Tenstorrent Wins a Global Recognition Award 2026
In the history of semiconductor engineering, one architect’s name appears at the origin of more consequential chip designs than any other: Apple’s A4 and A5. AMD’s Zen. Tesla’s custom AI chip. Intel’s Silicon Engineering Group. All Jim Keller, now CEO of Tenstorrent, is building the most credentialed NVIDIA alternative in the AI chip industry. For raising over $1 billion in an oversubscribed Series D backed by Samsung, Bezos, Hyundai, and Fidelity, building the only AI chip company with RISC-V IP licensing, driving most of its bookings, and delivering the platform that Japan, Korea, and Canada are using to build sovereign AI compute independence, Tenstorrent has earned a 2026 Global Recognition Award.
Technical Innovation and Architecture
The Tensix Core, Tenstorrent‘s AI compute building block, is a mesh-topology processor with five RISC-V processors per core, vector/matrix math engines, and built-in Ethernet interfaces enabling direct torus-network chip-to-chip clustering without external switch infrastructure. The generation roadmap (Grayskull → Wormhole → Blackhole → Grendel chiplet Gen 4) evolves consistently toward modular chiplet integration, with the Grendel architecture separating RISC-V CPU and Tensix AI compute into composable independent tiles under the Open Chiplet Atlas open interoperability standard, the semiconductor equivalent of USB for AI compute tiles.
RISC-V Ascalon CPU IP and Neo AI core IP are licensed to LG, Hyundai, and Japan’s LSTC sovereign 2nm AI accelerator program, with “most bookings from IP deals” confirming ARM-model royalty economics. TT-Forge (open-source PyTorch/JAX/ONNX MLIR compiler in public beta) and TT-Metalium (low-level hardware programming) create the developer ecosystem alternative to CUDA that the AI community has been waiting for since GPU lock-in became the dominant AI hardware constraint.
Market Strategy and Leadership
Founded in 2016, Tenstorrent is led by Jim Keller, the only active technology CEO who personally designed the chips inside the iPhone, AMD’s data center processors, and Tesla’s autonomous AI engine before building his current company. The $1B+ capital stack from Samsung Securities, AFW Partners, Bezos Expeditions, Fidelity, Hyundai, LG, and Baillie Gifford spans three continents, with both financial conviction and commercial procurement commitments — Samsung serving as both the lead investor and a $150M customer. Series E at $3.2B target in 11 months from the Series D close reflects commercial momentum ahead of already-strong expectations.
“Own Your Silicon Future” resonates with AI cloud providers (NVIDIA cost alternative), national programs (Japan LSTC, Korea Samsung/Hyundai), and SoC design teams (licensable IP) — three different customer motivations arriving at Tenstorrent’s open RISC-V architecture as the only platform that resolves their specific form of silicon dependency without creating new lock-in.
Industry Impact and Future Vision
NVIDIA’s 80%-plus AI accelerator market share, H100 supply constraints, and CUDA lock-in represent systemic fragility in global AI infrastructure that every government and industrial deployer has identified as a strategic risk. Tenstorrent’s RISC-V foundation, open chiplet architecture, IP licensing economics, and open-source compiler stack invert NVIDIA’s approach rather than replicating it — building the platform for AI compute independence at every level, from sovereign national programs to personal developer desktops.
At CES 2026, Tenstorrent presented a complete portfolio — Open Chiplet Atlas, TT-Ascalon, Blackhole, TT-QuietBox, and Razer AI box, that no AI chip challenger has assembled in one place. Tenstorrent earns the 2026 Global Recognition Award for giving every organization on the planet a genuine, architecturally open path to owning their silicon future.
Tensix Core: mesh topology AI compute building block with five RISC-V processors per core, vector/matrix math engines, and built-in Ethernet interfaces for torus-network cluster interconnect without switch infrastructure
Chip generation roadmap: Grayskull (Gen 1) → Wormhole (Gen 2, 70B LLM serving confirmed) → Blackhole (Gen 3, large-scale workloads) → Grendel (Gen 4, chiplet architecture separating RISC-V CPU and Tensix AI into composable tiles)
Open Chiplet Atlas: open interoperability protocol for heterogeneous chiplet integration, enabling Tenstorrent’s AI chiplets to combine with CPU chiplets from any compliant vendor
Ascalon RISC-V CPU: eight-wide high-performance RISC-V core, described as the world’s first general-purpose CPU with eight wide RISC-V cores, licensable IP
Neo AI Core IP: productized licensable AI accelerator core (September 2025); LG and Hyundai confirmed licensees
Japan LSTC partnership: RISC-V and Chiplet IP selected for Japan’s sovereign 2nm edge AI accelerator programme
TT-Forge: MLIR-based open-source AI compiler in public beta supporting PyTorch, JAX, and ONNX
TT-Metalium: open-source low-level hardware programming library enabling kernel-level Tensix optimization
CES 2026: full portfolio showcase — Open Chiplet Atlas, TT-Ascalon, Blackhole, TT-QuietBox, Razer compact AI box
$693M+ Series D (December 2, 2024): oversubscribed, Samsung Securities + AFW Partners lead, $2.0B pre-money / $2.6B post-money valuation
$100M strategic up-round (August 2023): Hyundai Motor Group lead, Samsung Catalyst Fund co-lead
Total raised: $1B+ as of Series D close
Series E discussions: $800M at $3.2B valuation, Fidelity leading (November 2025, The Information)
$150M in customer contracts: confirmed from Samsung and Hyundai
“Most bookings from IP deals”: Tenstorrent VP product strategy confirmed, implying ARM-model royalty gross margins
Series D investor roster: Samsung Securities, AFW Partners, Bezos Expeditions, Fidelity, LG Electronics, Hyundai Motor Group, XTX Markets, Baillie Gifford, Export Development Canada, Healthcare of Ontario Pension Plan
Series D oversubscribed: investor demand exceeded capital acceptance threshold
NVIDIA alternative with open architecture: RISC-V foundations + Open Chiplet Atlas + open-source compilers inverts CUDA lock-in at every level
IP licensing model (ARM-like): scales with licensee production volume without requiring Tenstorrent manufacturing capacity
Three distinct customer addressable markets: cloud/inference (chip sales), national sovereign programmes (IP licensing), and SoC design teams (IP licensing) — diversified revenue concentration across market types
Samsung three-layer relationship: equity investor + lead Series D investor + customer = maximum commercial alignment
Jim Keller’s unmatched track record: Apple A4/A5, AMD Zen, Tesla AI chip, Intel SVP — the only active CEO with multiple once-in-a-decade chip designs before founding current company
Valuation step-up: $2.6B (December 2024) → $3.2B target (November 2025) in 11 months confirming accelerating commercial momentum
TT-Forge public beta: any PyTorch/JAX/ONNX developer can compile and run models on Tenstorrent hardware via standard AI frameworks — no CUDA re-skilling required
TT-LoudBox workstation: four Wormhole n300s cards, eight processors, 96GB memory pool, confirmed 70B parameter LLM multi-user serving in production via Kubernetes
TT-QuietBox and Razer compact AI box: full-system configurations from data center to personal desktop, democratizing production-grade AI compute
PCIe acceleration cards: individual engineer access to Wormhole and Blackhole compute for model development and small-team inference
Open-source community: active GitHub, Discord, and public beta feedback channel for TT-Forge making Tenstorrent architecture accessible to the global developer community
Koyeb cloud partnership: Tenstorrent-powered inference accessible through cloud API without hardware procurement
Open-source philosophy: TT-Forge and TT-Metalium are publicly available, enabling academic research, independent development, and community contribution without proprietary license fees
RISC-V open ISA: architectural foundation that no company owns, enabling sovereign semiconductor development in any country without royalty payments to US chip IP holders
AI compute democratization: Razer compact AI box and PCIe cards extend production AI compute access to individual developers and enthusiasts at personal computer price points
Open Chiplet Atlas as industry standard: an open interoperability protocol that benefits the entire semiconductor ecosystem rather than Tenstorrent alone, fostering a heterogeneous chiplet ecosystem that creates more competitive options for all buyers
Canadian technology leadership: Export Development Canada and Healthcare of Ontario Pension Plan as Series D investors confirm Canadian institutional support for domestic AI chip sovereignty
