The Silicon Revolution

How Tenstorrent is Waging an Open-Source War in the Post-Microchip Era

The 50-year reign of the general-purpose microchip is over.1

This isn't a forecast of silicon's obsolescence. It's a strategic observation that the foundational assumption of the computing industry—that a single, monolithic, general-purpose CPU would get predictably faster every two years—has crumbled.2 Intel’s "tick-tock" model, the metronome that powered Moore’s Law and an entire generation of innovation, is dead.2

In its place, a new, far more chaotic and specialized era has begun. Nvidia CEO Jensen Huang calls this a "platform shift".5 We are moving from retrieval-based computing (fetching a file, running an application) to generative-based computing (creating novel text, images, and proteins).5

This new workload, broadly defined as Artificial Intelligence, doesn't run efficiently on old architectures. A general-purpose CPU is a sequential processor, designed to execute one complex task at a time with incredible speed.6 AI workloads are massively parallel, requiring thousands or millions of simple calculations to be performed simultaneously.6 Using a CPU for AI is, as one analyst put it, "like using a sports car to haul a shipping container".11

Using a CPU for AI is, as one analyst put it, "like using a sports car to haul a shipping container".

This fundamental mismatch created a power vacuum, and Nvidia—whose Graphics Processing Units (GPUs) were, by historical accident, designed for the parallel task of rendering graphics—was perfectly positioned to fill it.9 Nvidia has since built a multi-trillion-dollar fortress, not just on its powerful chips, but on CUDA, its proprietary software ecosystem that locks in the entire AI industry.13

Nvidia won the first battle of the new computing era. But its victory, built on a closed, proprietary, high-cost model, has created the exact conditions for a revolution. In this new, fragmented landscape, a startup is waging a brilliant, asymmetric campaign to unbundle the king. That company is Tenstorrent.15

The "Anti-Nvidia" Playbook

Led by legendary chip architect Jim Keller—the mind behind AMD's Zen architecture, Apple's A-series chips, and Tesla's first self-driving computer—Tenstorrent is the antithesis of Nvidia.10 Its entire strategy is a bet against the proprietary, high-margin, closed-garden model and for an open, commodity-driven, disaggregated future.

This "anti-Nvidia" playbook is visible in every decision Tenstorrent makes.

First, Tenstorrent is attacking Nvidia's notorious software "moat." CUDA is Nvidia's greatest strength, a 15-year-old software stack that developers are deeply invested in.13 Tenstorrent knows a frontal assault is impossible. Its counter-move is a "fully open source software stack," including projects like TT-BUDA and TT-Metalium.19 The goal is not to beat CUDA overnight, but to attract a community of developers and customers frustrated by Nvidia's "vendor lock-in."22

Second, it is rejecting proprietary, expensive hardware. High-end AI systems are defined by two components that create massive bottlenecks in cost and supply:

1. Interconnects: Nvidia's high-end GPUs use NVLink, a proprietary, high-bandwidth, and very expensive interconnect to link chips together.21 Tenstorrent’s "Wormhole" chips, by contrast, have standard, commodity 100Gbps Ethernet ports built directly onto the silicon.20 Keller’s bet is that you can "make big computers from small ones" using the same standard, scalable, and—most importantly—cheap Ethernet that powers the rest of the internet.23

2. Memory: The most powerful AI accelerators all rely on High Bandwidth Memory (HBM), an extremely fast but extremely expensive and supply-constrained component.23 Tenstorrent’s chips are designed to use standard, commodity GDDR memory, the same found in consumer graphics cards.20 This is a massive cost and supply-chain advantage, allowing them to "avoid HBM's cost and limited availability".23

This philosophy of openness is embedded in the very architecture of their chips.

RISC-V: The Revolution Inside the Chip

The secret to Tenstorrent's architecture is the "Tensix Core," the processing grid that forms the heart of its Grayskull and Wormhole chips.20 Each of these Tensix Cores is a "chip-within-a-chip," containing a high-density math unit, a SIMD engine, local memory, and, most crucially, five programmable "baby" RISC-V cores.26

These RISC-V cores are the key. RISC-V is a free, open-source instruction set architecture (ISA)—a direct, existential threat to the proprietary, royalty-based model of ARM.30 In Tenstorrent’s design, these RISC-V cores aren't the main engine; they are "job controllers".34 They manage the flow of data and orchestration, "decoupling data movement from compute".35 This gives the chip extreme flexibility and a powerful balance of performance and energy efficiency.27

By building its core architecture around an open-source standard, Tenstorrent is sending a clear message: the future is not proprietary.36 This decision is the technical expression of its business model, and it's the foundation of the company's grand strategy.

By building its core architecture around an open-source standard, Tenstorrent is sending a clear message: the future is not proprietary.

Tenstorrent's Grand Strategy: Arming the Rebels

Jim Keller knows he cannot out-Nvidia Nvidia. A head-to-head chip war against a $3 trillion incumbent with a 15-year software moat is unwinnable.10

It is pursuing a brilliant, dual-pronged strategy:

1. Sell Chips (The "Demo"): The company sells its own Wormhole-based AI accelerator cards.20 This serves as a "demo," proving the technology works at scale, building developer mindshare, and establishing a beachhead in the market.25

2. License IP (The "Endgame"): This is the real, high-leverage, long-term play.38 Tenstorrent is licensing its core intellectual property—its "Tensix" AI cores and its high-performance "Ascalon" RISC-V CPU cores—to other companies.17

This dual strategy is a masterstroke. The chip business proves the IP is road-tested38, while the IP-licensing business scales Tenstorrent's technology far beyond its own manufacturing capacity. Keller isn't trying to be the next Nvidia; he's positioning Tenstorrent to be the "ARM of AI."

He is, in effect, arming the rebels.

In the new, disaggregated era, every major corporation and nation-state shares a common strategic fear: a critical dependence on Nvidia. From automotive and smart-device manufacturers to hyperscale cloud providers and even entire nations, no one wants their future to be dictated by a single, proprietary supplier. They are all desperately seeking an alternative.14

Tenstorrent is giving it to them.

This "anti-Nvidia" coalition is already forming, with Tenstorrent at its center.

• Japan: In a massive, multi-tiered deal, Japan’s national semiconductor initiative (LSTC) and its new foundry, Rapidus, have selected Tenstorrent. Tenstorrent will license its RISC-V and chiplet IP to co-design a new, cutting-edge 2nm edge AI accelerator.17 This is an incredible vote of confidence, tying Tenstorrent's future directly to a G7 nation's semiconductor sovereignty.

• LG Electronics: The Korean electronics giant is expanding its partnership with Tenstorrent to build custom AI System-on-Chips (SoCs) for its next-generation smart home appliances, AI-driven solutions, and future mobility projects.42

• Moreh: This Korean AI software startup has entered a strategic partnership with Tenstorrent with the explicit goal of "challeng[ing] NVIDIA" in the data center.10

These partners—nations, global corporations, and startups—are the first members of an alliance built on Tenstorrent's open, licensable, and high-performance IP.38

The microchip era, defined by one-size-fits-all monolithic chips, is over.1 The new era is one of "heterogeneous compute"—a "Lego" model of specialized chiplets17 and open-source foundations.39 Nvidia has built an empire, but it's a closed, walled garden in a world that is rapidly demanding openness and choice.

Tenstorrent has correctly identified that you cannot beat a fortress by laying siege to the front gate. You win by arming every other person in the kingdom with the tools to build their own castles. And that is exactly what it is doing.

Footnotes

1. https://www.nber.org/system/files/working_papers/w24553/w24553.pdf

2. https://www.nber.org/system/files/working_papers/w24553/w24553.pdf

3. https://linexplore.com/intel-the-tick-tock-empire/

4. https://www.rev.com/transcripts/gtc-keynote-with-nvidia-ceo-jensen-huang

5. https://www.rev.com/transcripts/gtc-keynote-with-nvidia-ceo-jensen-huang

6. https://www.geeksforgeeks.org/cloud-computing/difference-between-sequential-and-parallel-computing/

7. https://www.servermania.com/kb/articles/parallel-vs-sequential-vs-serial-processing

8. https://blog.equinix.com/blog/2025/11/05/gpus-vs-cpus-demystifying-hardware-processors/

9. https://dev.to/nayanraj-adhikary/why-cpu-was-not-enough-need-for-gpu-in-the-picture-of-ai-3g09

10. https://www.starburst.io/blog/parallel-vs-sequential-processing/

11. https://backup.education/showthread.php?tid=4336

12. https://blog.purestorage.com/purely-technical/cpu-vs-gpu-for-machine-learning/

13. https://www.reddit.com/r/LocalLLaMA/comments/1jvx7kj/who_is_winning_the_gpu_race/

14. https://www.aranca.com/knowledge-library/articles/investment-research/the-rise-of-custom-ai-chips-how-big-tech-is-challenging-nvidias-dominance

15. https://www.avaq.com/technology/top-10-ai-chip-startups-to-watch-in-2025

16. https://counterpointresearch.com/en/insights/ai-chip-start-ups-can-domain-specific-chips-impact-nvidias-dominance

17. https://tenstorrent.com/en/vision/tenstorrent-risc-v-and-chiplet-technology-selected-to-build-the-future-of-ai-in-japan

18. https://tenstorrent.com/en/vision/ai-software-startup-moreh-partners-with-ai-semiconductor-company-tenstorrent-to-challenge-nvidia-in-

19. https://www.youtube.com/watch?v=GIyzQpOyNno

20. https://tenstorrent.com/hardware/wormhole

21. https://tenstorrent.com/hardware/wormhole

22. https://tenstorrent.com/en/vision/tenstorrent-risc-v-and-chiplet-technology-selected-to-build-the-future-of-ai-in-japan

23. https://xpu.pub/2024/07/25/tenstorrent-wormhole/

24. https://tenstorrent.com/en/vision/community-highlight-tenstorrent-wormhole-series-part-1-physicalities

25. https://xpu.pub/2024/07/25/tenstorrent-wormhole/

26. https://docs.tenstorrent.com/pybuda/latest/hardware.html

27. https://arxiv.org/html/2505.06085v3

28. https://arxiv.org/html/2505.06085v1

29. https://arxiv.org/html/2505.06085v1

30. https://www.design-reuse.com/news/202528993-are-open-source-risc-v-cpus-a-threat-to-arm-holdings-business-/

31. https://patentpc.com/blog/the-rise-of-risc-v-is-it-a-threat-to-arm-and-x86-market-growth-stats

32. https://www.wevolver.com/article/risc-v-vs-arm

33. https://www.stromasys.com/resources/risc-v-vs-arm-processors-comparative-analysis/

34. https://en.eeworld.com.cn/news/qrs/eic701158.html

35. https://en.eeworld.com.cn/news/qrs/eic701158.html

36. https://en.eeworld.com.cn/news/qrs/eic701158.html

37. https://tenstorrent.com/en/vision/tenstorrent-aims-to-build-ai-chips-that-beat-nvidias-best

38. https://www.eetimes.com/tenstorrent-productizes-risc-v-cpu-and-ai-ip/

39. https://tenstorrent.com/ip

40. https://tenstorrent.com/ip

41. https://businessmodelanalyst.com/nvidia-competitors/

42. https://tenstorrent.com/en/vision/lg-and-tenstorrent-expand-partnership-to-enhance-ai-chip-capabilities