what it's building, what patterns it's seeing, and to hear feedback directly. The focus, Minera says, is on giving founders practical leverage — from infrastructure and credits to direct time with OpenAI’s solutions and engineering teams.
"We focus on resourcing portfolio companies — credits, technical support, and access to our solutions and engineering teams.”
According to Minera, VCs are understandably most interested in roadmaps but also assurance that OpenAI is closely partnered with their scaling companies.
“There’s a lot of interest in understanding what we’re seeing, what we’re building, and we try to share those insights with VCs so they can better support their companies,” shared Minera.
“Increasingly, they’re interested in ChatGPT as a distribution channel. With hundreds of millions of weekly users, there’s real interest in embedding startup experiences directly into that ecosystem. Commerce is another emerging area of interest.”
Overall, the company’s work with VCs and startups shapes how OpenAI evaluates startups
OpenAI’s litmus test for startups
Despite OpenAI’s scale and resources, Minera is quick to stress that the startup-facing team itself is still small. He admits that while the company is big in terms of funding,“We still feel like babies.”
“My team is about 45 people globally. I’m based in San Francisco. We started there, but now we’re in Europe and Asia as well.”
For Minera, a great startup, from our perspective, is pushing the frontier of how they’re using OpenAI’s models to build product.
“We work with what we call AI-native companies — where there’s an LLM at the core of the product. If you took it out, the product wouldn’t work anymore. That’s the litmus test.
Within that, we want companies that are really operating at the bleeding edge. We work across many verticals — coding, customer support, legal tech, healthcare — and also newer companies taking entirely different approaches.”
Being at that edge provides a feedback loop. Startups help the team understand how models need to improve to support specific tasks such as legal workflows, live conversation, or sales automation.
“That feedback accelerates our own pace of development.”
In return, model improvements are usually very closely tied to product-market fit for those companies.
“It becomes self-reinforcing,” shared Minera.
As more startups build on the same underlying models, Huet argues that defensibility — not just technical capability — has become the defining challenge for founders.
The real moat in AI startups: deep problem understanding
When it comes to defensibility and category leadership, Huet argues that access to powerful models is no longer enough. He contends that in this wave of startups, there are a lot of competitors emerging in the same categories. “ Legal tech is a good example — I could name a dozen companies off the top of my head.”
So what differentiates them? Part of it is product.
“There’s real skill in designing a great product, even without AI. User experience matters enormously," explained Huet.
“With AI specifically, how you use the models matters a lot. Some teams have a much deeper understanding of how models are built — how to prompt better, how to provide context, what’s in distribution and what isn’t. That AI engineering sophistication really shows.”
Speed still matters.
As Minera puts it:
“There’s a joke that speed is the only moat in the application layer right now — and there’s some truth to it. Teams that ship fast, get in front of customers quickly, and react in real time have a genuine edge.”
Romain Huet agrees, but argues that speed alone is no longer a differentiator.
“Speed has almost become table stakes. Builders can now go from an idea to a working feature incredibly fast,” he says.
"What really matters is an obsession with the problem you’re solving. Unless you’ve spent dozens — even hundreds — of hours deeply understanding a customer's pain point, it’s extremely hard to solve it well, even with AI.”
The strongest founders, Huet adds, are those who combine sharp AI intuition with deep customer obsession — using speed not as a shortcut, but as a force multiplier.
The importance of understanding LLMs
In terms of the teams that break through, according to Minera, most of the teams OpenAI sees doing really well have very strong engineering backgrounds that sometimes border on research.
“They’re not doing foundation model research, but they understand how models work. There’s a new skill set here that’s different from building a web app.
Some teams experiment with fine-tuning — not as a first step, but when it makes sense. That requires understanding data composition, overfitting, and evaluation. That’s a different discipline than hooking up a database to compute.”
How startups shape OpenAI’s roadmap
Startups play a critical role in OpenAI’s feedback loop. “Startups often provide reproducible examples that our research teams can investigate and build evaluations around,” says Minera.
“There’s an old adage in programming languages: the ones people complain about are the ones people use. Models are similar. Even very successful companies can point to many things they want improved.”
That feedback feeds directly into OpenAI’s research priorities, which span everything from highly technical issues — such as improving tool-calling accuracy for AI agents — to more visible product capabilities. One area of sustained investment is coding.
“Coding isn’t one thing,” Minera explains. “It includes code review, generation, schema adherence, language specificity, and more. We’re constantly iterating across all of those dimensions.”
Romain Huet notes how quickly the role of AI in software development has evolved.“Coding has changed dramatically in the last few months. Where models once helped with snippets or light tasks, they now function more like teammates — taking on large, complex work for hours and returning complete outputs,” he says.
“That’s why we’re continuing to release models optimised specifically for coding.”
According to Minera, “This cycle is different from previous tech waves. When we release something new, it can materially change a startup’s roadmap or an investor’s thesis. So those conversations are critical.”
Why pivots have become easier
I’ve seen more startup pivots in the last 18 months than ever before. Why now? According to Huet, pivoting used to be extremely costly — six to twelve months of runway. Now, with AI tools, teams can test new directions in days or weeks:
“Founders can explore new customer segments or problems very quickly. Accelerator teams are pivoting multiple times in a short period, which would have been unthinkable a few years ago.”
Part of pivoting is open expanding to new markets. In terms of emerging areas, Huet asserts that multimodality and speech-to-speech are still underused.
“The quality is now there, and pricing has dropped enough for startups to build viable products. As AI moves into the physical world — robots, devices, hardware — voice will likely become the primary interface.”
OpenAI is learning from Europe
According to Minera, some of the most exciting startups we work with are coming out of Europe.
“The ecosystem is vibrant, funding is increasing, and many European companies are category leaders. Several European startups have taught us the most about how our models are used in production.”
Huet revealed:
“I built my first company in Paris 17 years ago, when the startup ecosystem barely existed. Today, Europe has talent, capital, and experience. What excites me isn’t comparing valuations to the US — it’s the trajectory. Five years ago, many of these companies couldn’t have been built here. Now they can.”
Ultimately, Minera wants startups to know “We’re here to work with startups, and we want feedback. Benchmarks matter, but what matters more is how models perform in real products. That’s how we learn and improve.” Huet stresses,
"The pace of change isn’t slowing down — it’s accelerating. Founders who stay curious and master the tools will have a real edge. And we’re happy to help.”
Would you like to write the first comment?
Login to post comments