Compute Capacity Becomes the Ultimate Bottleneck
Two deals. $1.5 billion per month in combined commitments. Google ($920M/month) and Anthropic (undisclosed, but similar structure) have both signed multi-year agreements with SpaceX for compute capacity — bypassing traditional cloud providers entirely.
This isn’t about cost savings. It’s about guaranteed access when frontier model training is outstripping available GPU supply. The labs that secure compute win; everyone else waits in line. SpaceX’s satellite-ground infrastructure hybrid offers something AWS and Azure can’t: dedicated capacity without contention.
The question for everyone else: what happens when the compute haves and have-nots diverge irreversibly?
Nvidia’s CPU Ambitions
Daniel Lemire reports that Nvidia is proposing a “beast” CPU system for Windows PCs — a move that would extend the company’s AI silicon dominance from data centers to consumer desktops. Details remain sparse, but the implication is clear: Nvidia sees an opportunity to own the entire AI stack, from training clusters to local inference.
For developers, this could mean optimized toolchains that span cloud and edge seamlessly. For competitors — Intel, AMD, Apple — it’s another front in the silicon wars.
EU’s €320 Billion Sovereignty Bet
The European Commission unveiled its Tech Sovereignty Package: €320 billion over 10 years, Chips Act 2.0, the Cloud and AI Development Act (CADA), and an EU Open Source Strategy. It’s the biggest European tech policy move in a generation.
The ambition is explicit: reduce dependence on US and Chinese infrastructure while creating opportunities for “trusted partners.” For New Zealand tech companies, the question is whether to position for the EU lane or wait for US clarity that may never come.
Zeroserve: eBPF Meets Web Servers
A new project called Zeroserve demonstrates what happens when you script web servers with eBPF: zero configuration, kernel-level performance, and the ability to modify behavior without restarting. It’s early, but the approach could reshape how we think about server architecture.