Bets on consumer chatbots and flashy side projects overshadowed AI coding opportunity, which rival Anthropic used to capture the lead.
By Berber Jin
OpenAI, the company that sparked the AI boom, fell behind rival Anthropic. Now, it is fighting for a comeback.
Growth of OpenAI’s flagship consumer product, ChatGPT, has slowed. The departure of Fidji Simo, Sam Altman’s heir apparent, prompted a reshuffle of responsibilities among the CEO’s top deputies. And sales staff are locked in a costly battle to win over lucrative business customers by offering volume discounts and other sweeteners.
Some of OpenAI’s largest investors have privately expressed concerns in recent months about the startup’s high cash burn relative to its growth, people involved in those conversations said, while others have hedged their bets on the company by pouring money into Anthropic.
Meanwhile, Anthropic’s revenue growth recently surpassed OpenAI’s, as did its valuation, which is now nearing $1 trillion, powered by the success of coding tool Claude Code. It is accelerating plans for a fall initial public offering, kicking off meetings with potential investors in which it has emphasized its lead over the ChatGPT maker, according to people familiar with those talks.
“We did not have our best last 12 months ever, which is mostly my fault, but we are about to have our best 12 months to date,” Altman posted on X earlier this month. “The team is doing amazing work and I think you’ll be very happy with what they’ve got cooking for you.”
OpenAI’s challenges in trying to regain its crown are rooted in an earlier misreading by company leaders of where the AI market was headed. Its predicament speaks to the intense competition that defines the race for AI supremacy.
Altman initially staked the growth of the business on ChatGPT, betting that more people would subscribe to the chatbot as AI became a bigger part of their lives. Instead, the overnight success of Claude Code made clear that the bigger prize came from selling tools to brainy software developers, and the deep-pocketed companies that employed them.
While OpenAI pursued a host of flashy projects from a video generator to consumer devices and chips, its smaller, more focused rival filled in the gap, developing a hit coding tool that helped it seize the lead.
As it chases its top competitor, OpenAI has released a series of new models focused on coding and other professional work, and put its president, Greg Brockman, in charge of revamping its product lineup. It also struck a deal with Amazon to sell AI tools to the cloud giant’s customers and hired former Slack CEO Denise Dresser to become its first chief revenue officer.
“I think that now we’re at a point where it feels like we have this really humming machine,” Brockman said at a July media lunch.
Altman is also in Washington this week meeting with Trump administration officials and lawmakers to preview a new model as the government debates how to further regulate the industry.
Executives are still trying to figure out the most effective pricing plans for its Codex enterprise product, including price cuts that could help it gain market share, though that risks eroding closely watched profit margins ahead of an IPO.
OpenAI might now wait until next year to go public, people with knowledge of the plans said, after earlier hoping to beat Anthropic. When it filed to go public a week after Anthropic, it hedged on timing, saying “it may be a while” because there are “things we want to do that are likely easier as a private company.”
Codex challenges
OpenAI was initially well positioned for the enterprise business it is now scrambling to land.
It launched a series of so-called reasoning models in the fall of 2024 that were able to work through problems step-by-step before answering, a capability that turned out to be well-suited for writing code. But researchers trained those models to ace coding problems from high-school competitions, not to handle the messier, more open-ended work of building software.
Anthropic took the opposite approach with its own reasoning model, called Sonnet 3.7, a few months later. In a February 2025 blog post, the company said it was focused on “real-world tasks” that reflected how businesses actually used AI.
OpenAI gave employees early access to a new version of Codex early last year, expecting it to be an instant hit, particularly among tech-savvy software engineers who had long been tracking the technology’s progress. Instead, usage was lower than anticipated, frustrating executives who repeatedly flagged the issue internally.
At first, the challenges with Codex appeared to be nothing more than a blip. ChatGPT’s weekly active users continued to climb, as did the company’s valuation. Last August, OpenAI brought in Simo, the former Instacart CEO, to help prepare it for a blockbuster IPO.
Instead, Codex’s troubles turned out to be an early warning sign of challenges to come. After OpenAI announced the tool publicly in May, it underperformed expectations. Software engineers flocked to Claude Code instead.
Many developers felt that Codex was too slow and clunky to use, leading OpenAI to change the product to more closely mimic Anthropic’s. The company also created a new team, called coding ninja, tasked with making sure new models were designed in a way that customers found useful.
But executives were also distracted by other, more pressing issues, such as fending off a talent raid launched by Meta CEO Mark Zuckerberg and fixing a worsening relationship with Microsoft, its largest investor. They spent precious computing resources on other projects that flopped, including the video-generator app Sora.
By the time the company turned its attention back to Codex, it was already behind.
Executives released a new model, GPT-5.2, in December, focused on coding and other professional tasks, overruling some employees who asked to push back its release so the company could have more time to make it better.
When the company heard Anthropic was in talks to partner with Cerebras, a startup that had built a specialized chip to process coding tasks at a rapid clip, it rushed to sign a deal of its own in an effort to box out its rival, people familiar with the matter said.
It wasn’t enough to stem the bleeding.
While Altman visited St. Barts in the Caribbean during the December holiday break, coders in San Francisco spent hours on “Claude-benders,” marveling at capabilities enabled by a new Anthropic model called Opus 4.5. Anthropic then went on a multimonth growth streak that surprised even its own leaders.
Meanwhile, ChatGPT’s growth suddenly tapered off after Google’s own consumer chatbot began to surge in popularity. The company missed an internal goal of hitting one billion weekly active users by the end of the year, although it hit the figure recently.
Before long, Anthropic’s growth rate surged past OpenAI’s, as did its valuation—and OpenAI found itself waging two separate battles as its lead in the AI race came under siege.
When executives pitched senior leaders at Blackstone on a partnership to create a new company to sell AI tools to the private-equity giant’s portfolio companies, they didn’t bite. Blackstone went with Anthropic for the project instead. OpenAI later struck a separate deal with other private-equity players.
By March, some OpenAI employees had grown frustrated, peppering a top executive with questions about the company’s future.
“It feels like Anthropic, a much smaller company by headcount and market cap, is consistently setting the frame technically and culturally, and we are reacting. Why does this keep happening?” one question read.
Another framed the problem in even starker terms: “What impact would Anthropic’s revenue surpassing ours have on our IPO plans?”
Simo, who recently left the company after a health condition worsened, said the company had lost its way chasing “side quests” and had fallen behind its rival in both research and product. Anthropic’s revenue “is a proxy for how well they are accomplishing their mission,” she said. “And the fact that they are accelerating such a big part of the economy should be a real wake up call for us.”
OpenAI is now trying to seize on its recent momentum to turn the tables in the AI race.
The company recently released a “super app” that integrates Codex with ChatGPT and a web browser, and said it has boosted the number of users for the new product, plus the stand-alone Codex app, to 10 million users. This month, it released a new model, called GPT 5.6 Sol, that became an instant hit with developers—and led Anthropic to extend access to its own powerful Fable model to compete.
It is also benefiting from a broader industry backlash against Anthropic, which has been accused of trying to keep cheaper, Chinese models out of the U.S.
OpenAI has recently “nailed the day-to-day experience for developers and knowledge workers,” said Dan Shipper, the chief executive of Every, a media and software company focused on AI. “I would put a lot of money on OpenAI winning back the mandate of heaven a bit from Anthropic.”
