Author：Bloomberg

Call it the season of IPO prep. Recent launches and announcements from OpenAI and arch rival Anthropic are aimed at laying the groundwork to go public in late 2026 or early 2027, and for OpenAI, which just closed a $122 billion funding round that values it at $852 billion, everything is riding on a model codenamed Spud.

OpenAI Chief Executive Officer Sam Altman needs a big win. ChatGPT has been losing consumer market share to Google’s Gemini, while its enterprise business has been squeezed by Anthropic, whose Claude Cowork service caused a stock market freakout in January.

Altman would no doubt love to release features that cause such hand-wringing, but he has suffered from Mark Zuckerberg syndrome. The Facebook founder has a habit of chasing shiny objects like the metaverse and crypto, and wasting billions. Altman has been similarly rash, steering ChatGPT into shopping, and spending much of OpenAI’s compute capacity on video generation for Sora.

Someone must have told Altman some hard truths, judging by his decision to close its video generator Sora last month and walk away from a $1 billion investment from the Walt Disney Co. It was a good move on his part. Altman has been in desperate need of the kind of focus displayed by Dario Amodei, the CEO of Anthropic. So deep is the tension between Altman and Amodei that they refused to hold hands in an awkward celebratory moment recently, but Altman now seems to be copying Amodei’s playbook of picking one area of business and giving it relentless focus.

Harnessing the computing power that went to Sora, the new Spud will underpin what OpenAI has called a “unified AI superapp,” which will combine the chat features of ChatGPT along with artificial intelligence-powered coding and browsing capabilities in a system that carries out tasks. “Users do not want disconnected tools,” OpenAI said in a blog post last Tuesday, subtly admitting it had in fact been creating disconnected tools. “They want a single system that can understand intent, take action, and operate across applications, data, and workflows.”

The idea of a superapp is well known in Asia, typified by services like China’s WeChat, where you can chat, play games, buy things and access government services. China’s Alipay combines food delivery, financial services and transportation.

But the format hasn’t taken off in the West, despite the efforts of Elon Musk to turn X into an “everything app,” which, he said in 2023, would allow users to communicate and conduct their “entire financial world.” That didn’t pan out for X, in part because American and European consumers have deeply formed habits around using specific apps for specific things. They don’t want or need a single gateway to everything.

That may pose some risks for OpenAI’s superapp ambitions, not to mention the fact it could be one-upped by Anthropic’s release of a new model codenamed Capybara. Altman could boost his chances if he capitalizes on his recent purchase of OpenClaw — a developer of AI agents — and tries to imitate Claude Cowork, Anthropic’s new desktop app that can chat, code, answer a person’s LinkedIn messages and operate their computer. That system has received rave reviews among coders and tech workers.

But Anthropic does have weaknesses — including one that Altman could exploit. The startup still can’t beat the distributive advantage of a giant like OpenAI’s partner Microsoft Corp.: Three-quarters of Fortune 500 companies use Microsoft 365 as their main productivity tool, which now includes Copilot, the company’s enterprise AI assistant, by default.

A better use of Altman’s time and renewed focus would be to repair his fraught relationship with Microsoft, which owns 27% of OpenAI’s for-profit arm, but with whom it has locked horns over the past year on competing products and the huge financial and computational costs of developing AI.

He ought to treat the software stalwart less like a cash machine and more like a product partner, whose massive distributive advantage OpenAI already exploits through revenue sharing. ChatGPT’s 900 million weekly users may dwarf the 450 million users of Microsoft 365 tools, but OpenAI’s consumer base is an enormous computing cost, while Microsoft’s enterprise customers offer a road to profits. (OpenAI said last week that it was generating $2 billion in revenue per month, but didn’t say how much it was losing. The company expects to burn through $115 billion of cash through 2029.)

One complaint I’ve heard about Copilot is that its user interface varies between consumers and enterprise users, no doubt the result of Microsoft’s byzantine structure of corporate silos, and that a single, cohesive interface would improve its reputation. Of course that’s not Altman’s remit. “It is 100% Satya’s job to figure out Copilot,” says David Rainville, lead manager of Sycomore Sustainable Tech, referring to Microsoft CEO Satya Nadella. But Altman’s teams could collaborate more closely on turning Copilot into a tool that acts as an AI agent as well as Claude Cowork does.

It doesn’t help the relationship that Microsoft recently started using Anthropic’s Claude in Copilot, not only so that its agentic technology can carry out multi-step tasks such as scheduling meetings and preparing reports, but also to fact-check answers from OpenAI’s GPT model before they reach users. That is a visible, embarrassing gap that OpenAI should be trying to close.

Spud may be an impressive new model, but user experience, focus and execution are proving to be more important than scientific benchmarks, especially as business customers grapple with plugging AI into their systems. Microsoft is exceptionally well placed to build those connections, and while Altman’s superapp ambitions are admirable, he might have a more successful path to an IPO if he works more closely with his partner. More from Bloomberg Opinion:

• Tech Bros Are Londonmaxxing For Good Reason: Parmy Olson
• There’s Method to China’s OpenClaw Madness: Catherine Thorbecke
• The White House’s AI Plan Is Anything But: Dave Lee

​​​​​Want more Bloomberg Opinion? OPIN