Gimlet Labs Raises $300 Million Series B Funding for AI Inference Platform

The surge in AI inference demand is reshaping the hardware landscape, pushing companies to adopt more specialized solutions.

• Gimlet Labs secured a $300 million Series B funding round, achieving a $3 billion valuation.
• Investors included major players like Andreessen Horowitz, Arm Holdings, and Microsoft’s M12 fund.
• The funding will enhance Gimlet's serverless inference platform and custom hardware development to meet rising AI demands.

• AI inference has surpassed training as the primary workload, with token generation increasing sixfold in the past year.
• Power consumption in AI data centers is projected to triple by 2030, creating a need for optimized solutions.
• Gimlet Labs offers a unique disaggregated AI inference platform that improves performance by 3-10X through heterogeneous hardware utilization.

Gimlet Labs is at the forefront of a significant shift in AI infrastructure, driven by the increasing complexity and demand for AI inference workloads. The company emerged from stealth in October 2025, quickly gaining traction with its innovative multi-silicon inference platform. This platform allows for the disaggregation of large language models (LLMs) into modular components, which can be efficiently scheduled across various hardware accelerators, including GPUs, CPUs, and specialized architectures.

The recent $300 million Series B funding round, led by Andreessen Horowitz, underscores the growing investor confidence in Gimlet's approach. With total funding now at $392 million, the capital will be directed towards expanding its serverless inference platform and developing custom inference-optimized servers. This is crucial as the demand for efficient AI inference solutions surges, particularly in light of the power constraints faced by data centers.

The context of this funding is rooted in the broader AI landscape, where inference workloads have become the dominant force. Monthly token generation has skyrocketed, and projections indicate a further 20X increase by 2030. Traditional homogeneous infrastructure, often repurposed from training or crypto mining, is proving inadequate for the multi-stage agentic AI workloads that require specialized hardware. Gimlet's platform addresses these inefficiencies by allowing for granular disaggregation, including prefill/decode separation and speculative decoding.

As power consumption in AI data centers reached approximately 18 GW in 2025, the urgency for optimized solutions has never been greater. The anticipated tripling of this figure by 2030 creates significant bottlenecks, making Gimlet's disaggregation approach not just advantageous but necessary. By leveraging AI agents and a custom compiler, Gimlet's platform can achieve performance improvements of 3-10X in throughput and interactivity for frontier workloads.

This funding round also highlights a strategic alignment with semiconductor firms like Arm, indicating a trend towards collaboration between software and hardware ecosystems. As Gimlet Labs continues to scale its operations and develop custom hardware, it positions itself as a key player in the evolving AI landscape, catering to the increasing demand for efficient and effective AI inference solutions.

• Tech companies: They will benefit from more efficient AI inference solutions, reducing operational costs.
• Data center operators: They face pressure to adapt to rising power consumption and efficiency demands.
• Investors: Those backing Gimlet Labs will see potential returns as the company scales and captures market share.

• Market adoption of disaggregated AI solutions: Increased uptake could validate Gimlet's model and influence competitors.
• Power consumption trends in data centers: Monitoring these trends will reveal the urgency for optimized infrastructure.
• Investor interest in AI infrastructure: Continued funding rounds could indicate a growing market for specialized AI solutions.