Positron AI Raises $875M for Inference Chips
Reno-based Positron AI has secured significant capital to expand production of its inference hardware, aiming to reduce industry reliance on scarce memory components.
Positron AI, a company based in Reno, Nevada, has completed an $875 million Series C funding round. The investment values the firm at $5 billion and provides the resources needed to scale up manufacturing and develop next-generation silicon. This move positions the company to compete directly with established players in the artificial intelligence hardware market.
The round was co-led by New Enterprise Associates (NEA) and Jim Clark, a veteran entrepreneur known for co-founding Netscape. Other participants included Atreides Management, Valor Equity Partners, Andra Capital, and SemiAnalysis Capital. According to GN technics/hardware (en-US), this influx of capital is critical for Positron as it seeks to address specific bottlenecks in current AI infrastructure.
Bypassing Memory Supply Constraints
Most AI accelerators rely on high-bandwidth memory, a component that is currently in short supply. Positron takes a different approach by using standard mobile memory known as LPDDR5X. This choice allows the company to sidestep the tight supply chains that are affecting rivals, including major chipmakers who have had to adjust their roadmaps due to scarcity.
By utilizing this alternative memory technology, Positron claims its systems can access over 90% of available bandwidth. This efficiency reduces the company's exposure to supply chain disruptions. However, the trade-off is that this approach requires careful architectural design to ensure that the lower-cost memory does not become a performance bottleneck for large language models.
New Products and Board Expansion
The funding will support the development of two new products: Asimov, a custom inference chip, and Titan, a multi-chip server system. Positron’s CEO, Mitesh Agrawal, stated that lessons learned from deploying their current Atlas system have directly informed the design of these new offerings. The goal is to deliver hardware that balances compute power, memory bandwidth, and energy efficiency.
Along with the capital, several new members will join the Positron board of directors. These include representatives from NEA, Atreides, Clark’s office, and SemiAnalysis. This expanded board brings significant experience in both venture capital and technical hardware development, aiming to guide the company through its next phase of growth.
Strategic Focus on Inference Costs
The core of Positron’s strategy is to lower the cost and power consumption of running AI models. Jim Clark, who co-led the investment, highlighted that the company’s architecture is optimized for the specific needs of large language models. He noted that the balance between compute and memory storage is near ideal for the next evolution of these systems.
NEA partner Forest Baskett emphasized that Positron is addressing a critical industry constraint. While competitors struggle to secure scarce memory components, Positron has built its systems to operate independently of that specific supply chain. This independence offers a potential advantage in a market where hardware availability often dictates deployment speed.
