Lambda Labs’ COO Has Left for Positron, a Startup Nvidia Competitor

• Lambda Labs COO Mitesh Agrawal has left to go AI {hardware} startup Positron.
• Lambda focuses on deploying cloud infrastructure to clients and is valued at over $2 billion.
• Positron goals to compete with Nvidia by providing quicker, energy-efficient AI {hardware}.

Lambda Labs, a Nvidia accomplice, has misplaced its chief working officer to a little-known firm constructing {hardware} for the AI trade.

Lambda COO Mitesh Agrawal instructed Enterprise Insider he stepped into a brand new function as CEO of Positron earlier this month. Positron builds {hardware} for transformer mannequin inference, which is how chatbots like ChatGPT reply to consumer requests.

Agrawal’s departure is critical given his function in shaping Lambda into one among Silicon Valley’s best-funded and most beneficial startups.

Throughout its Collection C spherical final February, the corporate was valued at about $1.5 billion. Agrawal declined to share the corporate’s actual valuation however stated it has grown to over $2 billion since then.

Agrawal instructed BI that when he joined Lambda in 2017, the corporate was centered on constructing machines for picture technology fashions. This was 5 years after twin brothers Stephen and Michael Balaban based it as an organization growing facial recognition expertise. It wasn’t lengthy after Agrawal’s arrival, nevertheless, that the corporate shifted its focus, designing infrastructure for full-scale information facilities and pivoting into cloud providers.

He stated Lambda’s enterprise now focuses on deploying cloud infrastructure to clients, renting out servers powered by Nvidia’s graphics processing models. It additionally presents the requisite software program, together with APIs for inference and machine studying libraries for purchasers.

Agrawal stated that his transfer to Positron comes amid a rising urge for food for inference — the capability for AI fashions to use their coaching to new information.

Between chatbots like ChatGPT and xAI’s Grok, and new reasoning fashions like OpenAI’s o1 tackling PhD-level issues, “the curve of expertise for inference is simply going up, which implies the computational requirement is actually going up,” Agrawal stated. So, he stated he is pondering quite a bit about “easy methods to resolve and easy methods to run these fashions with as a lot effectivity as attainable.”

He believes Positron is well-positioned to tackle that problem.

Positron was based in 2023 by Thomas Sohmers, whom Agrawal met in 2015. The 2 additionally overlapped at Lambda throughout Sohmers’s stint on the firm between 2020 and 2021. Sohmers, who will transfer into the function of chief expertise officer, instructed BI that, in easiest phrases, the corporate is “constructing {hardware} competing in opposition to Nvidia.”

Positron says its {hardware} outperforms Nvidia’s H100 and H200 GPUs — which fueled the AI race earlier than it launched its extra highly effective Blackwell chips — in efficiency, energy, and affordability.

Going up in opposition to a behemoth like Nvidia — which overtook Apple because the world’s most beneficial firm final week — isn’t any straightforward job for an up-and-coming firm. However by focusing extra narrowly on offering {hardware} for transformer mannequin inference, Sohmers stated Positron can differentiate itself from the competitors.

Transformer fashions — neural networks that study the context and that means of information to generate new information — are behind among the hottest generative AI functions. Not like convolutional neural networks, which underpinned earlier a long time of machine studying advances, transformer fashions have better reminiscence calls for. Sohmers stated he noticed a chance to capitalize on these calls for.

“I’d say the entire purpose we began Positron is we thought that there was a greater solution to do issues,” Sohmers stated. “Nvidia, as a big firm that additionally has loads of different product focuses wasn’t going to actually optimize and deal with the actual area of interest that we’re centered on, which is transformer mannequin inference.”

Agrawal, too, is assured within the efficiency and power effectivity of Positron’s {hardware}. Its compatibility with a spread of transformer fashions can even assist it entice clients from rivals, he stated.

“Nvidia has such a powerful ecosystem on this planet of AI fashions. You hear about their CUDA moat, and also you heard in regards to the software program moat,” he stated, referring to the software program community the corporate has constructed between its merchandise to retain clients.

“What Positron actually did was fully take away this friction of something,” Agrawal stated. Meaning an organization can take a mannequin skilled on an Nvidia GPU and “run that mannequin’s inference on a Positron card identical to you’ll run on an Nvidia GPU,” he stated.

Agrawal stated the bounce from a longtime participant like Lambda to a younger startup like Positron presents an “thrilling problem.”

“You get to compete in opposition to an trade veteran in addition to in a subject that’s simply so huge,” he stated.