A startup selling alternatives to Nvidia's famous graphics processing units is unlikely to really challenge Nvidia for a long time. But they rebelliously claim good performance, speed, energy efficiency, cost, or all of the above.
Gusto is a must for teams who want to compete with a 10-year head start, virtually unlimited resources, and over 70% market share.
The answer to those who are purchasing Nvidia alternatives focuses on the financial value of AI tools and the total amount of AI usage.
“Competers have windows to carve the niche,” Carl Mozlkiwich, senior principal at AI architects at Cloud Farm Valdi, told Business Insider.
For efficiency
Nvidia chips are considered at the cutting edge of accelerated computing. But they are expensive, require immeasurable strength on scale, and in some cases companies are looking for more targeted, energy or spatially efficient solutions.
If companies want to diversify away from nvidia, they often rely on AMD and also create GPUs. AMD CTO Mark Papermaster told BI that moving workloads from one GPU to another will always be the easiest move from a technical standpoint. Also, AMD is to direct resources towards software and make this as simple as possible.
However, some industries are particularly keen to look out of these two candidates.
If the volume of inference is high enough and uniform enough, it makes economic sense to invest in doing work on alternative chip architectures for a particular workload. Robert Wachen, co-founder of Etched, one of the new participants in Chips Space, told BI.
“When you're guessing, finding the right processor that matches your use case is really important,” David Drigers, CEO of Startups Cerebras, Sambanova Systems and Cirrascale Cloud Services, which offers chips for Qualcomm, Nvidia and AMD, told Bi.
Here are the industries that look to Nvidia's alternatives:
High frequency trading
Rodrigo Liang, CEO of Sambanova Systems, told BI that high frequency stock trading (HFT) is an area where calculation speed and accuracy are “mission-critical.” Companies like Citadel Securities, Susquehanna and Jane Street can make millions by moving the market a second ahead.
“The entry points for new chips will come first in these high value use cases,” Liang said.
HFT companies hire top machine learning talents and often compete directly with frontier companies like Openai and humanity.
Jane Street has thousands of H100 and H200 chips from Nvidia, according to the company's website. However, the company also participated in Etched's $120 million Series A in 2024. Etched bets on the type of transformer model used in chatbots. The company also raised funds from PayPal founder Peter Thiel and Github CEO Thomas Dohmke.
In addition to the need for speed, HFT companies often need at least some calculations to make them completely private.
“A hedge fund with its own code that has been a truly trade secret strategy for a long time was the last holdout of the cloud,” SourceGraph CEO Quinn Slack told BI. On-premises data centers are more likely to have the space and energy-related constraints that NVIDIA alternatives are trying to seize.
Targeting and recommendations
Some of the oldest returns of billions of dollars invested in AI came from within Tech Giants' social media and e-commerce business.
“AI has already made it better for many companies to target and find audiences who are more interested in their products than they are themselves. By using AI to improve recommendations across the platform, Instagram, threads and Facebook users have more time on their apps.
“There are also many opportunities to improve our core business by doing more calculations on advertising and recommendations,” says Zuckerberg.
Wachen said BI recommendation workloads are being prepared for Nvidia's alternatives, especially as ad targeting will be custom generated ads.
Ad targeting and content recommendations are also the main use cases for chips developed by cloud companies. For example, Google's TPU, a chip originally designed in partnership with Broadcom, is particularly suitable for these tasks.
Sovereign AI
Sovereign Cloud – State development data centers built for national security and other purposes often share some restrictions with the financial services sector.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
Saudi Arabian entities have been particularly committed to diversifying chips. For example, Saudi Arabia Aramco deals with Celebras, GROQ, Sambanova Systems (and AMD and NVIDIA).
G42, Saudi Arabia's leading AI data center project, is partnering with Cerebras, AMD and Nvidia. Humain, the latest organization to emerge from Saudi Arabia's greedy AI, first announced President Donald Trump's recent visit to the Kingdom, Groq, Nvidia and AMD as partners.
Canadian telecom company Bell Canada has announced GROQ as its inference provider. The Sambanova chip is located in one of SoftBank's data centers in Japan.
On the stack
When Meta introduced the Llama 4 model, the first type of API was also released. This is a direct way for developers to access models from Meta itself, and the computing power comes from Cerebras and Groq. Both companies have their own new chip architecture. They also offer their own inference services along with selling tips.
The reasoning market is busy. However, because they set up their own data centers and provide them as a service, chip startups can generate revenue faster than the process that is much longer than their own sales process. Still, there are trade-offs.
The choice of hardware poses both opportunities and risks for the enterprise, Driger said. Using an alternative to Nvidia can save you time and money in some cases, but most alternatives are less flexible than GPUs. The risk of AI's commitment in these early days is that many companies are trying not to buy non-nvidia chips, he said. In the meantime, startups are fighting against a very cumbersome reasoning market.
“If our offerings improve, our customers will stay,” says Liang. “If someone else has a better product, the market moves.”
