Stack L2

Tenstorrent

Merchant RISC-V AI accelerator; Blackhole GA, open TT-Metalium stack vs hyperscaler ASICs

HQ CA

Fit Score

avg of 4 segments

72/100

Growth

Comp

—

Learning

Stability

Culture

Recent Business Signals

No business signals recorded in the last 90 days.

Strategic Position

What they do best

Shipping an openly programmable merchant accelerator: Blackhole p150a (140 Tensix cores, 32GB GDDR6) hit GA in May 2025 (c5631efa) with a fully open TT-Metalium stack (b9b0709b).

Their bet

Win cost/performance against hyperscaler custom ASICs by riding Samsung SF2 2nm for the next-gen Quasar chiplet (c270298c) plus sovereign-AI compute mandates referenced in the company profile.

Top risk

TT-Metalium/TT-Forge may not reach out-of-box parity for mainstream LLM training/inference, leaving Blackhole GA cards (c5631efa) with no disclosed volume design win by 2026-H2.

Compared to peers

Direct competitor

AMD

AMD ships MI300/MI350 at volume with maturing ROCm; Tenstorrent is pre-volume, betting on open RISC-V and lower cost/performance.

Substitute

Groq

Groq's LPU is inference-only with deterministic latency; Tenstorrent Blackhole is a general train+inference programmable part.

Why someone would join

1.Silicon is real, not pre-tapeout: Blackhole p150a (140 Tensix cores, 32GB GDDR6) reached GA May 2025 (c5631efa) and scaled to 32-chip Galaxy systems by Sept 2025 (b9b0709b) — work is productization.
2.Next-gen 2nm program is staffing now: Samsung SF2 Quasar chiplet partnership signed July 2025 (c270298c) with 2 ASIC physical-design/DFT backend reqs opened April 2026, indicating an active tapeout pipeline.

Recent Hiring (60 days)

physical design
1
place and route
1
timing closure
1
UVM verification
1
AI accelerator verification
1

Reverse-Hype Watch

!Galaxy-class 'scaling to dozens of Blackhole accelerators per box' (2025-09-18) and systems 'target customers seeking an alternative to hyperscaler-internal' silicon, but zero named customer-win or revenue signal in 180d window — capacity/scale-out claims unbacked by customer signals.
!Scale-out capacity claims unbacked by customer signals: Galaxy systems 'scaling to dozens of Blackhole accelerators per box' and systems that 'target customers seeking an alternative to hyperscaler-internal training/inference silicon' — no customer win, design-win, or named production case study appears in any provided signal.
!Capacity/positioning claims unbacked by customer signals: systems 'scaling to dozens of Blackhole accelerators per box' and 'merchant...alternative to closed hyperscaler custom silicon' (2025-09-18), but no customer win, unit-shipment, or revenue signal; data point notes 'no provided signal cites named hyperscaler production case studies'.
!Positioning as 'merchant AI accelerator alternative to hyperscaler custom ASICs' and Galaxy scaling 'up to 32 Blackhole chips' / 'dozens of Blackhole accelerators per box' — but zero customer-win or design-win signals in business_signals_180d; capacity/scale-out claims unbacked by any customer evidence.

→ Get this data as JSONLast updated: May 16, 2026

Frequently asked about Tenstorrent →