The trillion‑dollar valuations are still moving, but the business models underneath them are breaking faster than the hype cycle can patch. AI pricing is built on cheap inference and patient capital, and this week both assumptions took real heat.

The labs are burning through compute. The alternative plays are raising to undercut them. And the researchers shaping the next wave are openly questioning whether today's architectures can deliver what the cap tables are pricing in.

What's left isn't chaos, it's clarity around which bets are actually durable when subsidy runs out.

This guide from Deel breaks down how to build one global hiring system. You’ll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.

Download the free guide today

Baseten is raising a large round to make AI models that cost less than today’s leaders. Reports say the company targets $1.5 billion to build lower-cost options to systems from OpenAI and Anthropic. A second report describes the same push as pursuing cheaper alternatives with a focus on inference costs and pricing.

The least obvious ripple is on cloud and GPU brokers, who could see pressure if cheaper inference shifts demand to smaller, more efficient stacks. Enterprises may accelerate multi-model strategies to avoid lock-in, forcing incumbents to cut prices or bundle more. Tooling vendors in evaluation, observability, and routing could benefit as buyers test and switch between models more often.

If you build AI products, recheck your unit economics: latency targets, throughput per dollar, and quality at a given token price. Design for model-agnostic routing and negotiate flexible contracts instead of heavy commitments, since price cuts could reset your gross margins. If you sell infra or dev tools, prioritize benchmarks, SLAs, and clear cost calculators that prove savings in real workloads, not just synthetic tests.

Your phone camera is lying to you.

And honestly… it’s kind of genius.

On paper, smartphone cameras now look almost identical to “real” cameras: 48MP sensors, f1.6 lenses, advanced image processing, cinematic modes. So why do pro cameras still feel different? And more importantly… would anyone actually notice on Instagram?

In this episode, Caya puts it to the test with a fake honeymoon across Costa Rica — comparing the same photos, same settings, same edits, shot on an iPhone and a Sony camera.

Along the way, we uncover how smartphone cameras actually work: Bayer filters, Quad Pixels, computational photography, sensor size, focal length equivalence, depth of field, and why Apple’s “48MP” claim is technically true… but also a little misleading.

This isn’t just a camera comparison.

It’s about how smartphones use software to simulate reality — and why your brain often can’t tell the difference.

Yann LeCun just turned the long-running Musk feud into a macro thesis. In a new interview, he labels xAI a failure and argues Musk can’t attract top researchers after the founding team walked, leaving a giant but underutilized compute machine.

LeCun points to xAI’s trillion‑dollar valuation, heavy infra losses, and reliance on renting Colossus capacity as evidence that “frontier” labs are funding usage with investor cash. He predicts AI bubble rupture unless leaders raise prices or cut costs, while AMI Labs quietly raised at a single‑digit billion valuation to chase “world models.”

For anyone building on top of these stacks, the takeaway is simple: AI economics are not settled. Business models that only work at today’s subsidized token prices are exposed if labs must reprice core services to survive.

Scale your growth: Join names like HubSpot, Ahrefs, and Deel. Put your product in front of 85,000+ startup founders. Let's talk.