Topic

AI ASICs

MarketCompaniesProductsHiringPersonasStrategy

ASICS-PROCURE-LEAD-01

Procurement / supply-chain lead negotiating accelerator allocation at large AI buyer.

Audience

  • · 10-20
  • Current: Hardware Procurement / Strategic Sourcing Lead
  • Pain: NVIDIA allocation visibility past 12 months
  • Pain: Merchant alternative pricing-power asymmetry

Product Needs

(none)

Channels

(none)

Competitor Lens

(none)

Fit Score weights — adjust to your priorities

20%
40%
10%
25%
5%
Top 5 for this segment
  1. 1. AMD45/100
  2. 2. Cerebras Systems43/100
  3. 3. Groq42/100
  4. 4. Rebellions40/100
  5. 5. Tenstorrent40/100

Full Persona Brief

Procurement / supply-chain lead negotiating accelerator allocation at large AI buyer.

Audience Profile

  • Age / Experience: 10-20 yrs (mid-senior strategic sourcing)
  • Current role: Hardware Procurement / Strategic Sourcing Lead at hyperscaler / large AI lab / sovereign AI fund
  • Top pain points:
  • NVIDIA allocation visibility past 12 months
  • Merchant alternative pricing-power asymmetry
  • CoWoS packaging capacity bottleneck flows to accelerator quotas
  • Top decision blockers:
  • Multi-year contractual commitments lock allocation
  • Strategic concentration risk on single vendor

What This Segment Needs

  • Information: booked vs aspirational backlog per vendor (GW / unit count + delivery dates), HBM generation and CoWoS/foundry slot exposure, anchor-customer subordination of your PO.
  • Tools: a scorecard separating contracted backlog (AMD 6GW OpenAI, 50k MI450) from management TAM targets; multi-vendor allocation hedge model.
  • Services: second-source qualification, take-or-pay exposure modeling, export-control force-majeure clause review.

Top 5 Companies for You (Fit Score)

| Rank | Company | Score | Why | |------|---------|-------|-----| | 1 | AMD | 75/100 | Record Q3 FY25 rev ~$9.25B +36% YoY (2025-11-04); booked backlog OpenAI 6GW Instinct (2025-10-06) + Oracle 50k MI450 Q3'26 (2025-10-14). Diversified CPU+GPU; offset by $800M MI308 China charge and OpenAI/Oracle concentration. | | 2 | Cerebras Systems | 72/100 | $1.1B Series G at $8.1B (2025-09-30, Fidelity/Tiger/Benchmark); 4 launches May–Aug 2025, OKC/Minneapolis build-out. Wafer-scale non-commodity supply; G42 single-customer/CFIUS exposure, pre-profit high burn. | | 3 | Groq | 70/100 | $750M at $6.9B post-money (2025-09-17), ~2.5x step-up; Helsinki DC, HUMAIN/PIF ~$1.5B (2025-05-13). No disclosed revenue/paid-token volume; mega-deal concentration; inference-only. | | 4 | Rebellions | 67/100 | Series C $1.4B, ~40% step-up (2025-06-12); REBEL on Samsung 4nm + SK hynix HBM3E (2025-08-26), Arm partner (2025-09-30). No revenue/named customers; HBM3E/4nm allocation contested vs NVIDIA. | | 5 | Tenstorrent | 66/100 | Blackhole GA p150a 140 Tensix/32GB GDDR6 (2025-05-12), Galaxy (2025-09-18), Samsung SF2 Quasar chiplet (2025-07-22). Merchant alt to hyperscaler silicon — hard wedge; no disclosed revenue/customer wins. |

Deal-Breakers (Your Hard Preferences)

No hard preferences declared for this segment.

How to Evaluate Any Company in this Niche (Checklist)

  • [ ] Check growth signals: demand contracted backlog stated in GW or unit count with delivery dates (e.g., AMD 6GW / 50k MI450 Q3'26), reject annual revenue-growth % as allocation proof.
  • [ ] Check comp/price data: none of these 5 disclose price publicly — require binding $/unit and price-step schedule vs NVIDIA Blackwell in the RFQ.
  • [ ] Check supply signals: confirm HBM generation (HBM3E vs HBM4) and CoWoS/SF2 packaging slot reserved against your PO, since packaging gates quotas.
  • [ ] Check stability signals: quantify vendor's own anchor concentration (OpenAI 160M-share warrant, G42, HUMAIN) — your allocation is junior to their warrant/anchor customer.
  • [ ] Check culture/exec signals: ask for written allocation-priority tiering and export-control force-majeure terms (AMD $800M MI308 precedent).
  • [ ] Check contract lock: model multi-year take-or-pay exposure against a 12-month re-bid option before signing.

Reverse-Hype Watch

  • AMD: ROCm/CUDA-substitute "watch-point" unbacked by hiring (5/5 silicon roles, zero compiler/ML-systems) — software-stack maturity risk for buyers.
  • AMD: FAD "35%+ growth 3-5yr" vs "$1T TAM by 2030" is an aspirational management target, not booked revenue.
  • Cerebras: "tens of millions tok/s" capacity unbacked by diversified customers; demand concentrated in G42.
  • Groq: Bell "500 MW build-out target" vs ~7 MW Kamloops live (~70x) — largely aspirational.
  • Tenstorrent: Galaxy scale-out claims with no named design win or production case study.

Under-reported for this segment: allocation priority tiering and anchor-customer subordination. Public coverage hypes headline GW and TAM numbers, but the binding constraint for a buyer is where you sit in the vendor's allocation waterfall behind warrant-holding anchors (OpenAI→AMD, G42→Cerebras, HUMAIN→Groq) and whether HBM4/CoWoS-L slots are physically reserved against your PO. That subordination, plus export-control force-majeure (MI308), is the real deal risk and is invisible in press releases.