When you buy through affiliate links in our content, we may earn a commission at no extra cost to you. Learn 
As the market size of the AI industry grows, so does the size of new players.
In December 2024, Amazon founder Jeff Bezos invested $700 million in funding into the AI chip startup Tenstorrent.
Tenstorrent sits on a valuation of $2.6 billion, which it plans to use to expand its engineering team and invest in its global supply chain.
The company is aiming to provide an alternative to Nvidia’s popular graphics processing units (GPUs) and plans to release new chips every two years.
With investors including Bezos, Samsung, LG Electronics, and Hyundai jumping on board and customer contracts totaling almost $150 million, the California-based startup has shown lots of promise since it was founded in 2016.
However, going head-to-head with the largest chipmaker in the world will be a big task, particularly when Nvidia’s GPUs are the go-to choice for leading vendors like OpenAI.
That being said, there are a number of elements that make Tenstorrent a must-watch wild card in the artificial intelligence chip market.
Key Takeaways
- Tenstorrent, valued at $2.6 billion, aims to rival Nvidia with open-source AI chips.
- Amazon founder Jeff Bezos invested $700 million in the AI chip startup in December 2024.
- Wormhole AI chips offer modular, scalable performance using RISC-V architecture.
- Tenstorrent’s chips boast lower costs and higher energy efficiency than Nvidia GPUs.
- The startup targets companies seeking alternatives to Nvidia’s high-cost solutions.
An Open Approach to AI Chips
As the generative AI race continues at breakneck speed, Tenstorrent aims to provide an alternative for organizations looking for a more cost-effective and open alternative to Nvidia’s GPUs.
Tenstorrent’s team is led by co-founder and CEO Jim Heller, who has worked for leading companies, including AMD, Intel, and Tesla. It is worth highlighting that Heller worked on the AMD Zen multi-chip architecture and Tesla’s self-driving chip.
His focus has been to create computers to support AI workloads with interoperable architecture that is engineer-friendly. Tenstorrent builds its computers with Tensix processors and Tensix cores.
Each of these cores includes an array math unit for tensor operations, a SIMD unit for vector operations, a network-on-chip (NoC) to transfer data between cores and chips, five baby RISC-v processors (open-source instruction set architecture) to assist the NoC, and 1.5MB of SRAM.
This approach is highly computationally and cost-efficient while offering solid performance on parallel computation tasks.
Likewise, the use of an open-source RISC-v processor makes the cores modular, easier to tweak, and opens the door to greater integration with third-party technologies.
A Look at the Tenstorrent Wormhole AI Processors
One of the biggest selling points of Tenstorrent is its Wormhole high-performance AI chips.
Wormhole chips use the RISC-V architecture and come in two different configurations: Wormhole n150 and Wormhole n300.
The n150 has a single processor with 72 Tensix cores and costs $900 (each core features five RISC-V baby cores).
It also has an add-in card of 12GB of GDDR6 memory. The N300 has two processors with 24GB of GDDR6 memory and costs $1,399. These improvements make the N300 significantly more powerful than the N150.
Tenstorrent aims to present Wormhole chips as a more scalable alternative to other processors on the market, capable of pooling the computing resources of multiple Wormhole processors together to handle more intensive applications when needed.
These chips can be found in Tenstorrent workstations such as the TT-QuietBox, a liquid-cooled desktop workstation with four Wormhole cards that costs $15,000. The TT-QuietBox is primarily designed to run quietly when running AI models.
Can Tenstorrent Take on Nvidia? Tenstorrent vs. Nvidia
How do Tensix processors compare to GPUS from vendors like Nvidia? Tenstorrent claims they are easier to program, more efficient, more scalable, and capable of performing conditional computation.
Tenstorrent claims that its upcoming processors offer performance on par with Nvidia’s AI GPUs, partly due to moving away from high-cost, high-bandwidth memory (HBM) and using GDDR6 as an alternative.
“You can’t beat Nvidia if you use HBM because Nvidia buys the most HBM and has a cost advantage,” Tenstorrent founder Jim Keller and CEO told Bloomberg.
“But they’ll never be able to bring the price down the way HBM is built into their products and their sockets.”
Likewise, the startup’s second-generation chip reportedly offers triple the energy efficiency and a 33% lower price than an Nvidia DGX server.
While these stats are impressive, the real selling point for Tenstorrent appears to be its open-source focus, which gives engineers much more transparency and control over their underlying infrastructure.
But are these elements enough to rival Nvidia? Well, given analysts estimating that Nvidia holds between 70% to 90% of the market for AI chips, it’s unlikely, although Tenstorrent is in a strong position to attract attention from companies that want a better base cost over the H100.
After all, not every company can afford to fork out for GPUs that can cost as much as $20,000-$30,000 each.
In addition, many companies like Microsoft and AWS have begun developing their own chips to decrease their reliance on Nvidia, so there could be an opportunity for Tenstorrent to work with these organizations.
The Bottom Line
Tenstorrent has everything it needs to be a key alternative to Nvidia. This is good for the company and its investors and, ultimately, for the market too. The more viable alternatives there are to Nvidia, the less monopolized the market will be.
While Nvidia is undoubtedly here to stay for now, providers like Tenstorrent provide viable alternative infrastructure for performing AI inference tasks.
