Positron AI Inc. secured $875 million in Series C funding to develop inference systems that use widely available smartphone-grade memory, promising faster LLM processing despite lower memory bandwidth than traditional server-grade solutions.
- Positron raised $875 million to advance inference hardware with consumer-grade LPDDR5X memory.
- Their Titan system supports large LLMs with up to 18.4 TB LPDDR5X and custom Asimov chips.
- Production is targeted for late 2027 with emphasis on LPDDR5X supply chain partnerships.
Market signal
Global demand for large language model inference capacity is intensifying, straining supply chains for traditional high-bandwidth memory critical to AI accelerators. Positron’s $875 million capital raise reflects strong investor confidence in technologies that can alleviate this bottleneck by using more accessible memory types.
By substituting costlier and scarce HBM with LPDDR5X, primarily used in smartphones, Positron aims to capture a significant portion of the inference hardware market. The startup's approach signals a shift in hardware design priorities toward maximizing real-world memory bandwidth efficiency rather than theoretical peak performance.
Operator impact
Operators running large scale AI systems may benefit from Positron’s inference appliances through potentially lower hardware costs and improved supply availability. The Titan appliance’s large LPDDR5X memory pools combined with the Asimov chip’s efficient utilization can enable inference on extremely large models with fewer systems and reduced energy demands.
Positron’s architecture, including its systolic-array-based Asimov accelerator with co-located memory and programmable CPU cores, provides flexibility for diverse LLM workloads. Enterprises should evaluate the opportunity to diversify their AI infrastructure providers as Positron moves toward production, potentially gaining access to more scalable and cost-effective inference solutions.
What to watch next
Key milestones will include the successful 2026 tape-out of the Asimov chip on TSMC’s 3-nanometer node and subsequent volume production of Titan appliances in the second half of 2027. Close monitoring of LPDDR5X supply commitments will be critical to ensure Positron can meet anticipated demand without interruption.
Industry adoption and performance benchmarks compared to incumbent GPU-based solutions like Nvidia’s Blackwell series will also be vital. The pace at which Positron can scale manufacturing and prove its technology claims in real-world deployments will determine its market impact in the evolving AI inference landscape.