SAN FRANCISCO, Oct. 22, 2025 (GLOBE NEWSWIRE) — Today Gimlet Labs, the Applied AI research and product company, launched with its platform that is the first to seamlessly decouple agentic AI workloads from underlying hardware, slice each workload into components and intelligently map each component to the hardware that is right-sized for its needs. Gimlet Labs has 8-figure revenues; its platform is deployed at AI-native and Fortune 500 companies, powering workloads that span multi-generation and multi-vendor hardware.
Industry luminaries on Gimlet Labs:
Sachin Katti, Chief Technology and AI Officer at Intel and Adjunct Professor at Stanford University, said: “The industry is hitting the limits of homogeneous, vertically integrated one-size-fits-all AI infrastructure. Agentic AI applications are inherently heterogeneous, reasoning across multiple models, modalities and data sources. These applications require heterogeneous hardware to scale but orchestrating workloads across diverse hardware remains a hurdle in practice. Gimlet provides the missing layer: abstracting heterogeneous hardware into a unified foundation that lowers cost, boosts performance and provides zero-friction deployments.”
Pete Warden, founding member of the TensorFlow team, said: “As one of the founders of TensorFlow, I know from personal experience how hard it is to move models from GPU platforms that are optimized for training, to run on production hardware. Developers want to work in the frameworks they know. Gimlet lets developers seamlessly deploy on the best hardware without having to rewrite their application. It’s one of the most game-changing technologies I’ve seen in AI, with the power to quietly unlock tens of billions of dollars in savings.”
The current model of deploying agentic AI workloads is broken and not scalable – it wastes limited compute power, drives up costs and sharply restricts available capacity. Typically, these workloads are deployed on the same type of hardware they were trained on because it is difficult to port models to different hardware and maintain performance. Gimlet Labs’ team saw that there needs to be a new paradigm for how AI workloads are run and optimized in order to scale.
“AI workloads have advanced rapidly but the infrastructure supporting them has not kept up. Gimlet’s mission is to make these workloads 10X more efficient and unlock new compute capacity for AI, giving customers choice and better performance per dollar. Our platform seamlessly decomposes and schedules AI workloads across many different types of hardware without requiring application changes,” said Zain Asgar, co-founder and CEO of Gimlet Labs and an Adjunct Professor of Computer Science at Stanford University. Previously he was a GPU architect at NVIDIA, an engineering lead at Google AI and the co-founder and CEO of Pixie Labs that was acquired by New Relic.
Gimlet’s unique approach unlocks new sources of compute for agentic AI workloads and provides superior performance which is critical given the shortage of GPUs and massive demand. Instead of being restricted to limited types of hardware, Gimlet can automatically target agentic AI workloads to many types of hardware, from GPUs to CPUs to custom accelerators of different sizes without code changes. The system can seamlessly run across both multiple generations and multiple vendors of GPUs, such as NVIDIA, Intel and AMD, providing significant operational flexibility.
- Autoporting without code changes: Unlike other solutions that require rewriting workloads into new frameworks, Gimlet automatically slices, maps and autoports the workload on behalf of the user using its novel technology in kernel generation and advanced compilation techniques. In addition, workloads written for a specific GPU can be ported by Gimlet to other hardware platforms.
- Intelligent orchestration across different hardware: Gimlet’s system automatically decomposes agentic AI workloads into different components and maps them to the most efficient device. Gimlet can also orchestrate agentic AI pipelines across both edge devices like laptops and cloud-based deployments, a hybrid approach that delivers significant application performance benefits.
- Agentic workload support: Gimlet goes beyond running individual models, it can support running entire agentic applications, spanning multiple models, custom code and custom data sources. Users can easily import their agents into Gimlet through frameworks such as LangChain and LangGraph or directly import models via APIs like Hugging Face transformers and PyTorch.
Gimlet Labs is offering its stack as an on-prem solution for datacenters, as well as a hosted developer platform.
About Gimlet Labs Research
Gimlet Labs drives foundational research across the stack in order to enable the next generation of efficient, scalable infrastructure. The research core of the company combines theory and practice to push the boundaries of AI efficiency via techniques such as automated GPU kernel generation, workload orchestration and heterogeneous execution across diverse hardware.
Gimlet Labs Raises Seed Funding Led by Factory
Today Gimlet Labs also announced that it raised a $12 million seed round led by Factory. Angels include Jon Feiber (Stanford Adjunct Professor), Dylan Field (co-founder and CEO of Figma), Amarjit Gill (Entrepreneur), Akshay Kothari (COO at Notion), Nick McKeown (Stanford Professor), Raghu Raghuram (former CEO of VMware) and Lip-Bu Tan (CEO of Intel).
About Gimlet Labs
Gimlet Labs’ mission is to drive breakthrough improvements in AI efficiency that result in massive increases of compute available for AI workloads. For more information, simply visit: https://gimletlabs.ai/.
Media and Analyst Contact:
Amber Rowland
amber@therowlandagency.com
+1-650-814-4560
