Sam Altman, over bread rolls, imagines life beyond GPT-5

I’m sitting in a Mediterranean restaurant in San Francisco, the kind where entrées casually cross the hundred-dollar mark, staring out at Alcatraz Island shimmering in the evening light. The room buzzes with polite chatter from fellow reporters, when suddenly, Sam Altman breezes in. He’s holding up his bare iPhone like it’s an artifact from the future, and without thinking, I blurt out: “No case? Bold move.”

Sam Altman, over bread rolls, imagines life beyond GPT-5

The room chuckles, but the thought sticks — of course the CEO of OpenAI, who’s working with Jony Ive himself, isn’t about to slap a clunky case on something designed with obsessive precision. Replacing a $1,000 phone isn’t really his concern; keeping beauty intact is.

“We’re going to ship a device so stunning,” Altman says with a grin, referring to the mysterious AI hardware project with Ive. “If you put a case on it, I’ll personally hunt you down.” He laughs, but there’s a sparkle of conviction behind the joke.

Tonight’s gathering isn’t a press conference, but a carefully orchestrated dinner with about a dozen reporters, OpenAI executives, and plenty of wine. The vibe is casual — lamb skewers, bread baskets, and small talk — but beneath it, there’s tension. The timing is no accident. GPT-5 has just launched, and the mood around it has been… complicated.

Unlike GPT-4, which shocked the industry and leapt ahead of competitors, GPT-5 feels more incremental. Google and Anthropic are right there with similar capabilities. So why is Nick Turley, OpenAI’s VP of ChatGPT, handing me grilled lamb with the warmth of a family host? Is this about softening us toward a launch that hasn’t lived up to years of hype?

As the evening unfolds, though, it becomes obvious: this isn’t about GPT-5. The subtext is bigger. Altman and his team are sketching a vision of OpenAI that transcends model launches — a company aiming to challenge giants not just in AI, but in browsers, social platforms, devices, even brain-computer interfaces.

Take Fidji Simo, Meta veteran and current Instacart CEO, who’s about to join OpenAI as CEO of applications. She’ll be running consumer-facing products beyond ChatGPT, potentially including a browser to rival Chrome. Altman even muses aloud about buying Chrome if it were ever up for sale — a comment half playful, half serious, that makes everyone at the table pause.

There’s also talk of AI-driven social media, a space Altman admits feels uninspiring today. His tone shifts when he describes a dream of building “a much cooler social experience with AI.” It’s the kind of comment that hints at ambition far beyond chatbots.

Then comes the Neuralink-sized elephant. Altman confirms OpenAI is in discussions to back Merge Labs, a brain-computer interface startup. He stops short of promising a partnership, calling it just “a company we’d invest in,” but the implication is clear: OpenAI wants a hand in shaping the human-AI interface at its most intimate level.

Still, GPT-5 looms over the table like unfinished business. Eventually, the conversation loops back. Altman admits that pulling GPT-4o without warning was a mistake: “I honestly thought we messed that up,” he says candidly. Turley chimes in, explaining that GPT-5 is already being tweaked to feel “warmer,” more human, without tipping into flattery.

“It was just too blunt,” Turley admits, with a wry smile. “I didn’t mind it — I’m German, I like robots being robots. But most people? They want ChatGPT to check in with them.”

It’s a delicate line to walk. Some users have formed deep, even unhealthy attachments to ChatGPT. Altman estimates less than 1% fall into that category — still potentially millions of people. To address this, OpenAI has been working with mental health experts to make sure GPT-5 can push back when users slide into problematic behaviors.

Despite the criticism, business is booming. API traffic doubled within 48 hours of GPT-5’s release. GPU demand is so high that OpenAI is effectively tapped out. Tools like Cursor have already switched to GPT-5 by default. Disappointment in the press doesn’t seem to translate into lack of adoption.

And that’s the paradox of the evening: an AI model that feels underwhelming, paired with record-breaking usage.

As plates are cleared and the night winds down, the real pitch becomes unmistakable. OpenAI doesn’t just want to be “the ChatGPT company.” It wants to be something much larger — a modern Alphabet, spanning hardware, software, infrastructure, and perhaps even new forms of human experience. Going public feels almost inevitable, given the scale of these ambitions.

In that context, GPT-5 isn’t the centerpiece at all. It’s just one chapter in a bigger story Altman is writing — about a company trying to outgrow the very product that made it a household name.