Sources: Open AI May Delay IPO Launch Until Next Year
OpenAI, the company that helped ignite the global artificial-intelligence boom, has been overtaken by rival Anthropic and is working to regain ground. Growth in OpenAI’s flagship consumer product, ChatGPT, has slowed sharply. Fidji Simo, Sam Altman’s designated successor, resigned, prompting senior executives under the chief executive to redistribute responsibilities. The sales team has had to offer bulk discounts and other concessions at a high cost to win high-value enterprise customers. Several people involved in related discussions said some of OpenAI’s major investors have privately expressed concern in recent months that the startup is burning cash too quickly and that revenue growth is not keeping pace. Other investors have sought to diversify their risk by also investing in Anthropic. Anthropic’s revenue growth and valuation have recently surpassed OpenAI’s, helped by the success of its Claude Code coding tool. Its valuation is now approaching $1 trillion. People familiar with the matter said Anthropic is accelerating plans for a fall initial public offering and is speaking with potential investors, emphasizing that it has moved ahead of OpenAI, the developer of ChatGPT. Altman wrote on X earlier this month: “Over the past 12 months, we have not delivered our best work, and the primary responsibility lies with me. But over the next 12 months, we may enter the best period of growth in our history. The team is working on major initiatives, and you will see some surprises very soon.” OpenAI’s effort to retain its leadership position has encountered mounting difficulties, rooted in earlier management misjudgments about the direction of the AI market. The company’s current predicament reflects the intensifying global competition for dominance in AI. Altman initially bet OpenAI’s growth on ChatGPT, expecting that as artificial intelligence became more deeply integrated into everyday life, more users would pay for chatbot subscriptions. But the overnight success of Claude Code showed the market that the larger opportunity lies in selling tools to high-end software developers and large companies that employ substantial numbers of programmers. As OpenAI pursued attention-grabbing projects ranging from video-generation models and consumer devices to custom chips, its smaller and more focused rival identified a market gap and launched a hit coding tool, overtaking OpenAI. To catch up with its top competitor, OpenAI released several new models focused on programming and professional office work. It appointed President Greg Brockman to lead a product-line overhaul. The company also reached a partnership with an unnamed cloud giant to sell AI tools to that company’s customers, and hired former Slack Chief Executive Dennis Dresser as its first chief revenue officer. Brockman said at a media luncheon in July: “I think our entire operating system is now functioning efficiently.” This week, Altman was also in Washington, meeting with officials from the Trump administration and members of Congress and previewing new models as the U.S. government considers how to further regulate the AI industry. The executive team is still working to determine the best pricing for the enterprise version of Codex. Lower prices could expand market share, but they would erode the profit margins closely watched by investors and create pressure for a future IPO. People familiar with the matter said OpenAI may delay launching its IPO until next year. The company had previously hoped to complete an IPO before Anthropic. One week after Anthropic filed its IPO application, OpenAI also submitted listing-registration documents, but it did not set a timetable. It said the timing for an IPO was “still too early” because “many strategic objectives are easier to pursue as a private company.” Codex faces steep challenges OpenAI initially had a first-mover advantage in enterprise services but is now scrambling to catch up. In fall 2024, OpenAI released a series of reasoning models capable of working through problems step by step, a capability well suited to coding. But researchers trained the models to solve high-school competition programming problems rather than the fragmented, open-ended and complex engineering tasks involved in real-world software development. Several months later, Anthropic took the opposite approach and released its Sonnet 3.7 reasoning model. In a February 2025 blog post, the company said its development work focused on “real-world business tasks” and was designed around how companies actually use AI. Early last year, OpenAI opened a new version of Codex to employees for beta testing, expecting it to quickly take off and attract heavy use from veteran engineers who closely follow AI developments. But activity fell short of expectations, and executives repeatedly raised warnings internally. At first, the problems with Codex were viewed as a short-term fluctuation. ChatGPT’s weekly active users continued to rise, and the company’s valuation climbed with them. In August last year, OpenAI recruited former Instacart Chief Executive Fidji Simo as it prepared for a highly anticipated IPO. The Codex slowdown later proved to be an early warning of a series of crises. After OpenAI officially released the product in May, its market performance fell short of expectations, and software engineers increasingly switched to Claude Code. Many developers said Codex was slow and cumbersome to use, forcing OpenAI to redesign the product along lines pioneered by Anthropic. The company also formed a dedicated team called the “coding ninjas” to ensure that new models were designed around customers’ actual needs. At the same time, executives were distracted by other urgent matters, including defending against a talent raid led by Meta Chief Executive Mark Zuckerberg and repairing a deteriorating relationship with the company’s largest investor, which the source does not identify. Large amounts of valuable computing capacity were devoted to projects that ultimately failed, including the Sora video-generation application. By the time OpenAI refocused its resources on Codex, it had already lost its market lead. In December, OpenAI released GPT-5.2, a new model focused on programming and professional work. Some employees advised delaying the launch to refine the product, but management rejected the proposal. People familiar with the matter said OpenAI learned that Anthropic was discussing a partnership with Cerebras, a chip startup that develops specialized processors capable of handling coding tasks at high speed. OpenAI quickly finalized a partnership of its own in an effort to block its rival. The move still failed to stop the slide. Over the December Christmas holiday, Altman vacationed on the Caribbean island of St. Barts while programmers in San Francisco worked through the night testing Anthropic’s new Opus 4.5 model and marveling at its capabilities. Anthropic’s revenue then grew rapidly for several consecutive months, exceeding management’s expectations. As consumer chatbot interest surged, growth in ChatGPT users suddenly slowed. OpenAI had set a goal of reaching 1 billion weekly active users by the end of last year. It initially missed that target and only recently achieved it. Soon afterward, Anthropic surpassed OpenAI in both revenue growth and valuation. As its position at the top of the AI industry came under pressure, OpenAI found itself fighting on two fronts. OpenAI executives had presented a partnership proposal to the unnamed group identified in the source only as a “___ Group.” The plan called for forming a new company to sell AI tools to companies in Blackstone’s portfolio, but Blackstone did not adopt it and instead chose to work with Anthropic. OpenAI later reached a separate partnership with other private-equity firms. By March this year, some OpenAI employees had become frustrated and repeatedly questioned an executive about the company’s future. One internal question said: “Anthropic’s employee count and market value are far below ours, but whether it is technology or industry opinion, they continue to define the rules while we can only respond passively. Why does this keep happening?” Another question was more pointed: If Anthropic’s revenue exceeded ours, what impact would that have on the company’s IPO plans? Simo recently resigned after her health deteriorated. She said the company had spent too much time pursuing side projects, lost its strategic focus and fallen behind its rivals across technology development and product execution. “Anthropic’s revenue performance is a direct measure of how effectively it is carrying out its mission. It is penetrating many core areas of the economy, and that should sound an alarm for us,” she said. OpenAI is now trying to capitalize on momentum from its latest products and reverse the balance of power in AI. The company recently launched a super app integrating Codex, ChatGPT and a web browser. The company said the new product and the standalone Codex application have more than 10 million users combined. This month, OpenAI released GPT 5.6 Sol, which quickly won favor with developers. In response, Anthropic expanded public access to its flagship Fable model and entered the competition. A shift in industry sentiment has also benefited OpenAI. Some companies have begun boycotting Anthropic, which critics accuse of trying to block more cost-effective Chinese AI models from entering the U.S. market. Dan Shipper, chief executive of AI-focused media and software company Every, said: “OpenAI is now genuinely optimizing for the everyday experience of developers and knowledge workers. I am very bullish on OpenAI’s ability to retake industry leadership from Anthropic.”
Growth in OpenAI’s flagship consumer product, ChatGPT, has slowed sharply. Fidji Simo, Sam Altman’s designated successor, resigned, prompting senior executives under the chief executive to redistribute responsibilities. The sales team has had to offer bulk discounts and other concessions at a high cost to win high-value enterprise customers.
Several people involved in related discussions said some of OpenAI’s major investors have privately expressed concern in recent months that the startup is burning cash too quickly and that revenue growth is not keeping pace. Other investors have sought to diversify their risk by also investing in Anthropic.
Anthropic’s revenue growth and valuation have recently surpassed OpenAI’s, helped by the success of its Claude Code coding tool. Its valuation is now approaching $1 trillion. People familiar with the matter said Anthropic is accelerating plans for a fall initial public offering and is speaking with potential investors, emphasizing that it has moved ahead of OpenAI, the developer of ChatGPT.
Altman wrote on X earlier this month: “Over the past 12 months, we have not delivered our best work, and the primary responsibility lies with me. But over the next 12 months, we may enter the best period of growth in our history. The team is working on major initiatives, and you will see some surprises very soon.”
OpenAI’s effort to retain its leadership position has encountered mounting difficulties, rooted in earlier management misjudgments about the direction of the AI market. The company’s current predicament reflects the intensifying global competition for dominance in AI.
Altman initially bet OpenAI’s growth on ChatGPT, expecting that as artificial intelligence became more deeply integrated into everyday life, more users would pay for chatbot subscriptions. But the overnight success of Claude Code showed the market that the larger opportunity lies in selling tools to high-end software developers and large companies that employ substantial numbers of programmers.
As OpenAI pursued attention-grabbing projects ranging from video-generation models and consumer devices to custom chips, its smaller and more focused rival identified a market gap and launched a hit coding tool, overtaking OpenAI.
To catch up with its top competitor, OpenAI released several new models focused on programming and professional office work. It appointed President Greg Brockman to lead a product-line overhaul. The company also reached a partnership with an unnamed cloud giant to sell AI tools to that company’s customers, and hired former Slack Chief Executive Dennis Dresser as its first chief revenue officer.
Brockman said at a media luncheon in July: “I think our entire operating system is now functioning efficiently.”
This week, Altman was also in Washington, meeting with officials from the Trump administration and members of Congress and previewing new models as the U.S. government considers how to further regulate the AI industry.
The executive team is still working to determine the best pricing for the enterprise version of Codex. Lower prices could expand market share, but they would erode the profit margins closely watched by investors and create pressure for a future IPO.
People familiar with the matter said OpenAI may delay launching its IPO until next year.
The company had previously hoped to complete an IPO before Anthropic. One week after Anthropic filed its IPO application, OpenAI also submitted listing-registration documents, but it did not set a timetable. It said the timing for an IPO was “still too early” because “many strategic objectives are easier to pursue as a private company.”
Codex faces steep challenges
OpenAI initially had a first-mover advantage in enterprise services but is now scrambling to catch up.
In fall 2024, OpenAI released a series of reasoning models capable of working through problems step by step, a capability well suited to coding. But researchers trained the models to solve high-school competition programming problems rather than the fragmented, open-ended and complex engineering tasks involved in real-world software development.
Several months later, Anthropic took the opposite approach and released its Sonnet 3.7 reasoning model. In a February 2025 blog post, the company said its development work focused on “real-world business tasks” and was designed around how companies actually use AI.
Early last year, OpenAI opened a new version of Codex to employees for beta testing, expecting it to quickly take off and attract heavy use from veteran engineers who closely follow AI developments. But activity fell short of expectations, and executives repeatedly raised warnings internally.
At first, the problems with Codex were viewed as a short-term fluctuation. ChatGPT’s weekly active users continued to rise, and the company’s valuation climbed with them. In August last year, OpenAI recruited former Instacart Chief Executive Fidji Simo as it prepared for a highly anticipated IPO.
The Codex slowdown later proved to be an early warning of a series of crises. After OpenAI officially released the product in May, its market performance fell short of expectations, and software engineers increasingly switched to Claude Code.
Many developers said Codex was slow and cumbersome to use, forcing OpenAI to redesign the product along lines pioneered by Anthropic. The company also formed a dedicated team called the “coding ninjas” to ensure that new models were designed around customers’ actual needs.
At the same time, executives were distracted by other urgent matters, including defending against a talent raid led by Meta Chief Executive Mark Zuckerberg and repairing a deteriorating relationship with the company’s largest investor, which the source does not identify. Large amounts of valuable computing capacity were devoted to projects that ultimately failed, including the Sora video-generation application.
By the time OpenAI refocused its resources on Codex, it had already lost its market lead.
In December, OpenAI released GPT-5.2, a new model focused on programming and professional work. Some employees advised delaying the launch to refine the product, but management rejected the proposal.
People familiar with the matter said OpenAI learned that Anthropic was discussing a partnership with Cerebras, a chip startup that develops specialized processors capable of handling coding tasks at high speed. OpenAI quickly finalized a partnership of its own in an effort to block its rival.
The move still failed to stop the slide.
Over the December Christmas holiday, Altman vacationed on the Caribbean island of St. Barts while programmers in San Francisco worked through the night testing Anthropic’s new Opus 4.5 model and marveling at its capabilities. Anthropic’s revenue then grew rapidly for several consecutive months, exceeding management’s expectations.
As consumer chatbot interest surged, growth in ChatGPT users suddenly slowed. OpenAI had set a goal of reaching 1 billion weekly active users by the end of last year. It initially missed that target and only recently achieved it.
Soon afterward, Anthropic surpassed OpenAI in both revenue growth and valuation. As its position at the top of the AI industry came under pressure, OpenAI found itself fighting on two fronts.
OpenAI executives had presented a partnership proposal to the unnamed group identified in the source only as a “___ Group.” The plan called for forming a new company to sell AI tools to companies in Blackstone’s portfolio, but Blackstone did not adopt it and instead chose to work with Anthropic. OpenAI later reached a separate partnership with other private-equity firms.
By March this year, some OpenAI employees had become frustrated and repeatedly questioned an executive about the company’s future.
One internal question said: “Anthropic’s employee count and market value are far below ours, but whether it is technology or industry opinion, they continue to define the rules while we can only respond passively. Why does this keep happening?”
Another question was more pointed: If Anthropic’s revenue exceeded ours, what impact would that have on the company’s IPO plans?
Simo recently resigned after her health deteriorated. She said the company had spent too much time pursuing side projects, lost its strategic focus and fallen behind its rivals across technology development and product execution. “Anthropic’s revenue performance is a direct measure of how effectively it is carrying out its mission. It is penetrating many core areas of the economy, and that should sound an alarm for us,” she said.
OpenAI is now trying to capitalize on momentum from its latest products and reverse the balance of power in AI.
The company recently launched a super app integrating Codex, ChatGPT and a web browser. The company said the new product and the standalone Codex application have more than 10 million users combined. This month, OpenAI released GPT 5.6 Sol, which quickly won favor with developers. In response, Anthropic expanded public access to its flagship Fable model and entered the competition.
A shift in industry sentiment has also benefited OpenAI. Some companies have begun boycotting Anthropic, which critics accuse of trying to block more cost-effective Chinese AI models from entering the U.S. market.
Dan Shipper, chief executive of AI-focused media and software company Every, said: “OpenAI is now genuinely optimizing for the everyday experience of developers and knowledge workers. I am very bullish on OpenAI’s ability to retake industry leadership from Anthropic.”