Infra-curious SWE / ML-platform engineer tracking the SpaceX IPO as a compute-capacity and career signal.

Audience Profile

• **Age / Experience:** Mid-career, 3–10 years building/operating systems.
• **Current role:** Software / ML-Platform Engineer (infra-curious), at a startup, scale-up, or big tech.
• **Top pain points:**
• Which suppliers actually benefit vs which is hype
• Separating the source-traced read from the 'SpaceX is an AI company' take
• Where to work if compute capacity is the binding constraint
• **Top decision blockers:**
• Signal-to-noise in IPO coverage for a non-analyst
• Hard to verify which claims are dated/sourced vs opinion
• Following the Anthropic–Colossus deal without an investing background

What This Segment Needs

• **Information:** A dated, source-traced map of who in the AI-infra supply chain has *filed* numbers vs who is narrative — not IPO hot-takes.
• **Tools:** A side-by-side of compute-exposed employers (silicon, GPU, cluster operators) keyed to verifiable filings and talent reqs.
• **Services:** A job-targeting read that treats *compute capacity as the binding constraint* — where deployed GPUs and funded power actually exist.

Top 5 Companies for You (Fit Score)

| Rank | Company | Score | Why | |------|---------|-------|-----| | 1 | Astera Labs, Inc. | 88/100 | Revenue +93.4% YoY to $308.4M, R&D 40.7% of revenue on UALink/UCIe/PCIe Gen6/CXL — real interconnect stack. Risk: ~90% top-5 customer concentration, China 29%. | | 2 | NVIDIA | 87/100 | Data-center ~$115.2B, +142% YoY; the GPU supplier under Colossus (100k→200k+). ~2–3% attrition. Offset: H20/China export-control charges. | | 3 | Anthropic | 82/100 | Run-rate ~$1B→$5B+ in 2025; $13B Series F at $183B post (2025-09-02); became first external Colossus customer. Not profitable; compute liabilities dwarf revenue. | | 4 | SpaceX | 82/100 | Confidential S-1 filed 2026-04-01, $1.75–2T, ~$75B raise, SPCX list ~6/12. Senior Datacenter Compute Engineer — Colossus reqs. External compute = single tenant (Anthropic). | | 5 | xAI | 79/100 | Colossus 100K H100 built in ~122 days toward a 1M-GPU target; a reported large 2025 raise; infra req 2026-05-10. High reported cash burn; reported leadership change; Memphis gas-turbine permitting challenge (all reported, not confirmed here). |

Deal-Breakers (Your Hard Preferences)

No hard preferences declared for this segment.

How to Evaluate Any Company in this Niche (Checklist)

• [ ] **Growth signals:** Confirm a *filed* revenue/run-rate number with a YoY % and a date (e.g. Astera +93.4%), not a valuation headline.
• [ ] **Comp data:** No comp was supplied for any of these — pull levels.fyi / Blind ranges yourself before negotiating; treat absence as a gap, not a green light.
• [ ] **Learning signals:** Check R&D as % of revenue and open *senior IC* ladders (Astera 40.7%; Anthropic/SpaceX show no Staff/Principal track).
• [ ] **Stability signals:** Look for customer concentration disclosure (Astera ~90% top-5; SpaceX external compute = one tenant) and power/permitting risk (xAI Memphis gas-turbine permitting challenge (reported, not confirmed here)).
• [ ] **Capacity vs commitment:** Separate deployed GPUs with a named customer from "intent to scale" language (SpaceX 1M-GPU = "ambition, not commitment").
• [ ] **Culture signals:** Ask in interviews whether attrition/Glassdoor data exists — for most here it was explicitly *not provided*; demand it.

Reverse-Hype Watch

• **SpaceX:** Memphis ~1M-GPU scale-up flagged "ambition, not commitment" — only the 100K+ tier has a customer.
• **Anthropic:** $30B Azure + ~1GW NVIDIA + up to 1M Google TPU commitments sized far above the >$5B run-rate; no model-training/infra hire in 90d talent signals.
• **xAI:** Colossus is a captive internal asset with no external customer case studies; business signals empty.
• **NVIDIA:** Spectrum-X Photonics announced (GTC Mar 2025) but zero matching optics engineering hires to substantiate it.

Under-reported for this segment: coverage obsesses over valuations and GPU counts but rarely surfaces the two things that actually decide whether the compute job *persists* — single-tenant/customer-concentration dependency, and funded, *permitted* power. A 1M-GPU headline behind an unpermitted gas-turbine fight is a career risk no IPO narrative will flag for you.