Platform Strategy (template + worked outline)
The vision + an executable roadmap. Concepts:
../WARMUP.mdCh. 2, 6. Fill in / adapt to your scenario.
Thesis
Decouple software from hardware (the JD's mandate). The HAL (Phase 09) is the spine; vendor lock-in is the enemy; the market is (a) customers wanting freedom from NVIDIA lock-in and lower cost, and (b) sovereign/regulated customers who need on-prem/air-gapped deployment. The moat is a performant, versioned, conformance-gated, multi-vendor HAL with an OEM ecosystem.
Roadmap (sequenced bets, each with a commercial outcome)
| Phase | Technical | Commercial outcome | Kill criteria |
|---|---|---|---|
| 0–6 mo | HAL + NVIDIA backend + serving (P07) + offline licensing (P11) | first sovereign customer in production | no sovereign pipeline by mo 6 |
| 6–12 mo | AMD backend + conformance suite (P09) + observability/release (P10) | OEM #1 certified; multi-vendor proof | AMD backend < 80% of vendor peak |
| 12–18 mo | Triton/MLIR backend + GPU sharing tiers (P06) + attestation (P11) | multi-vendor GA; tiered pricing; regulated-market expansion | kernel-velocity not improving |
Each milestone is commercial, not just technical. Dependencies are explicit (licensing before sovereign customer; conformance before OEM).
Build vs buy (stated once, defended consistently — Phase 09 ADR)
- Build: the HAL + control plane (the moat).
- Adopt: kernel compiler (Triton/MLIR), inference engines, observability stack.
- Buy: commodity infra, HSM/KMS.
Governance
- Pace governed by error budgets (P10), not just ambition.
- Each major investment carries a commercial thesis (P12 Ch. 6).
- Conformance (P09 Ch. 7) gates every backend; ABI versioning (P09 Ch. 6) gates releases.
Your additions
- The specific first customer/segment and why:
- The riskiest bet and how you'd de-risk it (a hands-on prototype, P12 Ch. 9):
- What you'd cut if you had half the team: