Nvidia's Perplexity Bet, XPeng's Robot Record, and Apple's Foldable Buzz

Nvidia's back in the headlines for basically owning every part of the AI food chain at once. XPeng just proved investors will throw serious money at a robot that can't even legally cross the street yet. And Apple's foldable phone is winning fans before anyone outside Cupertino has really held one. Here's the rundown, and what it actually means once you get past the headlines.

Quick answer

Nvidia is talking about putting money into Perplexity at a valuation over $30 billion, on top of a separate deal worth around $7 billion to help build open-weight AI models. Both moves come down to the same thing: keeping demand for its chips growing no matter who wins the AI race. XPeng's robotics arm just landed the biggest single funding round any Chinese humanoid robot company has ever pulled off, even while the car business underneath it is losing more money than it did last year. And Apple's first foldable iPhone is already getting glowing early reviews weeks before it's even been announced, which is exactly the kind of story worth reading with one eyebrow raised.

Nvidia is playing every side of the board

Reuters and The Information both report that Nvidia is in talks to join a funding round for Perplexity that would value the AI search company above $30 billion. That's more than 50% higher than its last raise. The number actually tracks with Perplexity's growth: annualized revenue went from under $250 million in January to more than $750 million now, mostly thanks to its "Perplexity Computer" agent product. Nothing's signed yet, and at one point Nvidia was reportedly weighing just acquiring Perplexity's tech and team outright instead of investing.

There's a second move here too, and it's honestly the more interesting one. Nvidia struck a deal worth roughly $7 billion with an AI startup called Poolside, made up of a $6 billion licensing fee plus a $1 billion investment, valuing Poolside at $12 billion before the money even lands. More than 100 Poolside staff are moving over to Nvidia as part of it. The goal is to build a genuinely capable open-weight model, one that can go toe to toe with DeepSeek, Kimi K3 and Qwen, the Chinese open models that have quietly become a lot of developers' default pick. It fits into a bigger plan too. Nvidia already committed to spending $26 billion over five years on open-weight models, and it's reportedly training its own trillion-parameter model in house on top of that.

None of this is Nvidia being generous. Jensen Huang's logic is pretty simple. Every good open model built on Nvidia's chips sells more chips, no matter which lab ends up winning at the top. It's also worth staying a little suspicious of the pattern. When Nvidia funds a company that then spends a chunk of that money on Nvidia hardware, that's basically vendor financing wearing a venture capital costume. The more of these deals pile up, the more that circularity starts to matter.

Zoom out and that's really the whole story. Nvidia is turning into an octopus with a hand in every tank: the chips, the open models running on those chips, the apps built on top of those models, and now maybe a stake in one of the biggest consumer AI products around too. It's not placing one big bet to win. It's making sure it gets paid no matter who does.

Here's what that actually means if you're not a shareholder. If you use Perplexity, or pretty much any product built on an open-weight model, you're increasingly paying the same company at every layer without really noticing. It also means a decent chunk of what people call the "AI bubble" is really just Nvidia's own money going out to startups and coming back as chip orders, which makes the whole industry's growth numbers look bigger than actual outside demand would justify on its own. And that's not just a Wall Street problem. Nvidia is one of the biggest single holdings in most index funds and retirement accounts, so if this ever stops working, plenty of regular people feel it too, not just Nvidia.

XPeng's robot arm just outraised its own car business

XPeng's robotics unit raised over $900 million in its first outside funding round, valuing it at more than $6.3 billion. That's the biggest single private funding round any company has landed in China's embodied AI industry so far. IDG Capital led it, with Gaorong Ventures, Tencent and Alibaba also chipping in. XPeng keeps control of the unit and it stays on the company's books.

The money is going toward mass producing IRON, XPeng's humanoid robot. It has 76 degrees of freedom in the body and 21 in each hand, runs on three of XPeng's own Turing AI chips putting out 2,250 TOPS combined, and is built to work without someone remotely operating it. Mass production is supposed to start by the end of 2026, first showing up in XPeng's own stores and campuses before wider deliveries in 2027. The pitch is that XPeng can bring actual car manufacturing discipline to a category that's mostly been flashy demo videos so far. That's a fair point too, since most humanoid robot companies have never mass produced anything in their lives.

Here's the part nobody's hiding though. XPeng's net loss widened 179% year over year to about $199 million last quarter, and R&D spending jumped 32%, all while investors keep throwing money at anything with "physical AI" attached to it. The funding is real and the ambition is real. Whether IRON's spec sheet actually holds up once it's out of a controlled demo and into the real world is still very much an open question, same as it is for every other humanoid robot out there right now.

The iPhone Ultra "early tester" stories are doing exactly what they're supposed to

Bloomberg's Mark Gurman reports that the small group of people who've actually handled Apple's unannounced foldable, internally called iPhone Ultra and expected to show up alongside the iPhone 18 line around a rumored September 9 event, like how small it folds down for your pocket, how solid the hinge feels, and how the software uses the bigger unfolded screen with iPad style layouts. The camera is getting more mixed reactions, and the leaks agree it skips a telephoto lens entirely, which is a real gap on a phone rumored to start above $2,000 and maybe hit $2,500.

Also worth mentioning: Face ID is reportedly gone, swapped for Touch ID built into the side button. That's not a small trade off on what's supposed to be Apple's most premium phone yet.

None of this is confirmed. There's no battery data, no thermal numbers, no idea how the hinge holds up after a year of getting shoved in and out of someone's pocket. A steady stream of "early testers love it" stories landing right before a rumored reveal is a pretty familiar move. It builds excitement without Apple having to say anything official. Worth reading. Not worth pre-ordering off of.

The TGK verdict

None of these three stories are really about the product on the surface. They're about who's paying for the AI boom and who's betting it keeps going. Nvidia is spreading its money across the entire stack so it wins no matter who else does. XPeng is betting that building robots the same disciplined way it builds cars will matter more than flashy demos eventually, even while its core business bleeds money to prove it. And Apple, as always, is letting a slow drip of good press do the selling before it has to commit to anything on the record.

Worth watching all three. Worth believing none of them completely yet.