Yesterday's signals, distilled.

Yesterday's signals, distilled, A look back at March 3.

The market sold off “AI trades” while second-tier fabs raised prices and boosted capex. Defense ministries tightened vendor lists while AI labs clarified what “classified” actually means. A robotaxi blew a basic social norm. A viral app went dark because one cloud vendor hiccuped.

The connective tissue isn’t hype or capability. It’s dependency.

Who you depend on for compute. Who your customers depend on for trust. Who your investors depend on for macro beta. And increasingly, who your regulators depend on for plausible deniability.

If your AI roadmap assumes stable inputs, cheap GPUs, tolerant regulators, patient capital, and invisible infra, yesterday was a reminder: the real risk isn’t that models get smarter. It’s that your dependency graph is already too fragile for the systems you’re trying to build on top.

BLUF

At Neue Alchemy, we support leaders navigating inflection points, when tech, capital, and policy converge. If your roadmap is already in motion and you’re pressure-testing execution, we’re open to conversations.

We also reserve capacity for education, SMBs, and mid-market leaders, those starting, mid-flight, or seeking outside perspective before systems harden.

INFRASTRUCTURE / COMPUTE

AI hardware is a seller’s market, and second-tier fabs just confirmed it

Asia’s smaller chipmakers are joining larger peers in hiking prices as AI demand drives projected semiconductor capex up 25% YoY to more than $136B in 2026, per Nikkei Asia.

These are not the flagship foundries setting the tone, this is the second tier realizing they can raise prices and still justify aggressive expansion.

The Bet: AI demand will stay strong enough that even non-leading-edge fabs can both raise ASPs and grow capacity without triggering a demand shock.

So What? AI infra is no longer just constrained at the bleeding edge. When trailing and specialty nodes are repricing and expanding, the entire hardware stack, accelerators, controllers, networking, storage, is repricing with it. If you’re assuming “GPU prices will normalize” in your 2027–2028 models, you’re fighting both capex momentum and vendor pricing power. The structural shift: compute is behaving like energy, a strategic input whose price is set by global demand and capex cycles, not your individual contract.

The Risk: If macro or regulatory shocks slow AI workloads, late-cycle capex could overshoot, flipping bargaining power back to buyers, but with a 2–3 year lag. In the meantime, over-optimistic infra startups that underprice capacity on the assu