Lab hardware acquisition shift — buy all 8-10 servers at once, Texas DC over Reno, NVIDIA+AMD outreach

Situation

Committed in Nathan 1:1 5/21 to a hardware acquisition strategy shift: buy all needed lab hardware at once (8-10 servers) instead of one-per-quarter. Cost is not the barrier; speed of development is. Reno office ruled out (1Gbps shared, no after-hours support, 5-6 server power/cooling cap). Texas DC preferred — to be re-evaluated against Reno after Nathan defines the full hardware list. Peter to email Scott at NVIDIA (H100/B200/B300) and discuss strategy with Steve directly. Nathan to email AMD for equivalent GPU hardware. Hardware sits inside the broader CI/CD acceleration goal — shift to Koji for automated signed kernel builds in a secure enclave to handle frequent builds and zero-day vulnerabilities. Hardware-acquisition shift is what unblocks the Koji shift.

Reasoning

Upstream cause: the 5/12 three-pillar CVE overhaul committed to tooling = Koji-based automated signed builds in a secure enclave. Koji needs hardware. Quarterly purchasing means Koji-by-late-2027, which is unacceptable given the slip pattern showing CVE-response already eating release deadlines. Capital deployment as a velocity lever — same pattern as Rakuten front-loading investment to compress timeline, but applied to internal capacity rather than customer commitment. Cost-is-not-a-barrier is deliberately prioritizing time-to-velocity over capex efficiency, consistent with the same-week Rippling time-over-cost decision (D8). Reno-as-unsuitable is structural not tactical (1Gbps + no after-hours + 5-6 server cap are inherent, not solvable by waiting) — naming it now closes a debate before it absorbs more time. NVIDIA + AMD in parallel because both internal testing AND customer demos matter; lab hardware that runs modern build pipelines is the same hardware that demos H100/B200/B300 to customers, which ties to the AI-as-a-product positioning.

Additional Context

Inside the broader CVE three-pillar overhaul from 5/12 (process + tooling + strategic kernel review). Koji enablement has been gated on hardware for ~10 days. AWS Mirror Manager unblock for Steve (the 5/12 commitment to chase Greg) is still open — and Greg has not acted, per Sarah 1:1 5/21. Hardware decision moves regardless of AWS state — the internal lab does not depend on AWS access.

Observed Evidence

Direct Nathan 1:1 summary quotes. Three Peter-owned action items: email Scott at NVIDIA, discuss hardware strategy with Steve, schedule multi-hour CI/CD working session with Nathan and Justin. Nathan-owned: define target hardware list, evaluate Reno-vs-Texas, email AMD.

Matching Patterns

Confidence Breakdown

Reasoning Depth Analysis