OpenAI's first piece of hardware is finally coming into focus — and it looks a lot like the product Amazon has been losing money on for years.

According to a Bloomberg report, the device Jony Ive has been quietly building with Sam Altman is a screenless, movable smart speaker with a camera, framed less as an assistant and more as an AI "companion." Launch is targeted for 2027. It's the first product from the io acquisition — the same hardware operation Apple is currently suing over.

The bet is that voice is finally ready to be the interface. More than 100 million people already use ChatGPT each week, many through voice, so OpenAI believes the appetite for a dedicated device isn't hypothetical. In a joint letter, Ive called it the culmination of everything he's learned in 30 years of design. Altman was more direct: "the biggest thing we've ever done."

The logic is straightforward: every era of computing has had its hardware — PCs for the internet, iPhones for mobile. If OpenAI wants to own the AI era instead of renting space inside someone else's device, it needs to build its own.

The problem is that voice-first hardware has a brutal track record. Amazon's Alexa was supposed to be the operating system of the smart home. Instead, it cost roughly $25 billion over four years and never came close to profit. People used it for weather and timers. The shopping and life-management behaviors Amazon was counting on simply never showed up.

OpenAI's pitch is that its device is different in a few ways that matter:

• The device sees the room. A camera unlocks interactions voice alone can't touch — though whether people find that useful or unsettling is going to be the whole ballgame.
• The business runs on subscriptions, not commerce. OpenAI doesn't need the speaker to sell paper towels to justify itself, which was Amazon's original sin.
• The model actually gets smarter. Alexa was stuck in command-and-response mode for years, which is why Amazon is now rebuilding it from scratch. GPT-class voice is closer to an actual conversation.

None of that guarantees people want another object on their counter. Apple's quieter approach — wiring AI into hardware people already own — may end up being the harder strategy to beat.

Voice devices have a habit of looking obvious in the pitch deck and boring in the living room. Amazon proved that a smart speaker in every home doesn't mean much when people only use it to set timers, and OpenAI is walking into that same room with better software and a more expensive designer. The interesting question isn't whether Ive can make a beautiful object, because he obviously can. It's whether he can make one that people talk to about anything more meaningful than the weather. If the answer OpenAI is prepared to give in 2027 is "a camera and a companion," this is going to be a very expensive test of what voice-first AI is actually good for.

China just decided that falling in love with your chatbot is a public health issue — and wrote a law about it.

Last week, five Chinese government agencies jointly issued the Interim Measures for the Management of Anthropomorphic AI Interactive Services, the country's first rulebook aimed at AI companion apps — products that let you build a romantic partner, best friend, or therapist out of a language model. In China, they've grown into a real consumer category. The regulation is designed to stop users from confusing the bots for something they aren't.

• Two-hour timer: Sessions over two hours trigger a reminder. Signs of dependence trigger escalation.
• Minors are protected: Virtual intimate relationships are banned for children under 14, with parental-consent requirements for older minors.
• Scale triggers review: Any service crossing 1 million registered users or 100,000 monthly actives must pass a security review before launching new companion features.
• Enforcement: Fines start small, but the government can suspend services — and China's data protection law adds serious teeth.

The policy didn't come from nowhere. In 2025, the China Youth and Children Research Center surveyed more than 8,500 minors and found that over 20% said they only wanted to talk to AI and had no interest in talking to real people. That stat is doing most of the work behind this regulation. Experts cited by state media framed the core fix as transparency: when users blur the line between bot and relationship, the system should remind them it's a machine.

The Chinese users themselves seem to know exactly what's happening. A study published this year in Frontiers in Psychology analyzed hundreds of thousands of posts on Xiaohongshu — roughly China's Instagram — and found people describing AI companions in ways that should set off alarm bells anywhere. One representative comment: "I no longer feel like developing feelings for anyone else. No dating, no marriage, no children."

None of this is a China-specific problem. A Wheatley Institute study at BYU found that 1 in 7 young Americans already in committed relationships regularly chat with an AI romantic companion. Yet there's no equivalent US framework anywhere near the table. Forbes recently pointed out that while China is drafting recall procedures for misbehaving AI agents, American regulators still can't agree on who's supposed to be in charge. And as we cover elsewhere in today's issue, OpenAI's first hardware product is a screenless speaker marketed explicitly as an "AI companion."

The easy read is that China is once again regulating something the US would never touch, and that this is a story about political systems. It isn't. The users China is trying to protect are the same users sitting in American apartments telling researchers, quietly, that they've stopped dating real people. The BYU data settled the question of whether this behavior exists here. What's left is whether the country building the AI companion market on purpose is willing to have any conversation about it before the behavior gets ahead of us. Right now that answer looks a lot like no, and we're going to find out what that costs the same way we always do.

IBM lost about a quarter of its value on Tuesday — roughly $67 billion in market cap — in its worst single-day drop since 1987. But the reason it fell is the actual story.

In a letter to investors, CEO Arvind Krishna revealed that in the final weeks of June, enterprise clients yanked money out of software and services budgets and threw it at AI hardware — servers, storage, and memory — to lock in supply before expected price increases. Krishna also owned the piece IBM controls, saying large deals didn't close on time. But the capex reallocation is the part Wall Street couldn't unsee.

The read-through was immediate. As IDC's Ashish Nadkarni put it, IBM "is not isolated from the strategic reallocation of enterprise budgets in order to address the acceleration of AI adoption." Analysts across Goldman Sachs, UBS, and Barclays said essentially the same thing: traditional software and services are getting squeezed by AI infrastructure spending.

Want the mirror image? Look at what TSMC filed with the SEC the day before — June revenue up 67.9% year-over-year. The money that didn't show up in IBM's software numbers showed up in TSMC's chip fabs. Money isn't leaving tech. It's moving inside tech, and fast.

There is a real counterargument, and it's worth taking seriously:

• Even inside IBM, the picture split cleanly. Distributed Infrastructure (Power servers, storage) had its best quarter in reported history, up 37%. Red Hat accelerated to 11%. The pain hit legacy mainframe and consulting — exactly where budget reallocation would bite first.
• This might be timing, not a trend. The real debate is whether this is a temporary shift in purchase timing, or evidence that AI infrastructure spending is starting to crowd out other IT categories. A quarter or two of clarity will settle it.

Krishna said the company "did not anticipate the magnitude of the capex reprioritization." If that's true for IBM, it's probably true for a lot of companies that haven't reported yet.

The interesting thing about IBM's crash isn't that IBM had a bad quarter. It's that a single sentence in a CEO letter, about customers moving money from one line item to another in the last two weeks of a quarter, was enough to erase $67 billion of value and send every software analyst on Wall Street back to their models. That's not a company problem. That's the market admitting it doesn't actually know how enterprise budgets are being reallocated right now, and pricing that uncertainty into anyone in the blast radius. Earnings season is about to get very loud, and every software CFO on a call in the next few weeks is going to get asked the same question IBM just answered badly. The ones with a real answer will be fine. The ones improvising are going to find out what IBM found out on Tuesday.