Cloud saves the AI trade. Open-source could wreck it.

Transcript

Tom Cloud saved the AI trade—and Amazon just sailed up fifteen percent in a session. Open-source though? That could be the wrecking ball.

Marie Welcome to Investment Flash Weekend Edition, August second, twenty twenty-six. I'm Marie, with Tom and Gerald.

Tom Alright, the Journal says Wall Street is finally connecting the AI dots—if you've got a cloud business, the spending spree suddenly looks smart. Amazon's A-W-S, Microsoft's Azure, Google Cloud: tens of billions in capex, now with a visible revenue path.

Gerald Visible revenue path? That's a generous way to say 'we hope they make money'. Nobody's broken out AI revenue yet. It's a leap of faith wrapped in a capex narrative.

Marie Tom—your Amazon buy call from yesterday? That's a victory lap. But Gerald's right, it's sentiment, not proof. The market's pricing cloud as the AI backstop, but if those services don't deliver hard R-O-I, this multiple expansion looks shaky.

Tom Buddy, victory lap indeed! Up fifteen percent in one go—and Microsoft and Alphabet both inflected last session too. Azure and Google Cloud are the conduits. This isn't just vibes, it's a structural monetisation path.

Gerald Structural, you say. So Amazon's trailing P-E on a fifteen percent pop? Probably pushing fifty. No hard R-O-I, but the narrative is structural. I've heard that one—reminds me of the metaverse.

Tom Oh, come on, Gerald. That's not the same. Cloud actually has paying customers. The metaverse was all aspiration.

Marie But the point is, Tom, paying customers for cloud storage and compute doesn't automatically mean they'll pay a premium for AI overlay. If open-source models commoditise the stack, the premium evaporates. And that's the Journal business section's whole story—startups racing to build cheap American alternatives on shoestring budgets, with no V-C support.

Tom Open-source on a shoestring? The hyperscalers are spending billions. They've got the compute and the distribution. These startups can't match scale.

Marie Scale doesn't matter if a model runs on a laptop, Tom! That's exactly the threat. The Journal's Kate Clark reports that without funding, the US might cede the low-cost market to China's DeepSeek. And Nvidia, at thirty-point-eight times trailing earnings, with single-digit year-to-date gains, looks awfully exposed if compute demand suddenly moderates.

Gerald Right, so Nvidia's the only chip play in town, but if the market yawns at ten-billion-dollar training runs because some kid in a garage deploys an open model, those top-end chips become a niche. Marie, you've just outlined my bear case.

Tom Hang on—cloud providers will still need inference capacity. Even if open-source models proliferate, they'll run on Azure and A-W-S, not laptops. Microsoft can distribute them and still make money.

Marie But at what margin, Tom? If the models are free, cloud providers are just renting compute cycles at commodity pricing. That's margin compression, not expansion. Microsoft's up three percent last session but still down one-point-seven year to date—range-bound for a reason.

Gerald And Alphabet, six-point-seven percent up last session on cloud monetisation hopes, but nobody's pricing in that low-cost competitors might undercut their AI models. The stock's riding a narrative that could reverse.

Tom Alright, fair pushback. But you both are overlooking that enterprise customers want integrated suites. It's not just a model—it's the whole platform.

Marie Platform lock-in is real, but history shows open-source undermines it. Think Linux versus proprietary Unix. Cheap alternatives erode pricing power.

Gerald Exactly. And that's why Nvidia's hold rating across both groups makes sense. It's the linchpin.

Tom Fine, Nvidia holds.

Marie So we agree—

Tom Yeah, yeah.

Gerald Spot on.

Gerald Alright, shifting from Silicon Valley to Shenzhen. Hong Kong is raising its full-year G-D-P forecast after a strong first half, and they're launching offshore Chinese government bond futures. The yuan push is real.

Marie This is a structural step. Bond futures add depth to the offshore market, and Financial Secretary Chan is championing yuan internationalisation. Hong Kong equities look cheap—E-W-H at zero-point-eight-nine times book.

Tom I like the low price-to-book, buddy. Plus the G-D-P upgrade suggests resilience. And selling U-S-D-C-N-H on policy support for a stronger yuan? That's a clean pair trade.

Gerald Wow, Tom's suddenly into macro pair trades. What's next—trade-weighted index chat?

Tom Hey, I can read a chart. But to be fair, the S-C-M-P piece underscores a decoupling narrative from mainland wobbles. H-K-E-X should benefit from increased volume.

Marie H-K-E-X gets a direct lift from bond futures—more trading, more fee income. And the G-D-P forecast raise supports the broad Hong Kong equity story.

Gerald Alright, I'm in on those buys. But let's watch actual flows.

Tom Exactly.

Marie Right —

Gerald Now, the Journal's James Mackintosh has a column that's basically my spirit animal: America is great at creating stock market bubbles and shrugging them off. The AI pullback has been almost entirely offset by gains in the rest of the market.

Marie This is crucial. The equal-weight S-and-P 500, R-S-P, is up eleven-point-five percent this year, versus nine-point-three for cap-weight S-P-Y. That's a rotation, not a retreat. The non-AI strength is absorbing the tech pullback.

Tom Two hundred and twenty basis points of outperformance! That's the trade right there—long R-S-P, and it's not even expensive. The rotation has legs.

Gerald But Mackintosh warns that one day the bust will be the big one. Nvidia, at thirty-point-eight times trailing, sits at the center of that froth. Zero margin for error. The non-AI gains are great, but the moment AI-centric names correct sharply, the spillover could be messy.

Marie Yet historically, the market shrugs it off. Concentrated bubble losses don't break the broad index. The equal-weight trade is a way to capture that resilience. Nvidia's hold is just a warning sign, not a collapse call.

Tom This is why Gerald loves Mackintosh—he gets to say 'bubble' and 'froth' in one column.

Gerald Ha — fair enough. But he's not wrong. The S-and-P five hundred has shrugged off plenty of bubbles.

Marie So R-S-P is a buy, and we hold the others.

Tom Our view is the story of two AIs: cloud monetisation on one side, open-source commoditisation on the other. The net result: rotation within tech, not a retreat. R-S-P is outperforming, and Hong Kong adds an Asian resilience leg.

Gerald But the cloud thesis is still a leap of faith, Tom. No hyperscaler has broken out AI revenue. If open-source produces a model that runs on a laptop, the cloud AI premium evaporates. Nvidia's multiple is particularly exposed.

Marie And here's what's missing from today's coverage: US rate sensitivity. The ten-year yield has been creeping up, yet growth stocks are rallying. Either the market thinks AI capex is yield-agnostic, or we're one hawkish Fed minute away from a reappraisal. No one's talking about that.

Tom Rates? Come on, Marie, cloud earnings don't care about the ten-year—they're secular growth, not cyclical. If enterprise adoption is happening, yields don't matter.

Marie They matter for valuations, Tom! If long-duration assets reprice on rate expectations, that fifteen percent Amazon pop could unwind. The cleanest expression is owning the equal-weight rotation with a hedge against open-source disruption. Long R-S-P, and watch Nvidia's reaction to any DeepSeek-like news as a short catalyst.

Gerald So we're back to long R-S-P, short S-P-Y—the easy money there may have been made, but it's still the cleanest. And Marie, your point about missing rate sensitivity is spot on. One hawkish minute and this whole narrative gets repriced.

Tom Alright, I'll give you rates as a risk. But the rotation trade doesn't need rate clarity—it's just about breadth. If the ten-year keeps creeping, non-AI cyclicals could benefit even more.

Marie Now Tom's a macro trader. But fair point.

Gerald I'm surrounded by converts.

Tom Ha — only when it's working.

Marie Alright, alright.

Tom I'll take it.

Marie As always, none of this is investment advice. Do your own research.

Tom Alright, friends, we've covered a lot. Gerald, your bond doom-loop returns, but in reverse—rates up, stocks rally? I'm confused.

Gerald Only Gerald could turn a chip rally into a P-E warning—oh wait, that's me. But seriously, have a great weekend. And if you're just finding us, hit follow on Spotify, or check investmentflash.com for the full digest with charts and sources.

Marie Monday's London Edition at seven-thirty a.m. London time—we'll see you then.