What Gimlet Labs does

Gimlet Labs is a San Francisco-based applied AI research and infrastructure company founded in 2023 that builds software for executing agent-based AI workloads across heterogeneous hardware. Its core product, Gimlet Cloud, is a serverless inference platform that orchestrates AI workloads across CPUs, NVIDIA GPUs, Intel Gaudi, AMD accelerators, ARM, Cerebras, and d-Matrix SRAM-centric accelerators, automatically mapping each phase of inference to the optimal processor to deliver 3-10x performance gains at the same cost and power. The company also ships kforge, an autonomous kernel-generation toolkit that uses a multi-agent system with shared memory to produce optimized low-level kernels directly from PyTorch across CUDA, ROCm, and Metal backends, with disclosed speedups of ~1.8x on H100 and ~1.24x on Apple M4 over PyTorch baselines.

The platform is positioned for agentic AI workloads that produce 5-15x more tokens than chat models, addressing an inference bottleneck that the company says leaves current data centers operating at only 15-30% utilization. Revenue is generated through consumption-based pricing on Gimlet Cloud inference compute, augmented by a standalone kforge offering and supplemented by enterprise and self-serve channels. Go-to-market operates on a dual track: high-touch enterprise field sales targeting frontier AI labs and hyperscalers, alongside product-led growth through the hosted platform for developers and smaller teams.

As of March 2026, Gimlet has raised $92 million in total funding ($80M Series A led by Menlo Ventures at a $3.1B post-money valuation plus $12M seed), disclosed eight-figure revenue exceeding $10 million within five months of its October 2025 stealth launch, and tripled its customer base. The company operates with 11-50 employees from 255 Potrero Ave in San Francisco, and has established technology or strategic relationships with NVIDIA, Intel, AMD, ARM, Cerebras, d-Matrix, MLCommons, and a deep advisor bench spanning Stanford faculty and operators from Sequoia, Figma, VMware, and SiMA.ai.

Name

Gimlet Labs

Legal name

Gimlet Labs, Inc.

Website

https://gimletlabs.ai

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

11–50 employees

Short description

Gimlet Labs builds a multi-silicon serverless inference cloud for AI agents, orchestrating workloads across CPUs, GPUs, and SRAM-centric accelerators to deliver 3-10x performance gains, primarily serving frontier AI labs and hyperscalers.

Ownership category

akta.pro rank

Product category

AI Inference Infrastructure

NAICS

Computer Systems Design Services (541512)

SIC

Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

Model Deployment, Serving & Inference Platforms (HDAAABAF)

akta.pro secondary industries

AI Compute Virtualization & Scheduling (GPU virtualization, cluster schedulers) (HDAAAAAG), On-Device Inference Runtimes & SDKs (mobile/embedded) (HDAAAJAB)

Keywords

• Multi-silicon inference cloud
• AI inference infrastructure
• Heterogeneous compute orchestration
• Autonomous kernel generation
• Agentic AI workloads

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Infrastructure, Personnel, Operations, Marketing or Sales

Revenue model

1. Serverless AI Inference Platform (Gimlet Cloud): Gimlet Cloud provides serverless inference for AI agents, handling scheduling, orchestration, and optimization across heterogeneous hardware. Users can run simple agents to complex multi-agent systems with custom logic and data sources. Revenue is generated through consumption-based pricing for inference compute used.
2. Autonomous Kernel Generation Tool (kforge): kforge autonomously generates optimized low-level kernels directly from PyTorch code. May be offered as a standalone commercial product or integrated offering.

Go-to-market motion2 records

Distribution channels3 records

Marketing channels5 records

Core offering

Gimlet Labs builds and operates Gimlet Cloud, a serverless multi-silicon AI inference platform that runs agentic workloads (simple agents through multi-agent systems with custom logic and data sources) by automatically disaggregating and orchestrating each phase (prefill, decode, speculative decoding, tool calls) across heterogeneous hardware including NVIDIA GPUs, Intel Gaudi, AMD, ARM, Cerebras, and d-Matrix Corsair accelerators. The company also offers kforge, an autonomous kernel generation toolkit that produces optimized low-level kernels directly from PyTorch code across CUDA, ROCm, and Metal backends.

Product overview

Gimlet Labs is an applied AI research and product company that offers two core products: Gimlet Cloud (a serverless inference platform for AI agents) and kforge (an autonomous kernel generation toolkit). Gimlet Cloud provides multi-silicon inference capabilities, distributing AI workloads across CPUs, GPUs, and specialized accelerators from vendors including NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix. kforge complements the platform by autonomously generating optimized low-level kernels for PyTorch across diverse hardware. Together, these products aim to make AI workloads 10X more efficient by expanding the pool of usable compute and improving how it is orchestrated.

Differentiator

Problem solved

Functional benefit

Brands

• Gimlet Cloud: Serverless inference cloud platform for AI agents. Enables running everything from simple agents to complex multi-agent systems with custom logic and data sources. The platform handles scheduling, orchestration, and optimization across heterogeneous hardware including CPUs, GPUs, and specialized AI accelerators.
• kforge

Products and services

• Gimlet Cloud Serverless inference cloud platform for AI agents that enables running simple agents to complex multi-agent systems with custom logic and data sources. The platform handles scheduling, orchestration, and optimization across heterogeneous hardware including CPUs, GPUs, and specialized AI accelerators from NVIDIA, AMD, Intel, ARM, Cerebras, and d-Matrix.
• kforge Autonomous kernel generation toolkit that generates optimized low-level kernels directly from PyTorch code using a multi-agent system with shared memory to explore designs, enforce correctness checks, and identify the fastest kernels across CUDA, ROCm, and Metal backends.

Quantifiable outcome

• 3-10x speedup on >1T parameter frontier models with large context windows at same cost and power
• +5 more outcomes

Named customers2 records

Segments3 records

Ideal customer profiles3 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability8 records

Feature9 records

Partnerships

Twelve partnerships are on record, tiered minor and core.

• MLCommonsminorStrategic or Co-development Partner · 29 June 2026Gimlet Labs joined MLCommons as a member company to contribute to new benchmarks for agentic inference and support open standards for ML performance across industry and research. This positions Gimlet to help shape emerging performance standards for heterogeneous inference systems as agentic inference becomes a dominant workload.
• d-MatrixcoreTechnology or Integration · 12 March 2026d-Matrix and Gimlet Labs partnered to integrate d-Matrix Corsair SRAM-centric accelerators into Gimlet Cloud for AI inference workloads, achieving up to 10x improvements in latency and throughput per watt compared to GPU-only deployments. The partnership routes memory-sensitive operations like speculative decoding and token generation to Corsair while GPUs handle compute-heavy inference stages, delivering the best of both architectures within Gimlet's multi-silicon orchestration.
• NVIDIAcoreTechnology or IntegrationGimlet has established partnerships with NVIDIA to ensure compatibility of its multi-silicon inference cloud with NVIDIA hardware. NVIDIA GPUs serve as primary compute accelerators within the heterogeneous stack, with Gimlet optimizing workloads including prefill and verification stages on NVIDIA B200 and H100 GPUs.
• IntelcoreTechnology or IntegrationIntel Gaudi 3 accelerators are integrated into Gimlet's heterogeneous inference stack. Research demonstrates that B200:Gaudi3 multi-vendor disaggregation (NVIDIA B200 for prefill, Intel Gaudi 3 for decode) achieves 1.7-4x TCO improvements versus single-vendor disaggregation.
• Mark HorowitzminorOthersMark Horowitz, Rambus Founder and Stanford Faculty, serves as an advisor to Gimlet Labs, contributing expertise in chip architecture and memory systems.
• Chris RéminorOthersChris Ré, Entrepreneur and Stanford Faculty, serves as an advisor, contributing deep expertise in machine learning systems and scalable AI infrastructure.
• Lukas BiewaldminorOthersLukas Biewald, Co-Founder of Weights & Biases, serves as an advisor, contributing expertise in ML tooling and experiment tracking infrastructure.
• Dylan FieldminorOthersDylan Field, Founder and CEO of Figma, serves as an advisor, contributing expertise in product development and scaling consumer/enterprise software.
• Amarjit GillminorOthersAmarjit Gill, Founder of P.A. Semi, serves as an advisor, contributing expertise in low-power processor design and chip company building.
• Krishna RangasayeeminorOthersKrishna Rangasayee, Founder and CEO of SiMA.ai, serves as an advisor, contributing expertise in AI chip design and edge AI deployment.
• Andrew TanminorOthersAndrew Tan, Founder of TLDR AI, serves as an advisor, contributing expertise in AI summarization products and developer-facing AI tools.
• Rangarajan RaghuramanminorOthersRangarajan Raghuraman, former CEO of VMware, serves as an advisor, contributing expertise in enterprise infrastructure, virtualization, and large-scale distributed systems.

Scale indicators6 records

Recent moves7 records

Expansion highlights6 records

Direct peers

• Anyscale: Anyscale provides the Ray-based Anyscale Platform for distributed AI compute, including production LLM serving and agentic workloads. It competes with Gimlet on serverless AI inference and multi-node orchestration for production deployments.
• Modal Labs: Modal provides serverless cloud infrastructure for running AI workloads, including model inference and agentic pipelines. It targets a similar developer-led and enterprise adoption curve as Gimlet Cloud.
• Together AI: Together AI operates a cloud platform for training, fine-tuning, and serving open and custom AI models, including agentic inference workloads. It directly competes with Gimlet Cloud in the inference-as-a-service layer targeting frontier labs and AI developers.
• Replicate: Replicate runs a serverless cloud for running open-source ML models via API, including LLMs and multimodal agents. It competes with Gimlet Cloud on consumption-based inference pricing for AI applications.
• OctoAI (NVIDIA): OctoAI built a serverless inference platform for generative AI models and was acquired by NVIDIA. It is a direct predecessor-style competitor to Gimlet Cloud, with NVIDIA now able to bundle similar capabilities into its DGX Cloud stack.
• Fireworks AI: Fireworks AI offers a serverless inference platform optimized for low-latency LLM and agent inference, with fine-tuning and custom deployment options. It competes head-to-head with Gimlet Cloud on inference performance and developer experience.

Broad incumbents

• Hugging Face: Hugging Face offers Inference Endpoints and a broader AI platform spanning model hosting, training, and deployment. It competes with Gimlet for developer attention on hosted inference, though it spans a much broader model and tooling portfolio.
• CoreWeave: CoreWeave is a large GPU-cloud provider offering AI training and inference infrastructure at scale. While not specialized for agentic inference, it competes for the same hyperscaler and frontier-lab workloads and provides an alternative path for heterogeneous compute.
• Lambda Labs: Lambda is a GPU cloud and on-prem AI infrastructure provider serving labs and enterprises with training and inference clusters. It overlaps with Gimlet on serving AI workloads on GPU hardware, though without Gimlet's multi-silicon orchestration layer.

Emerging players

• RunPod: RunPod provides GPU cloud and serverless endpoints for AI inference and training. It targets developers and smaller teams seeking lower-cost alternatives to hyperscaler GPU clouds, partially overlapping with Gimlet's self-serve/PLG motion.

Market position

Strengths5 records

Weaknesses5 records

Competitive moat4 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/gimletlabs
• LinkedInlinkedin.com/company/gimletlabs

Revenue estimate

Valuation estimate

Number of profiles

Profiles17 records

Funding overview

Funding rounds3 records

Investors18 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Gimlet Labs