Quick Hits — September 5, 2026
The late round: Broadcom turns VMware into an AI factory with governance bolted on, DigiCert's own survey shows post-quantum work stuck in first gear, and a 24-minute deep dive into the hash tables inside Go's most-used data structure.
Broadcom used VMware Explore to launch an AI Factory aimed squarely at the on-premises buildout — and paired it with a control plane for agents. VMware AI Factory bundles Cloud Foundation with certified servers from Dell, Cisco, Lenovo and Supermicro plus a customer's choice of accelerators, and Broadcom claims its automation cuts the path from bare-metal server to first served model from weeks to hours; a new MetalSoft integration handles the physical provisioning. The agent-relevant parts are three services: multi-tenant model sharing through isolated namespaces so teams stop buying duplicate GPU capacity, an AI Gateway that rate-limits tokens and routes prompts between on-prem and cloud models, and secure sandboxes that isolate code an agent writes before it runs. AMD is on board as an accelerator partner — Suresh Andani pitched Instinct GPUs with ROCm as "predictable economics and no per-token pricing," which is the whole pitch in one line. The take: the interesting tell is that a virtualization vendor now sells token governance as a feature — the bill, not the model, is what's pulling enterprise AI back onto private hardware.
DigiCert's own research says the post-quantum migration has barely moved: 87% of organizations are planning or testing PQC, but only 7% have deployed quantum-safe or hybrid certificates at scale — up two points in a year. The survey of 1,001 IT and security decision-makers in the US, UK and Australia found the blockers are practical, not philosophical: legacy complexity (26%), performance impact (19%) and budget (19%), while executive buy-in and standards uncertainty sit at 8% each. DigiCert is pushing the theme at its third annual World Quantum Readiness Day on September 17, and analyst Paul Nashawaty's framing is the one worth keeping — quantum readiness is becoming an application architecture problem, with crypto-agility and automated certificate management moving into CI/CD pipelines rather than a compliance project for later. The take: two percentage points a year is not a migration, it's a pilot that never ends — and the AI buildout is now building encrypted systems that assume the old math holds.
Go's built-in map got a Swiss-table rewrite in Go 1.24, and VictoriaMetrics has published the 24-minute walkthrough of how it actually works. The old map chained overflow buckets off each bucket, so a lookup meant following dependent pointers; the new one allocates groups of eight slots together, keeps a control byte per slot, and filters candidates with SIMD before comparing full keys. Large maps are split into tables of at most 1,024 slots behind a directory indexed by the high bits of the hash, so growth rebuilds one table instead of the whole map. The Go team's microbenchmarks showed map operations up to 60% faster than Go 1.23 with some cases slower, while full-application benchmarks landed around a 1.5% geometric mean CPU improvement. The take: a 1.5% whole-app win is a quiet kind of enormous — it's free performance for every Go service in production, including the ones serving AI inference.
