Tenstorrent

Tenstorrent designs open-source RISC-V AI processors and licenses chiplet IP for scalable AI compute. Founded by chip legend Jim Keller; raised $200M+ including strategic from Hyundai and Bosch. Wormhole and Blackhole chips announced.

Tenstorrent combines AI silicon with open RISC-V architecture, positioning as the 'Linux of AI chips' against proprietary stacks.

tenstorrent.com AI Accelerators

Canadaprivate

At a glance

Total raised

$913.7M

last round series d

Headcount

—

no public estimate on record

Headquarters

Toronto

Canada

Founded

2016

10 years active

Status

Private

AI Accelerators

Latest valuation

$2.6B

Series D · 2024

For subscribers

Subscribe to see the full profile

Including Funding rounds, Bull / Bear thesis, Stock + earnings, Roster changes, Patents, News, and Open roles.

Founding

50% off

/month

91 of 100 spots left

Full

$10

/month

Available once all 100 Founding Member spots are claimed.

Get full access

Already subscribed? Sign in →

Funding history

Total raised

$913.7M

Rounds tracked

Latest stage

Series D

Latest valuation

$2.6B

Date	Stage	Amount	Valuation	Lead investors
Dec 2, 2024	Series D	$693M	$2.6B	Samsung Securities, AFW Partners
May 1, 2021	Series C	$200M	$1B	Fidelity Management & Research Company
Feb 1, 2019	Series B	$20.7M	—	—

How they make money

Hardware (one-time)Pattern

Tenstorrent sells AI accelerator chips and licenses chiplet intellectual property on a per-unit or per-project basis. Customers include hyperscalers, automotive OEMs like Hyundai and Bosch, and enterprise AI labs that buy Wormhole or Blackhole processors for inference and training workloads. Revenue is recognized upon shipment of silicon or execution of an IP licensing agreement, consistent with standard hardware and licensing revenue recognition. The model is capital-intensive upfront — wafer fabrication, packaging, and tape-out costs are sunk before any revenue arrives. Gross margins on chip sales are generally lower than software, though IP licensing carries a much higher margin profile once the R&D is complete. The hardware business scales with volume and node maturity; the IP-licensing side scales with design-win count and royalty streams from partners who integrate Tenstorrent's chiplets into their own silicon. There is no recurring subscription or services component disclosed in the current public record.

Generated by deepseek-v4-flash · June 2, 2026Analytical summary · not investment advice

Founder backgrounds

Ljubisa Bajic

Co-founder (former CEO/CTO, now advisor)

A chip architect with prior experience at Nvidia and AMD working on GPUs, CPUs, and AI processors. Co-founded Tenstorrent in 2016 to build a new class of AI compute chips, serving as both CEO and CTO during the company's formative years. Attracted Jim Keller — legendary chip architect behind AMD's K7, K8, and K12 architectures as well as Apple Silicon — as CEO in 2023. When Raja Koduri joined Tenstorrent's board, Bajic scaled back his role and transitioned to an advisor capacity. He subsequently co-founded Taalas, a new AI chip startup, alongside early Tenstorrent engineering leaders.

Ivan Hamer

Co-founder

An early Tenstorrent engineering leader who co-founded the company in 2016 alongside Ljubisa Bajic and Milos Trajkovic to develop a novel AI processor architecture. Contributed to the hardware engineering and architectural foundation of Tenstorrent's Grayskull and Wormhole AI accelerator chips, which use a RISC-V-based tile architecture with a NoC (network-on-chip) fabric enabling flexible scaling from single chips to large multi-chip inference and training clusters. Tenstorrent has raised over $700 million and counts Jim Keller as CEO and Raja Koduri as a board member.

Milos Trajkovic

Co-founder

Co-founded Tenstorrent in 2016 alongside Ljubisa Bajic and Ivan Hamer, contributing early chip design and architecture expertise to the company's AI processor development. Tenstorrent's chips — including Grayskull and Wormhole — employ a RISC-V-based mesh architecture optimized for efficient AI inference and training workloads. The company has attracted notable industry figures including legendary chip architect Jim Keller as CEO and former Intel and AMD executive Raja Koduri to its board.

AI bull / bear

Bull case

Jim Keller gives Tenstorrent a gravitational pull that money alone cannot buy. His track record at Apple, AMD, Tesla, and Intel means top silicon talent will join and customers will take meetings. The open RISC-V thesis is potent: NVIDIA's CUDA lock-in is the single biggest vulnerability in AI compute, and a genuinely open alternative could capture the same antimonopoly energy that Linux captured against Solaris and Windows NT. Strategic investments from Hyundai and Bosch signal real industrial demand, not just VC signaling — these are companies that need AI at the edge and in vehicles, precisely where NVIDIA is weakest. The chiplet IP licensing model is a second vector: if AI hardware fragments across dozens of custom designs, Tenstorrent's open building blocks become the standard. Wormhole and Blackhole give them hardware that exists and runs today. In a market where hyperscalers are desperate for a third option, Tenstorrent has the most credible path to being that option.

Bear case

The open-source RISC-V pitch sounds compelling until you try to break NVIDIA's CUDA moat. NVIDIA ships 90%+ of AI accelerators and spends more on R&D in a quarter than Tenstorrent has raised in total. Jim Keller is magnetic, but he is one person — and chip companies live or die on execution across thousands of engineering decisions. The Wormhole and Blackhole chips exist, but no credible independent benchmarks show them competing with NVIDIA's B200 or AMD's MI300X on the workloads that matter. Hyundai and Bosch are meaningful strategic backers, but they are automotive and industrial players — not cloud hyperscalers that drive the majority of AI compute spend. Without a hyperscaler anchor customer, Tenstorrent will struggle to achieve the volume that justifies the enormous cost of leading-edge foundry wafers. Meanwhile, RISC-V software tooling remains immature compared to CUDA's twenty years of optimization. The risk is that Tenstorrent ends up as a well-funded niche player in a market where the winner takes nearly everything.

Generated by deepseek-v4-flash · June 2, 2026Refreshes quarterly · see disclosures

In the news

Tenstorrent · last 30 days

No articles ingested yet for Tenstorrent. Once the hourly news pipeline is live, every article the classifier tags as mentioning this company appears here with its one-line AI summary and sentiment.

Watch

Leadership

Ljubisa Bajic

Co-founder (former CEO/CTO, now advisor)

—

Ivan Hamer

Co-founder

—

Milos Trajkovic

Co-founder

—

Jim Keller

Chief Executive Officer

in 𝕏

Keith Witek

Chief Operating Officer

—

David Bennett

Chief Customer Officer

—

Products

Wormhole
Tenstorrent's second-generation AI processor, built from a grid of Tensix cores that combine compute, SRAM, and packet-based networking so chips scale out natively over Ethernet without external switches. Sold on PCIe cards (n150/n300) and in air-cooled developer workstations and servers, Wormhole targets AI training and inference and is fully programmable through Tenstorrent's open-source software stack.
Blackhole
Tenstorrent's latest-generation standalone AI accelerator, pairing upgraded Tensix Neo cores with onboard 'big' RISC-V CPU cores so each card can run workloads independently. With high-bandwidth Ethernet scale-out and GDDR6 memory, Blackhole (in p100/p150 cards and the Galaxy server, generally available April 2026) targets cost-efficient AI compute as an open alternative to Nvidia GPUs.
RISC-V CPU IP (Ascalon)
Tenstorrent's licensable RISC-V processor IP, including the high-performance out-of-order Ascalon core, which it sells to chipmakers and governments building custom silicon. This IP business, led by veteran architect Jim Keller, underpins design wins and partnerships (including with Japan's national AI/edge efforts and automakers) and embodies Tenstorrent's 'open' strategy of pairing RISC-V CPUs with its AI accelerators.

Most recent patents

19 patents on file, but none with both an extractable figure and an abstract on Google Patents yet.