After the upheaval, how is OpenAI faring now? What is its biggest revenue driver?
TechFlow Selected TechFlow Selected
After the upheaval, how is OpenAI faring now? What is its biggest revenue driver?
This interview combines CNBC's conversation with COO Brian Lightcap following the Dev Day event, as well as discussions after last week's leadership changes.
Navigating Web3 tides with focused insightsTranslated by: Alpha Rabbit
This article combines CNBC’s interview with COO Brad Lightcap following the Dev Day event and OpenAI’s internal communications after recent leadership changes. It addresses several key questions: What is the current state of the company? What are its primary revenue drivers today? What aspects of AI are currently most overhyped and most underestimated? What does this technology truly mean for businesses?
A few weeks before OpenAI launched ChatGPT in November 2022, the executive team spent significant time discussing one critical question: Should they release this tool at all?
If you know Sam Altman, you know he prefers fast, efficient discussions—so spending so much time on a single topic signaled that it was deeply important. No one could be 100% certain whether releasing it was the right move or worth the investment of time.
Brad Lightcap recalled that at the time, OpenAI had very limited GPU resources and saw itself primarily as a company building tools for developers and enterprises. Yet CEO Altman strongly advocated for giving it a try, based on the belief that text-based interaction with models was fundamentally important and carried personalized elements.
This decision ultimately paid off. Data shows that ChatGPT broke records as the fastest-growing consumer application, now boasting around 100 million weekly active users, with over 92% of Fortune 500 companies using ChatGPT. According to PitchBook, Microsoft made an additional $10 billion investment in the company earlier this year, making it the largest single AI-focused investment of the year.
However, recent developments have overshadowed these past milestones. Last month, OpenAI’s board abruptly removed Sam Altman as CEO, triggering widespread opposition from nearly all employees—including threats to resign and a signed open letter—as well as shock among investors including Microsoft.
Within less than a week, Altman returned to the company. On Wednesday last week, OpenAI announced a new board, including former co-CEO Bret Taylor, Larry Summers, and Quora CEO Adam D’Angelo. Microsoft holds an observer seat on OpenAI’s board.
This article draws from CNBC’s interview with COO Brad Lightcap after the Dev Day event and follow-up discussions after the leadership transition.
Q: We’re approaching ChatGPT’s first anniversary. This time last year, just weeks before ChatGPT’s debut, DALL-E was still in beta, Stable Diffusion was gaining attention, and ChatGPT didn’t yet exist. What was your team like back then?
A: At the time, we saw ourselves strictly as a company building tools for developers. So the idea of creating something “any ordinary person could pick up and use” felt quite novel for OpenAI.
We’d had a similar experience with DALL-E. After people used it, we noticed how excited many users were. But we assumed DALL-E, being inherently visual, would naturally attract more attention and represent the peak of consumer interest in such tools. When we began exploring ChatGPT, we used DALL-E as a benchmark: How many people would use it? Who would care? Would ChatGPT feel like a real tool—or just a toy people played with briefly?
Back then, the OpenAI team even placed bets on how big ChatGPT might become. My own bet was that we’d see a peak of one million concurrent users at any given moment. We planned accordingly, and as a finance lead, I modeled various financial scenarios based on that assumption. That was our reality—but looking back, that estimate was far too low.
In terms of business opportunity, what did OpenAI expect from ChatGPT at the time?
A: We simply couldn’t foresee all its potential uses—that’s the paradox of this technology. ChatGPT turned out to be so broadly useful, embedding itself into every corner of the world and people’s lives, often in ways users didn’t even realize they needed.
So we tried to do early business analysis: “Okay, what will people actually use ChatGPT for? What will drive ongoing paid usage?” Then we started imagining practical applications—trying to picture people using ChatGPT for writing, or for this task or that. In hindsight, we now understand many of the ways people use it, but back then, we couldn’t possibly imagine them all, nor could we prove why ChatGPT would succeed.
Perhaps the interesting lesson here is that business analysis doesn’t always tell the full story. Sometimes, placing a bold bet—and genuinely discovering where something new resonates widely in practice—can matter more than any analytical forecast.
Q: In August, 80% of Fortune 500 companies were using ChatGPT. Now, as of November, that number has reached 92%. For the remaining 8% that haven’t adopted it, is there a trend?
A: The 8% tend to be companies in heavy industry—large, capital-intensive sectors like oil and gas, or industries with extensive physical machinery, where work revolves more around producing physical goods rather than information or services.
Q: In your view, what aspects of AI today are most overhyped and most underestimated?
A: I think the overhyped aspect is the belief that AI can instantly deliver transformative business outcomes. We’ve spoken with many companies eager to tie long-standing goals to OpenAI—like “We want to restore year-over-year revenue growth to 15%,” or “We want to cut X million dollars from this cost line.” But there’s no silver bullet—no single AI solution that magically solves complex business challenges. This underscores how vast and messy the real world is, and how these systems are still evolving—they’re truly in their infancy.
What I believe is underestimated is the level of personal empowerment and agency these systems give end users. What we hear consistently from users and customers is that tools like ChatGPT have given them superpowers—abilities they simply didn’t have before.
Q: Let’s talk about the business of generative AI. Critics say consumer apps are multiplying rapidly—could this market become oversaturated? What does this technology really mean for enterprises?
A: We’re still in the very early days of AI. It’s crucial to maintain a high rate of global experimentation and iteration. Looking back at historical tech transitions, there’s always a vital experimental phase. Mastering a new technology from scratch is extremely difficult. We’ll eventually reach the final form of this technology—converging on what works—but only after trying countless approaches, learning what fails and what succeeds, and building the next best thing on top of what proves effective.
My view is that the most important things built on this technology haven’t been created yet. It takes cycles of actual usage to fully grasp these tools’ capabilities and figure out how to combine them with other technologies to create something greater than the sum of their parts. So this explosion of experimentation is expected—and I think it’s very healthy.
Q: A few years ago, people were surprised by AI’s potential in trucking—a field seen as highly traditional. Now, AI is part of nearly every industry. Based on recent trends, is there a common thread? Has any industry adopted AI in a surprising or novel way?
A: The tech sector has shown strong appeal. We’re seeing ChatGPT serve as a powerful technical assistant—whether for software engineers, mechanical engineers, chemists, or biologists. In each discipline, there’s a vast knowledge base on the other side; your productivity depends on how well you can access and apply it.
Throughout their careers, professionals strive to master their fields and absorb as much domain knowledge as possible. Especially in areas like biology, chemistry, or AI itself, where literature constantly evolves with new discoveries and research. I’m not sure if this is the most surprising thing, but one of the coolest things we’ve seen is ChatGPT acting almost like a junior assistant—a research aide. ...The pull from these industries is something I never anticipated back in November 2022.
Q: ChatGPT Enterprise has been live for several months. I remember you launched ChatGPT in under a year and had over 20 companies (like Block and Canva) in the pilot program. Specifically, how has usage grown? Since launch, who has been your biggest customer?
A: Enthusiasm has been overwhelming. But we remain a small team, and much of our focus over the past two months has been ensuring the first wave of clients see real value. ...We’re still working through thousands on the waitlist. Reaching everyone is our goal for 2024.
Q: With ChatGPT Enterprise now available, what is currently OpenAI’s biggest revenue driver? How do you expect this to evolve?
A: OpenAI almost never takes a revenue-first approach to product development. We always take a product-first approach—meaning we prioritize building products that excel in one of two areas: 1. Being genuinely useful tools for developers to build with, or 2. Providing valuable abstractions that help users extract more value from the product.
For example, consider GPT—it almost perfectly maps to the second category: it’s a way to abstract the intelligence within ChatGPT and direct it toward specific tasks, giving it the right context, tools, and connections so it can truly excel at solving particular problems. Whether that’s something useful for your job, your life, or just something fun.
Q: Multimodal ChatGPT—offering image generation and other tools within the same service—is a major priority outlined at Dev Day. Why is this so important?
A: The world is multimodal. Think about how humans perceive and interact with the world—we see, hear, and speak. The world is far richer than text alone. So relying solely on text and code as the single mode—the only interface—to understand how powerful these models are and what they can do has always felt incomplete.
That’s why we began layering in visual capabilities. The fact that computers can now see what’s happening in the world, describe it, engage with it, and reason about it may be the most astonishing thing I’ve witnessed in my five years at OpenAI. I still can’t fully grasp its implications. But when you pause to think, things that were once impossible are now becoming feasible.
Consider simple applications—like helping visually impaired individuals better understand their surroundings with low latency and high quality. Or companies gaining deeper insights into their equipment and creating consumer experiences where pointing a camera reveals how something works. In education, we can help people better understand and analyze concepts—many of whom are visual learners—by presenting information in ways that match their preferred learning styles. This unlocks a completely new set of capabilities.
What excites me is that this gives us a way to use technology that aligns more closely with how humans interact with the world—and ultimately makes technology more human-centered.
Q: We know OpenAI’s GPT-4 large language model may be more reliable than GPT-3.5 but also potentially more prone to bias. Can you explain how the new Turbo model announced at Dev Day differs, if at all, and what your mitigation plans are?
A: We’ll release a Turbo model card [a transparency tool for AI models], which will likely be a better reference point for technical benchmarks.
Q: What are your biggest hopes for the coming year? What capabilities might future versions of GPT achieve that current versions cannot?
A: Progress is advancing along the axis of reasoning quality. Fundamentally, what humans do well is combine diverse concepts creatively to produce meaningful outputs for tasks we face or are asked to perform. We do this daily—it’s an art form and the foundation of how we shape the world.
That’s the direction we want the technology to go—significantly enhanced reasoning; handling increasingly complex tasks by breaking them down into necessary components and executing them with high proficiency; and doing all this safely alongside those tasks. From a research perspective, we emphasize getting safety right. As systems grow more capable, we must raise safety standards in parallel, because these systems will become increasingly autonomous over time. If we don’t advance safety equally, it won’t work.
Q: Over the past year, which day stood out to you the most?
A: The day we launched GPT-4. People may not realize how long we had worked on it beforehand. Internally, there was immense excitement—we knew this would mark a true shift in model capabilities and redefine what people considered a high-quality language model. Once we had it, we couldn’t wait to share it with the world. Our team draws tremendous energy from how people respond. Seeing our customers, developers, and users engage and get excited is incredibly motivating.
For the seven or eight months leading up to it, we knew this moment was coming, so anticipation had been building...
Back then, we didn’t hold large launch events like Dev Day. Right after the release, there was a moment—I think we were all gathered in the café space—and everyone looked around, sharing a mix of excitement, relief, and exhaustion—but everyone was smiling. It was something truly special… moments like that are rare.
Q: When you went home, what did you personally do to celebrate?
A: I celebrated by working late into the night.
Q: In under ten years at OpenAI, people have seen it evolve from a nonprofit into a “research and model deployment” company. Questions arise about what this means, your organizational structure, and Microsoft’s stake. Can you shed light on this journey?
A: At a high level, we wanted the core structure of the company to preserve the original OpenAI (the nonprofit). When we formed the company, the challenge was figuring out how to achieve that. That was actually my initial role at OpenAI: to determine whether there was a way to place OpenAI’s mission—and the nonprofit embodying that mission—at the center of our new structure.
So I think this is the first thing to understand about OpenAI: in this sense, it’s not a typical company. Literally, structurally, and spiritually, it’s an extension of the nonprofit mission. Its primary duty is to carry out the nonprofit’s mission: to build safe, broadly beneficial artificial general intelligence. Granted, this may sound crazy—there are certainly simpler structures and technical approaches to forming a company, ones with lower legal costs and fewer complications—but for us, getting this right mattered deeply. I don’t know if we succeeded; time will tell. One benefit is that the structure is highly adaptable. As we learn more and the world changes, we can ensure the structure remains aligned with success. But at its core, we want to preserve OpenAI’s founding mission as the company’s very reason for existing.
Q: Talk about Microsoft’s ownership?
A: I won’t comment on specific structural details, but the setup is designed for collaboration with the world, and Microsoft happens to be a strong partner. We continually think about how this structure can extend outward, interact with the world, and stay aligned with the nonprofit’s mission. I think this is also part of the rationale behind the profit-cap model.
Q: You’ve worked with Sam Altman since OpenAI’s early days. What are the main differences in how you operate? How do you complement each other’s strengths and weaknesses?
A: Sam moves incredibly fast. One thing we share is a love for speed across everything we do.
Where I think we balance each other is this: Sam is absolutely future-oriented—he lives in the future. And frankly, he should. He excels at that. My role is to ensure the way we build the company, run operations, and establish partnerships with customers and partners reflects not only where we believe the world is headed over the next five-plus years, but also delivers what we need to achieve today.
Our challenge is that technology evolves rapidly. So we place great emphasis on communicating to the world how to use the technology, the types of work we do—from safety to capability—the way we think about our products, and how they continuously evolve. Doing this well while moving fast on shifting ground requires strong coordination. I hope my added value lies here—focusing on doing this well, building a strong team to support it. If we get that right and stay grounded, we’ll ultimately head in the right direction.
Q: OpenAI underwent massive changes in just one week. Now Sam is back, and the new board structure is announced. What impact do you expect this to have on day-to-day operations? Are further structural changes expected in the coming months?
A: Day-to-day operations won’t change. OpenAI’s mission remains unchanged. Our focus continues to be delivering excellent research, building outstanding products, and serving our customers, users, and partners. We’ve already shared that we now have an initial board and plan to add more members over time.
Q: What’s the overall mood at the company now?
A: Over the past few weeks, the company has come together in a way that’s hard to describe. I’m deeply grateful to our team and profoundly thankful to our customers and partners for their unwavering support throughout. Their support energizes us to work even harder toward our mission. Personally, I’m extremely focused.
(Lightcap and OpenAI declined to comment further on the specifics of Altman’s removal and reinstatement.)
References:
1. https://www.cnbc.com/2023/12/04/openai-coo-brad-lightcap-interview-with-cnbc.html
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
