Corp-strategy / supply-chain lead at an L1-L5 supplier or neocloud testing the captive→merchant-compute pattern.

Audience Profile

• Age / Experience: 10–20 yrs; senior Director / BizDev-lead tier
• Current role: Corporate Strategy / Supply-Chain / BizDev Lead (optical/networking/power supplier or neocloud)
• Top pain points:
• Whether 'first external customer' generalizes to a merchant compute market
• Falsifiable triggers (a second external Colossus customer before listing)
• Capacity planning under buyer multi-provider sourcing posture
• Top decision blockers:
• Pattern-matching prior buildouts too hard on a single data point
• S-1 internals not public — read built on reported, not confirmed, figures

What This Segment Needs

• Information: Named-tenant count + dated S-1 disclosures (SPCX public S-1 expected 5/20/2026) to test whether Colossus captive→merchant is a market or a one-off.
• Tools: A demand-tracker tying xAI GPU/networking/power scale-out to your order book, plus a per-customer concentration model.
• Services: Counsel that separates reported-vs-confirmed gaps (power permitting, 1M-GPU "ambition") before you commit capacity.

Top 5 Companies for You (Fit Score)

| Rank | Company | Score | Why | |------|---------|-------|-----| | 1 | Astera Labs | 84/100 | 10-Q Q1 2026 revenue $308M, +93.4% YoY; R&D 40.7% of revenue; ~$1.18B cash, no debt. But top-5 customers ~90%, China 29% under export risk — concentration mirrors your buyer-risk question. | | 2 | Anthropic | 78/100 | ~5x run-rate ($1B→$5B in 2025), $13B Series F at $183B. The named first external Colossus tenant — your single data point. $30B Azure liabilities dwarf the >$5B run-rate. | | 3 | SpaceX | 78/100 | Public S-1 expected 5/20, SPCX NASDAQ 6/12/2026 at $1.75–2T; Anthropic first external Colossus customer (2026-05-06). External revenue rests on one tenant that is diversifying. | | 4 | Coherent | 74/100 | Q3 FY26 revenue $1.81B (+20.6%), capex $547M (+77%) ahead of AI optics; equity $11.0B. Offset: operating cash $10M vs $503M; active 2023/2025 restructuring. | | 5 | xAI | 70/100 | Colossus 100K→1M-GPU target; new Training-Cluster req 2026-05-10. Captive asset, no external case studies; reported Memphis gas-turbine permitting challenge could pressure core power (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)

• [ ] Check growth signals: count *named* external Colossus tenants beyond Anthropic before SPCX lists 6/12/2026 — your falsifiable trigger is ≥2.
• [ ] Check comp data: all 5 report "no comp data"; pull the SpaceX public S-1 (expected 5/20/2026) for segment revenue + tenant concentration, cross-check Levels.fyi for the supplier.
• [ ] Check learning signals: demand R&D ≥ ~15% of revenue and a UALink/UCIe/PCIe Gen6/CXL or RDMA-AI-networking stack — legacy optics ≠ merchant-compute pull.
• [ ] Check stability signals: model top-5 customer concentration; flag any supplier whose external story rests on one named tenant that publicly multi-sources.
• [ ] Check power signals: confirm MW + on-site generation is S-1/permit-confirmed, not "reported" — verify Memphis gas-turbine permitting status.
• [ ] Check culture signals: ask whether a Staff/Principal/Distinguished IC ladder exists — absent in SpaceX/xAI/Anthropic signals.

Reverse-Hype Watch

• Anthropic: $30B Azure + ~1GW NVIDIA + ~1M Google TPU sized far above a $5B run-rate — forward conviction, not customer-backed.
• SpaceX: xAI ~1M-GPU Memphis scale is reported ambition, NOT confirmed; external Colossus revenue = single tenant (Anthropic) that multi-sources.
• xAI: 1M-GPU Colossus 2 target with no external customer case studies; captive asset, not yet market-validated.

Under-reported for this segment: every public read of the captive→merchant pattern hangs on ONE dated data point — Anthropic as "first external customer" (2026-05-06) — while S-1 internals stay private until ~5/20 and power/permitting figures are reported, not confirmed. The metric you actually care about — a *second* externally-sold, named tenant and per-tenant concentration — is exactly what nobody publishes. Treat single-tenant generalization as the null hypothesis until the S-1 prints.