Mitesh Agrawal, the former Chief Operating Officer of Lambda Labs—a prominent Nvidia partner valued at over $2 billion—has made a significant career move to become CEO of Positron, an emerging AI hardware startup focused on transformer model inference. This transition marks a notable shift in the AI infrastructure landscape as Positron positions itself as a direct competitor to Nvidia’s dominance in the GPU market.
Lambda Labs’ Journey and Agrawal’s Impact: Agrawal joined Lambda Labs in 2017 when the company was primarily focused on image generation models. Under his leadership, Lambda transformed from a facial-recognition technology company founded by twin brothers Stephen and Michael Balaban into a cloud infrastructure powerhouse. The company pivoted to designing full-scale data center infrastructure and cloud services, now focusing on deploying servers powered by Nvidia GPUs along with supporting software, APIs for inference, and machine-learning libraries. During Lambda’s Series C funding round in February 2024, the company achieved a valuation of approximately $1.5 billion, which Agrawal confirmed has since grown to more than $2 billion.
Positron’s Mission and Technology: Founded in 2023 by Thomas Sohmers (who will now serve as CTO), Positron is building specialized hardware for transformer model inference—the process that enables AI chatbots like ChatGPT and xAI’s Grok to respond to user requests. Agrawal explained his move was driven by the exponential growth in inference demand, noting that “the curve of technology for inference is just going up, which means the computational requirement is really going up.” Positron claims its hardware outperforms Nvidia’s H100 and H200 GPUs in performance, power efficiency, and affordability.
Challenging Nvidia’s Dominance: Taking on Nvidia—which recently overtook Apple as the world’s most valuable company—represents a formidable challenge. However, Positron’s strategy focuses on a specific niche: transformer model inference. Transformer models, the neural networks behind popular generative AI applications, have greater memory demands than previous convolutional neural networks. By specializing in this area, Positron aims to differentiate itself from Nvidia’s broader product portfolio. Crucially, Positron’s hardware is designed to be compatible with models trained on Nvidia GPUs, effectively bypassing Nvidia’s famous “CUDA moat”—the software ecosystem that typically locks customers into Nvidia products.
Key Quotes
The curve of technology for inference is just going up, which means the computational requirement is really going up.
Mitesh Agrawal, new CEO of Positron, explained his rationale for joining the startup. This statement highlights the explosive growth in AI inference demand as chatbots and reasoning models become more sophisticated and widely deployed.
I would say the whole reason we started Positron is we thought that there was a better way to do things. Nvidia, as a large company that also has a lot of other product focuses, wasn’t going to really optimize and focus on the particular niche that we’re focused on, which is transformer model inference.
Thomas Sohmers, Positron’s founder and incoming CTO, articulated the company’s strategic positioning. This quote reveals Positron’s belief that Nvidia’s broad focus creates an opportunity for specialized competitors to outperform in specific use cases.
What Positron really did was completely remove this friction of anything. A company could take a model trained on an Nvidia GPU and run that model’s inference on a Positron card just like you would run on an Nvidia GPU.
Agrawal emphasized Positron’s key competitive advantage: compatibility with Nvidia-trained models. This addresses one of Nvidia’s strongest competitive moats—its CUDA software ecosystem—by making it easy for customers to switch hardware without retraining models.
You get to compete against an industry veteran as well as in a field that is just so big.
Agrawal described his decision to leave an established, multi-billion dollar company for a startup. This reflects both the enormous market opportunity in AI inference and the confidence that specialized solutions can compete against dominant players.
Our Take
Agrawal’s move from Lambda Labs to Positron represents more than just an executive transition—it’s a bellwether for the AI hardware industry’s evolution. The timing is particularly significant as the industry shifts from the “training era” dominated by building ever-larger models to the “inference era” focused on efficiently deploying AI at scale. Positron’s strategy of creating Nvidia-compatible hardware specifically optimized for transformer inference is shrewd, potentially offering customers a lower-friction path to diversify their hardware suppliers. However, the challenge remains formidable: Nvidia’s ecosystem advantages, manufacturing scale, and continuous innovation (like the new Blackwell chips) create high barriers to entry. The real test will be whether Positron can deliver on its performance claims while scaling production and building the customer relationships necessary to compete. If successful, this could catalyze a wave of specialized AI hardware companies, fundamentally reshaping the competitive landscape and potentially reducing the infrastructure costs that currently constrain AI deployment.
Why This Matters
This executive transition and Positron’s emergence signal a critical evolution in the AI hardware market. As AI applications shift from training massive models to deploying them at scale through inference, specialized hardware solutions are becoming increasingly valuable. The inference market represents the next major battleground in AI infrastructure, as companies seek more efficient and cost-effective ways to run chatbots, reasoning models, and other AI applications that serve millions of users daily.
Positron’s challenge to Nvidia’s near-monopoly could have far-reaching implications for the AI industry. If successful, it would provide AI companies with alternatives to Nvidia’s expensive GPUs, potentially reducing costs and increasing competition. The fact that a senior executive from a $2 billion Nvidia partner is willing to bet his career on this challenge suggests genuine confidence in the market opportunity. For businesses deploying AI at scale, more competition in the hardware space could mean lower costs and better performance options. This move also highlights how the AI industry is maturing beyond the initial training phase into operational deployment, where efficiency and specialized solutions become paramount.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- Jensen Huang: TSMC Helped Fix Design Flaw with Nvidia’s Blackwell AI Chip
- Encharge AI Secures $21.7M Series A Funding to Revolutionize AI Chip Efficiency
- Pitch Deck: TensorWave raises $10M to build safer AI compute chips for Nvidia and AMD
- OpenAI’s Valuation Soars as AI Race Heats Up
- Tech Workers Are the Real Winners in the AI Talent War, With Pay Set to Soar by 2024
Source: https://www.businessinsider.com/lambda-labs-coo-mitesh-agrawal-positron-ceo-nvidia-ai-chips-2025-1