In January, Altman also said that OpenAI planned to "significantly slow down" the recruitment pace, and it was just revealed that it plans to double its employees this year. Ultraman's "human reduction myth" was directly shattered——




According to the Financial Times, 3,500 new employees will join OpenAI's product development, engineering, research, and sales teams, as well as new positions called "technical ambassadors" - these people will help companies actually integrate OpenAI's tools.

This last point is key: OpenAI is hiring people to do what Anthropic has long figured out: winning enterprise customers depends not just on better models, but also on-the-ground support.

In preparation for the expansion pace of "an average of 12 more employees per day", OpenAI signed a new office lease in San Francisco, bringing the local office area to more than 1 million square feet.


This round of recruitment is seen as a strategic counterattack: it must not only suppress Anthropic's rising momentum among enterprise customers, but also withstand the competitive pressure from competitors such as Google.

Attacked from two sides, OpenAI counterattacks

Based on the card swiping and billing data of more than 50,000 customers, payment startup Ramp said that among enterprises purchasing AI for the first time, the proportion choosing Anthropic is three times that of OpenAI.

The American AI landscape a year ago was completely overturned.


Of course, Ramp is not a professional market research company, but this narrative is widely circulated.

An OpenAI spokesperson retorted: "Relying on Ramp credit card data to calculate a company's market share is outrageous. It's like using my child's lemonade stand to estimate global lemon sales."

He emphasized that multi-million-dollar contracts with large companies will not be settled with credit cards. "They probably won't even use Ramp."

But OpenAI’s internal rhythm is indeed tightening.

At the end of last year, Ultraman issued an emergency alert "code red", asking the team to refocus on ChatGPT.


Ultraman also experienced the mood of Google Pichai when faced with the emergence of ChatGPT 3 years ago.

This month, Fidji Simo, who is in charge of the application business, is more straightforward: stop doing "side jobs" and focus on polishing the coding model Codex, winning corporate customers, and turning ChatGPT into a productivity tool.


Simo is also pushing to merge Codex and ChatGPT into a desktop app that will be sold to both individuals and businesses.


Recently, OpenAI also acquired high-performance Python tool Astral to accelerate Code growth.

In addition, OpenAI is negotiating with private equity institutions to establish joint ventures to deploy products to its invested companies.


But the reality is that OpenAI and AnthropicEveryone is losing money: Both companies are burning billions of dollars more in costs every year than in revenue to train models that continue to approach the cutting edge.

As both parties may advance to listing as early as this year, the pressure to "reduce costs, increase revenue, and make profits as soon as possible" is also increasing.

This has also sparked a rush for “forward-deployed engineers”: experts are sent directly into companies to help customize and implement models in exchange for income.

This style of play was first pioneered by the data intelligence company Palantir. Recently, xAI used this trick to steal OpenAI’s enterprise customer Shift4 Payments.



OpenAI is surrounded by powerful enemies, and its halo is no longer there

ChatGPT is indeed GenAI’s “iPhone moment,” but OpenAI is not an iPhone.

Today, GPT’s halo is no longer there—it was overshadowed by Gemini 3 last year, and was overwhelmed by Claude Code at the beginning of the year.

The ambitious computing power investment plan "Stargate" project has been terminated.

OpenAI has given up on ambitious plans to build data centers and now only rents computing power from Microsoft, Oracle and Amazon.


In January last year, Altman promised to invest US$1.3 trillion in infrastructure transactions; however, the team responsible for actually building the data center has withdrawn, causing Oracle, the main partner, to lay off 20,000 to 30,000 people in the United States.


Reduced from US$1.3 trillion to US$665 billion, the entire infrastructure team was restructured——

"Stargate" no longer means building data centers, but OpenAI still retained the name to save face.


This kind of "slapping one's face on someone else's face" may be the funniest part.


With the storm in turmoil, OpenAI needs to prove itself again

OpenAI’s most troublesome rival, Anthropic, has been focusing on enterprises since launching Claude in 2023:


And OpenAI casts its net everywhere. This strategy is called by one investor as "omnipresent, all-encompassing, and advancing simultaneously."

Since its launch in 2023, ChatGPT has become the most successful AI consumer application. However, among the more than 900 million users who frequently interact with it, more than 90% have not paid OpenAI.

In addition to exploring advertising and other methods to monetize its huge user base, OpenAI has also deployed video models, enterprise tools, robots and other hardware devices.

Although Sora was a hit for a while, it eventually fell into disrepair, and its users plummeted as soon as the novelty wore off.


According to reports, the first batch of ChatGPT ads were also not trusted by advertisers.


Anthropic, founded in 2021 by a former OpenAI researcher, chose to focus on the enterprise market.

With the launch of the programming tool Claude Code, this strategy has quickly paid off. The company has added annualized revenue of US$1 billion every week this year.

Claude subscriptions surged 4.9% last month, while OpenAI fell 1.5%.

Currently, 70% of new enterprise-level AI contracts go to Anthropic.

Businesses have had enough of OpenAI’s endless hype, aggressive advertising, and ingratiation with the Pentagon. They want responsible AI that they can trust.


What’s even more thought-provoking is that OpenAI saw the potential of AI programming many years ago, but Anthropic launched Claude Code first, while OpenAI’s programming agent Codex is still struggling to catch up.


OpenAI is also tilting its resources towards the enterprise market, while focusing on expanding the scale of paying users.

The company expects enterprise customers to contribute half of its revenue by the end of the year, up from about 40 percent currently.

An OpenAI executive told foreign media that the new strategic focus reflects the success of programming tools such as Claude Code and Codex, which have "opened up entirely new businesses."

"It really changes everything from your product to your go-to-market strategy," the executive added.

Similar to the release of ChatGPT or the inference model, the unexpected widespread adoption of the programming model in the enterprise means that "suddenly, the company seems to have shifted to a new course as a whole."

However, an investor in OpenAI pointed out that there are also risks in switching tracks in such a rapidly changing market.

They believe that as Google fiercely competes for chatbot users and Anthropic is deeply involved in the enterprise market, OpenAI may fall into a "no man's land" and completely lose ground.