China Blocks Meta’s $2B Manus Deal, Meta Bets on Space-Based Solar | Diet TBPN
TL;DR
China moved to unwind Meta’s $2 billion Manus acquisition even after closing — TBPN says Beijing wants funds returned, ownership re-registered, and Meta to stop using Manus algorithms, showing that relocating a startup to Singapore didn’t remove Chinese leverage.
OpenAI’s cloud exclusivity with Microsoft appears to be loosening in a meaningful way — the hosts note OpenAI can now potentially serve products across AWS, Google TPUs, and Amazon Trainium, which they frame as a major flexibility shift in the infrastructure battle.
Meta is making an aggressive energy bet for AI data centers with space solar and 100-hour batteries — the company announced a partnership with Overview Energy to beam up to 1 gigawatt of solar power from orbit and another with Noon Energy for 1 gigawatt of ultra-long-duration storage.
A placebo sleep study became their case for ‘golden retriever mindset’ as a performance tool — in a 164-person experiment, participants told they had above-average REM sleep performed better on some cognitive tests even when self-reported sleep quality didn’t predict the results.
Early labor-market evidence still doesn’t show the clean AI job-collapse story people expected — despite Filipino call centers being ‘hard to think of a more AI-exposed job,’ the show says employment in the Philippines was up 4% in 2025 while U.S. call-center declines mostly predate ChatGPT.
The hosts think inference capacity is becoming more strategically important than model weights alone — citing Noam Brown, they argue the real moat may shift from stealing weights to controlling enough chips and data-center capacity to run many more agents than your rival.
The Breakdown
Four big headlines, plus Musk v. OpenAI
The show opens in classic TBPN mode: quick banter about a brighter studio screen, then straight into the day’s four stories. First up is Elon Musk’s $134 billion lawsuit against Sam Altman, with jury selection underway in Oakland and reporters already noting the courtroom vibe around Sam, Greg, and the still-unseen Elon.
OpenAI’s cloud loosen-up changes the infrastructure chessboard
They quickly pivot to Microsoft-OpenAI news, where the key takeaway is that OpenAI can now serve products across more cloud providers instead of being boxed into one setup. The hosts call out Andy Jassy’s excitement, then spell out the real implication: Google TPUs, Amazon Trainium, and other chips are now more plausibly in play.
China blocks the Manus deal and exposes where power really sits
The biggest drama is China ordering Meta to unwind its $2 billion acquisition of Manus, even though Manus had already moved to Singapore and Meta had already started integrating the product. The hosts keep coming back to the human pressure point: team members still have family in China, some were reportedly briefly detained, and whatever the cap table says, Beijing still has leverage.
A very serious detour into Diet Coke reels and semiconductor confusion
Then the show swerves into a long, very TBPN-esque bit about Diet Coke Instagram reels, a fake ‘study’ proving Diet Coke drinkers are ‘Sigma Chads,’ and a hilariously inefficient mansion fridge stocked with nowhere near the claimed $4,000 of soda. From there they bounce into a deleted post saying, ‘Got to be honest, bro. I have no idea what a semiconductor is,’ which sets up a surprisingly useful explainer: copper conducts constantly, silicon can be switched on and off, and that’s the root of digital computing.
The placebo sleep study becomes the case for ‘golden retriever mode’
A viral paper gives them their next meme-worthy thesis: mindset measurably changes cognition. In a study of 164 participants, people randomly told they had above-average REM sleep scored better on some tasks, which the hosts translate into one line: ‘Golden Retriever mode, Golden Retriever mindset,’ followed by a Jim Carrey ‘Yes Man’ riff as the cinematic version of that worldview.
Meta’s space-solar plan sounds wild, but the data-center need is real
Back to Meta: they dig into the company’s plan to use space-based solar power at night via Overview Energy and pair it with Noon Energy batteries offering 100 hours of storage. The hosts are clearly fascinated by the idea of mirrors or lasers beaming 24/7 energy from orbit, but they also stress the brutal operational reality that data centers can’t tolerate intermittent power without wrecking utilization economics or even equipment.
AI jobs, Poland’s rise, and why the labor story is still messy
On AI and employment, they argue the cleanest place to see disruption should be offshore call-center markets like the Philippines, yet employment there was reportedly up 4% in 2025. That leads into a broader point: U.S. call-center decline began long before ChatGPT, likely driven more by outsourcing, while Poland gets a surprise victory lap for rapidly rising incomes and even attracting British workers.
Inference is the new strategic weapon, and AI assistants are getting weirdly personal
Citing Noam Brown, the hosts say weights may matter less over time than the ability to run inference at scale, because chips and cluster capacity decide who can deploy more agents in the real world. They pair that with a Semafor story about a founder whose AI reads messages, listens to family conversations, schedules meetings, and effectively ‘is’ him — weird, yes, but also close enough to today’s alarm clocks, maps, and calendars that they think this kind of mediation could feel normal faster than people expect.
Manus fallout, Chinese leverage, and a glimpse of generative UI
In the closing segment, they return to Manus and read through the Financial Times framing: Beijing may mainly be trying to send a message and deter future cross-border AI deals. They end on a lighter future-facing note with a generative weather map made in ChatGPT Images 2.0 and Ben Thompson’s bull case for Meta AR — the bigger idea being that AI-generated interfaces, not static app chrome, may be what starts to feel truly magical.