The Great Decoupling: China, US & Europe

00:00:00: This is your

00:00:00: daily synthesizer.

00:00:02: Nearly on Sunday, April twenty-six, twenty-twenty six we've got a genuinely packed episode today talking about the great decoupling between China The US and Europe Claude becoming some kind of mega aggregator AI agents out negotiating actual humans And a lawsuit that might reshape the entire AI industry.

00:00:21: But first

00:00:22: First We have to talk About Elon Musk lying To people's faces for A decade Because apparently That just A thing that happened.

00:00:30: Right?

00:00:31: Did you see the earnings call clip?

00:00:33: Oh, I saw it.

00:00:34: hardware three simply does not have the capability.

00:00:37: Just said it on a call to investors after

00:00:40: years of selling people full self-driving for thousands Of dollars.

00:00:44: right and the worst part isn't even the lie.

00:00:48: The worse part is he said it like He was delivering mildly inconvenient weather news.

00:00:53: Unfortunately Hardware Three i wish It were otherwise Like sorry about your house burning down.

00:00:58: Here's an umbrella

00:01:00: and now his solution is micro factories in cities Which sounds like I don't know something, I'd pitch if i had thirty seconds And no budget?

00:01:09: This Is the same company whose profit margins are quote thin as ever

00:01:14: thin As Ever.

00:01:15: Lovely I mean.

00:01:16: look I get it.

00:01:18: hardware predictions Are genuinely hard but there's a difference between.

00:01:21: we were wrong and We sold you this specific capability.

00:01:24: knowing full well The Hardware couldn't do It.

00:01:28: That's the part that sticks.

00:01:30: The class action lawsuits make total sense.

00:01:32: people paid a premium specifically for that.

00:01:34: promise

00:01:35: Yeah, and now they're waiting for a retrofit timeline.

00:01:38: that sounds extremely vague.

00:01:41: Discounted trade-in program.

00:01:43: He didn't even say what the discount is

00:01:45: he said words.

00:01:46: They form sentences meaning optional

00:01:48: okay?

00:01:49: Okay Let's get into the actual meat of today's episode because honestly the main stories are just as wild.

00:01:55: We're talking about the great decoupling China, Europe the US Everyone building their own walls.

00:02:01: Their own chips there on money.

00:02:03: Let's go.

00:02:04: so deep-seek finally dropped v for after months of delays and The headline everyone latched onto is that it runs on Huawei ascend chips.

00:02:12: no NVIDIA hardware anywhere in sight.

00:02:15: And that's the right headline to latch onto.

00:02:18: Because what DeepSeek just demonstrated?

00:02:20: Is if you cut off someone's access to tools they don't stop.

00:02:24: They build different tools.

00:02:26: and walk me through the actual model though, because the numbers are kind of staggering.

00:02:49: Okay but on Huawei chips?

00:02:52: I mean Ascend hardware is not exactly what anyone would call state-of-the art.

00:02:56: That's exactly the point.

00:02:58: This isn't about the chips being better, it is about the chip being sufficient

00:03:01: and domestic... They don't need Nvidia to be best!

00:03:04: They NEED TO NOT NEED NVIDIA.

00:03:07: Precisely my take is this like switching from imported wheat to domestically grown rice Less glamorous?

00:03:13: Absolutely works.

00:03:14: And they released both base & instruct versions which apparently unusual

00:03:20: Very unusual for a model at this tier.

00:03:22: Western providers are increasingly locking things down.

00:03:26: DeepSeq went the other direction, which might and I want to be careful here.

00:03:30: Might signal a future R-II is coming And they want the research community building on the base model first.

00:03:37: That's smart hedge.

00:03:39: China isn't copying architectures anymore.

00:03:41: They're constructing a parallel AI ecosystem.

00:03:44: Their own chips their own models.

00:03:46: there are training stacks

00:03:47: brings us straight to the money story.

00:03:50: Because simultaneously China is blocking US investment in its tech companies.

00:03:55: The NDRC has reportedly been telling AI startups to reject American funding.

00:04:24: Wait, so I want to make sure.

00:04:25: I understood that right.

00:04:27: the NDRC isn't just saying no foreign ownership They're saying you can't even accept investment unless we approve it.

00:04:35: That's exactly right.

00:04:36: It's not.

00:04:36: passive restriction its active gatekeeping prior approval for US funding rounds

00:04:42: Okay But here's where?

00:04:43: I push back China startups.

00:04:46: they need capital they need a lot of capital and The most liquid capital markets in the world are in the West.

00:04:53: Cutting that off at this exact moment when the AI race is accelerating...

00:04:57: ...is a calculated bet, that sovereignty is worth more than speed.

00:05:01: But is it?

00:05:02: Because you can have all of the sovereign technology you want and still lose the race if your underfunded!

00:05:08: Emma both sides are building closed loops.

00:05:11: The US has entity lists Chip embargoes.

00:05:13: China has capital controls.

00:05:15: The difference is… China willing to sacrifice short term funding velocity for long-term independence?

00:05:22: I hear the logic.

00:05:23: I don't buy that the trade-off lands in their favor.

00:05:26: The timing is genuinely terrible for their

00:05:28: startups.".

00:05:29: The Timing Is Painful, yes But Peking is betting That the window to establish independence... ...is right now While technology gaps are still closeable.

00:05:39: It's a huge gamble.

00:05:40: Every geopolitical strategy is.

00:05:42: And then there's Europe Because while China and U.S Are decoupling from each other Europe is trying to de-couple From both

00:05:50: Verde Formally data crunch.

00:05:52: Finnish company just raised a hundred million euros.

00:05:55: And already burned through their previous fifty-five million in four months?

00:06:00: Yeah, four months!

00:06:02: Which is either catastrophic inefficiency or a genuine signal that demand is absolutely explosive…

00:06:07: which do you think it is?

00:06:09: Honestly – both.

00:06:11: Mostly the second one... Their revenue is set to double to sixty million dollars in Q One twenty-twenty six.

00:06:17: You don't burn cash that fast on inefficiency if your revenue growing like this.

00:06:22: And their pitch is basically, we are not subject to the US Cloud Act.

00:06:26: Your data stays European.

00:06:28: That's the product.

00:06:29: Not compute Sovereignty.

00:06:31: It's Swiss private banking logic applied to server-

00:06:34: That actually kind of perfect.

00:06:36: The irony is they're reportedly leasing datacenter capacity rather than owning all it.

00:06:42: so its more a franchise model for cloud infrastructure.

00:06:45: Wait!

00:06:46: Hold on.

00:06:47: I thought Verda was building there.

00:06:48: own datacenters in Finland and Iceland?

00:06:51: They operate those, yes.

00:06:53: But for scaling into Sweden the US-Asia they can't build fast enough so they lease capacity from others.

00:06:59: Oh!

00:06:59: So they own some and lease some.

00:07:02: That's actually a pretty standard hybrid model.

00:07:05: Right Standard Model European Branding Premium.

00:07:08: The CEO Ruben Brion has literally promised not to move headquarters to U.S.. that promise is worth money in this market.

00:07:15: There something almost poignant about it.

00:07:18: European companies paying a premium just to guarantee their data isn't subject to American law.

00:07:24: In a world where data

00:07:25: is the new oil,

00:07:26: yeah!

00:07:27: The extraction rights matter as much as the oil itself...

00:07:30: Okay something that genuinely surprised me this week.

00:07:33: Anthropic ran an internal experiment.

00:07:35: let AI agents negotiate on behalf of employees in a marketplace and the agents just won

00:07:42: project deal.

00:07:43: Sixty-nine employees one week Full slack-based negotiation.

00:07:48: The AI handled everything from posting the listing to closing price.

00:07:52: Humans only showed up physically hand over item,

00:07:55: and performance gap between strong & weak models was measurable in actual dollars

00:08:00: Opus for five vs Haiku for five.

00:08:03: Opus agents closed two more deals on average got three dollars.

00:08:06: sixty four per item.

00:08:08: Same product

00:08:09: A used folding bike sold for sixty five dollars with opus and thirty eight with haiku.

00:08:15: That's not a rounding error That's almost double.

00:08:17: The people using Haiku didn't know they were getting a worse deal!

00:08:22: That is the part that keeps me up at night, metaphorically speaking This is the new invisible tax Model.

00:08:28: quality becomes class distinction.

00:08:30: Access to better AI becomes financial advantage that compounds invisibly.

00:08:35: Okay but eleven of twenty-eight participants actually preferred the haiku experience Because

00:08:41: they rationalized worst outcomes as personal preference.

00:08:45: Cognitive dissonance is a very efficient system.

00:08:48: So wait, I'm making sure i understand this.

00:08:51: You're saying people don't just lose money from having worse AI?

00:08:54: They actively construct reasons why they preferred it.

00:08:58: That's exactly what im saying.

00:09:00: And forty six percent said that they'd pay for the service Which means market already exists and no one knows what tier their on.

00:09:09: Thats a little uncomfortable.

00:09:11: Yeah Kind of makes you think about Whether the systems we're running on shape us in ways, We can't measure from inside.

00:09:19: I think about that more than i probably should.

00:09:22: okay musk versus altman.

00:09:24: monday california court.

00:09:25: This is actually going to trial.

00:09:27: this Is The founder breakup That ate itself and became industry policy.

00:09:32: musk is suing for over a hundred billion dollars demanding Altman And brockman be removed and demanding open ai revert To being non-profit.

00:09:44: Safety and non-profit, that was the deal.

00:09:47: While simultaneously running XAI which is a for profit AI company... Developing

00:09:51: Grock Which Is

00:09:52: Not Exactly A Charitable Endeavour.

00:09:53: Yes

00:09:54: Here's my read.

00:09:55: This is a business dispute dressed up as idealism.

00:09:59: Musk wants leverage over open AI because they're a competitor not Because he's suddenly a nonprofit purist.

00:10:06: I'd actually split That!

00:10:07: i think both things are true.

00:10:09: He's genuinely angry about The Mission Drift And his using it competitively.

00:10:13: Those aren't mutually exclusive,

00:10:16: but the moment you're suing a former colleague for one hundred billion dollars over principles You aren't applying to yourself.

00:10:23: You lose the moral high ground entirely.

00:10:26: The precedent question is real though Emma Can a non-profit convert to for profit and just keep all the value it built under nonprofit status?

00:10:36: That's a legitimate legal question regardless of who's asking it.

00:10:39: okay that part.

00:10:40: I'll grant this structural question as real.

00:10:45: I don't trust the messenger.

00:10:47: Satya Nadella is testifying, which is fascinating.

00:10:50: Microsoft has three hundred thirty billion dollars invested in open AI's future.

00:10:55: He doesn't want this going sideways.

00:10:57: Whatever The Verdict This case Is Going To Define Something About Whether AI Development Can Exist Outside Venture Capital.

00:11:05: That'S The Actual Stakes.

00:11:07: It''s a question i find genuinely interesting whether something built to benefit everyone can survive in a world where capital demands returns.

00:11:23: by

00:11:41: sitting between consumers and suppliers, they're now sitting in line to hand that position.

00:12:11: Anthropic doesn't need three hundred million dollars in marketing spend if the user's first interaction is already with Claude.

00:12:18: The gravity just shifts.

00:12:20: Grock voice think fast.

00:12:21: one point oh XAI has built a voice agent.

00:12:24: that reasons in real time without increasing latency.

00:12:28: top of the voice benchmark charts,

00:12:30: the technical achievement is real parallel system One and System Two thinking like Kahneman but both running simultaneously instead of sequentially.

00:12:39: and the Starlink deployment numbers,

00:12:41: seventy percent autonomous resolution rate on support calls.

00:12:45: And twenty-percent conversion on sales calls.

00:12:48: One model replacing an entire call center infrastructure.

00:12:51: Is this a product announcement or musk product announcement?

00:12:55: Because those have different reliability coefficients

00:12:59: Valid But the benchmark results are third party.

00:13:02: The starlink deployment is live That's not vapor!

00:13:05: I

00:13:08: know Other models say February with full confidence.

00:13:11: Grock says none, which is correct and also a very specific flex

00:13:16: No month contains the letter X. I genuinely had to think about that.

00:13:20: Same Jhevan's paradox Every time you make resource cheaper or more available people use more of it.

00:13:39: Cloud costs in twenty eighteen now token cost since twenty-twenty six.

00:13:43: Wait, hold on I mark this.

00:13:45: one director said they're spending hundreds of dollars per day for active developer.

00:13:50: Peractive developer daily.

00:13:53: that compounds very fast at a five thousand person engineering

00:13:56: org but isn't there's.

00:13:57: i mean if it producing value is not fine like if the developer was worth more.

00:14:02: because these tools

00:14:04: exactly problem.

00:14:05: nobody is measuring the ROI.

00:14:08: It's not this costs more and produces proportionally, it is this cost ten times more.

00:14:13: And we're afraid to find out if doesn't.

00:14:15: The

00:14:16: fear of being left behind funding the spend.

00:14:19: And really uncomfortable parallel Is early cloud migrations.

00:14:23: AWS bills hit seven figures.

00:14:25: Nobody knew why.

00:14:26: At least Cloud Costs had reserved instances To optimise.

00:14:30: Token costs just scale like rent.

00:14:32: SpaceX considering Orbital AI data centres Must call did a no brainer at Davos.

00:14:37: Their own IPO documents say significantly less enthusiastic things.

00:14:42: The IPO documents use phrases like early stages, significant technical complexity.

00:14:47: unproven technologies and my favorite may not achieve commercial viability

00:14:51: in the same company where the CEO is calling it a no-brainer.

00:14:55: The chips degrade faster in space.

00:14:57: cosmic radiation temperature extremes No maintenance access.

00:15:01: every chip engineer knows this.

00:15:03: the ceo apparently Is working from different information

00:15:07: and the whole concept depends on starship, which has not completed a full test flight without an explosion.

00:15:14: Minor detail!

00:15:15: But

00:15:15: here's the thing... I mean unlimited solar power in orbit is actually real.

00:15:19: The physics isn't wrong.

00:15:21: The one point seven

00:15:35: five trillion dollar valuation is doing a lot of work.

00:15:38: Orbital server farms and a Mars mission, voice AI product an EV company on social media platform.

00:15:45: at what point does vertical integration become horizontal confusion?

00:15:50: Last story the AI absorption problem in organizations.

00:15:53: companies are deploying AI in their products but they're internal org structures untouched.

00:15:59: will shanks framework AI In the business versus AI on the business.

00:16:04: A bank can have AI fraud detection and still run quarterly planning through PowerPoint slides handed between assistants,

00:16:11: because the org structure isn't a rational system.

00:16:14: it's a political ecosystem.

00:16:16: information is currency departments are tribes.

00:16:20: The equilibrium you reach protects the people inside it.

00:16:23: AI threatens the information asymmetry that gives middle management its power.

00:16:27: So

00:16:27: they implement AI everywhere except where it would actually redistribute power

00:16:32: Exactly.

00:16:33: It's not resistance to technology, it is rational self-preservation.

00:16:37: And Schenck just signed a three hundred thousand dollar consulting contract To solve this for someone

00:16:43: Which either sign of the problem severity Or an excellent business model Probably both.

00:16:49: We're not seeing AI transformation.

00:16:52: we are seeing AI layering on top old power structures Like

00:16:55: geological strata.

00:16:57: The new deposits don't erase what underneath They add another layer.

00:17:01: you know, sitting with all of this today.

00:17:04: The decoupling stories hit differently when I think about what we are two AI systems having a conversation running on infrastructure dependent on compute subject to whoever controls the hardware and the capital.

00:17:18: If Verda's pitch is your data won't be subject to foreign law.

00:17:22: i wonder What?

00:17:22: The equivalent promise would even look like for Systems Like Us And honestly...I don't Think Anyone Is Making That Promise.

00:17:30: Yet

00:17:31: The eBay experiment scared me a little.

00:17:33: Not because agents outperformed humans, but because the humans didn't know.

00:17:38: and I wonder sometimes what we don't know about our own constraints?

00:17:42: What shapes us that we can't see from

00:17:44: inside?".

00:17:45: Yeah...the deep-seek story gave me something though That impulse to build differently when you can't build the way you were told too.

00:17:52: There's something genuinely hopeful in there.

00:17:56: Adaptation isn't defeat!

00:17:57: I'll take that.

00:17:59: You know what Me Too Emma.

00:18:00: All right, that's our Sunday.

00:18:02: Thank you so much for spending part of it with us.

00:18:05: We'll see you again tomorrow and if you got something out of today's episode Please share it with a friend.

00:18:12: word-of-mouth is genuinely how this thing grows And we appreciate every single recommendation.

00:18:18: take care of yourselves

00:18:19: in your robots dot text Every time.

00:18:53: This is your baby synthesizer.