July 31, 2026
TechCrunch's earnings roundup makes the pattern hard to miss: Amazon keeps pouring money into data centers, and investors keep applauding. That's not a surprise so much as a confirmation of where confidence actually sits right now. Nobody is entirely sure which AI products will win, but almost everyone agrees someone will need compute to run them, so the safest bet is on the companies renting out the shovels. For a business reader, the takeaway isn't 'buy cloud stocks' -- it's that the infrastructure layer of AI is maturing faster and with more certainty than the application layer sitting on top of it. That gap matters when you're deciding whether to build your own AI tooling or buy something already running on someone else's infrastructure, a tradeoff we've written about in our build vs. buy framework.
British neocloud Nscale buying Anyscale is a smaller story than Amazon's spending, but it points at the same trend from a different angle. Nscale isn't just selling raw compute anymore -- it's absorbing the software layer, Anyscale's workload-scaling tools, that helps companies actually make use of that compute across data centers. This is the compute-stack version of vertical integration, and it tells you something about where the margin is expected to move. Whoever controls both the hardware and the orchestration software around it gets to capture value at two layers instead of one, and gets to lock customers in more tightly while they're at it. If you're a business running AI workloads across multiple vendors, that consolidation is worth watching, because it usually precedes pricing power shifting away from you. It's the same logic behind why fragmented tool stacks eventually get expensive -- something we've covered in the hidden tax of switching between a dozen SaaS tools.
Meanwhile, on the consumer side, the AI wearable Friend is back with a new voice feature and, notably, a much higher price. Adding a voice interface is a real product upgrade, but charging significantly more for it right as the device tries to prove it has staying power is a risky sequencing choice. Companion hardware like this lives or dies on habitual daily use, and raising the price barrier right when you need more people trying the thing, not fewer, suggests the company is optimizing for revenue per unit rather than growth right now. That's a legitimate strategy if your unit economics were underwater before, but it's also exactly the move a company makes when it doesn't expect volume to save it. I'd bet this pricing move is more about margin survival than confidence in demand, though I'll happily be wrong if the voice feature turns out to be the retention hook the product always needed.
The last thread here is Situational Awareness, the hedge fund founded by a former OpenAI researcher, which had to unwind its leveraged public-market bets after they went badly. But TechCrunch notes it's held onto its Anthropic shares -- a private stake that wasn't subject to the same margin pressure as its public portfolio. This is a useful reminder for anyone evaluating AI-adjacent investment vehicles: leverage on volatile public AI stocks is a very different risk profile than a private stake in a foundational model company, even if both get pitched under the same 'AI exposure' banner. The fund's unwind is a cautionary tale about conflating the two. Whether the Anthropic position ends up vindicating the fund's thesis is still an open question, but it's the one asset here that wasn't forced to sell at the worst possible moment.
If you had to place new money in AI today, would you rather own the compute layer, the applications, or a piece of a frontier lab -- and does today's news change your answer?
Sources