SHORTList
AI/ML
Data/Analytics
DevTools/Cloud

doubleAI

AI systems for expert-level software optimization

CORE INFO

$220M
Total Funding
Series A
Round
2023
Founded
~40 employees
Team Size
Ramat Gan, Israel
Headquarters

doubleAI builds AI tools for organizations running compute-intensive software, including AI infrastructure, data engineering, scientific computing, and semiconductor design. Its Maxwell system optimizes existing programs on CPUs and GPUs without requiring source-code changes, while WarpSpeed searches for verified software and hardware improvements.

  • Series A: $200 million, co-led by Lightspeed Venture Partners and Bessemer Venture Partners, with the company reportedly valued at more than $1 billion in October 2026.calcalistech.com
  • Compute-cost reduction: Maxwell reportedly cut costs by 67% for one unnamed early customer; the workload, baseline, and measurement period were not disclosed.thenextweb.com
  • The Wall Street Journal covered Maxwell's launch, the reported customer cost reduction, and doubleAI's purchase of 1,000 NVIDIA B300 chips.wsj.com

WHY WE WOULD WORK AT DOUBLEAI

Scale Expert Intelligence

Help build AI that captures and scales deep expertise in science and engineering. DoubleAI's mission is to remove the bottleneck of scarce human specialists.

Transform How Code Runs

Work on Maxwell, a Code Execution Model that optimizes existing software on CPUs and GPUs without requiring source-code changes. Its systems analyze programs, hardware, and data while formally verifying behavior.

Shape Frontier Research

Explore ambitious problems across GPU kernels, mathematics, cryptography, quantum circuits, and chip design. WarpSpeed has already reported a 3.6× geometric-mean speedup across tested cuGraph algorithms.

Make Compute Go Further

Your work can translate directly into lower infrastructure costs and faster workloads. Maxwell has reportedly reduced one early customer's compute costs by 67%, while doubling inference speed for an LTX workload.

Learn With Experts

Join a roughly 40-person, highly technical team that includes AI researchers, engineers, and PhDs across computer science, mathematics, and physics. The culture emphasizes ownership, collaboration, and continuous learning.

Grow With Real Momentum

DoubleAI has raised approximately $220M, including a reported $200M Series A backed by Lightspeed, Bessemer, NVIDIA, Dell Technologies Capital, and others. You'll help shape a young company with substantial resources and an ambitious product roadmap.

MARKET AND TRACTION

GROWTH TACTICS

  • Expanding through early-access enterprise and developer programs for Maxwell, with potential delivery through CLI, IDE integrations, web interfaces, and AI pipelines.

  • Using a performance-linked business model that may price access partly as a share of measurable computing-cost savings.

  • Targeting organizations with compute-intensive workloads across AI infrastructure, data engineering, scientific computing, cloud platforms, and semiconductor design.

    • MARKET OPPORTUNITY

  • doubleAI addresses the broad science-and-engineering market, focusing on organizations that need faster software and hardware execution without rewriting existing source code.

  • Relevant application areas include AI inference, GPU computing, data pipelines, cloud infrastructure, scientific workloads, cryptography, quantum circuits, and chip design. Specific total-addressable-market figures were not found.

    • PUBLIC USERS & TARGETS

    ✦ KEY METRIC
  • No named paying enterprise customers have been publicly disclosed.

  • LTX has been reported as testing or using Maxwell, with coverage describing doubled model-inference speed for an LTX workload; the commercial relationship was not clearly disclosed.

  • Public optimization targets and demonstrations include NVIDIA cuGraph, pandas, NVIDIA GPUs, and the NVIDIA CVDP hardware-design benchmark. These should not automatically be treated as customers.

    • KEY METRICS

    ✦ KEY METRIC
  • doubleAI has raised approximately $220M in total funding, including a reported $200M Series A, and has approximately 40 employees.

  • The company reports a 67% reduction in computing costs for one unnamed early customer, more than 38× faster data pipelines, and more than 4.5× faster AI inference kernels.

  • In company-reported WarpSpeed testing, optimized cuGraph workloads achieved a 3.6× geometric-mean speedup; 100% of tested algorithms improved, 55% exceeded 2× improvement, and 18% exceeded 10× improvement. These results have not been independently verified.

    • COMPETITIVE ADVANTAGE

  • Maxwell operates at the runtime and machine-instruction level, optimizing existing CPU and GPU execution without requiring source-code changes.

  • doubleAI combines long-horizon search with formal verification and behavioral-equivalence checks, aiming to produce optimizations that are both high-performing and correct.

  • Its Artificial Expert Intelligence and Code Execution Model approach is designed to scale specialized expertise across software optimization, hardware design, mathematics, and scientific engineering.

    • PRODUCT AND TECH

    Maxwell Code Execution

    Maxwell is a Code Execution Model that analyzes programs, hardware, and data to optimize execution on CPUs and GPUs without requiring source-code changes. It formally verifies that optimized programs preserve their original behavior.

    WarpSpeed Optimization

    WarpSpeed searches for, tests, and verifies performance optimizations for GPU kernels, software libraries, and hardware designs. It has been applied to technologies including NVIDIA cuGraph and pandas.

    Artificial Expert Intelligence

    doubleAI develops Artificial Expert Intelligence systems designed to perform deep, specialized work in science and engineering. Its approach emphasizes long-horizon search, domain expertise, verification, and continuous self-improvement.

    Formal Verification Systems

    The company uses formal verification, Lean proof formalization, SAT solving, proof certificates, and equivalence testing to validate software optimizations and technical solutions. These methods help ensure that performance improvements are correct rather than merely plausible.

    COMPANY CULTURE

    Values

  • Deep expertise over broad, shallow capability

  • Scientific rigor and evidence-based progress

  • Correctness through formal verification

  • Continuous self-improvement

    • Operating Principles

  • Search broadly across the space of ideas

  • Explore, backtrack, and build on prior results

  • Validate solutions rather than accepting plausible outputs

  • Turn research breakthroughs into practical products

    • Work Style

  • Highly technical and research-driven

  • Product-oriented with a focus on measurable outcomes

  • High ownership and individual agency

  • Collaborative across research and engineering

  • Comfortable working across code, systems, algorithms, and domains

    • Team Cadence

  • Fast-moving environment at a frontier-AI company

  • Broad, generalist problem-solving across science and engineering

  • Close collaboration between AI researchers and engineers

  • Hybrid work model for Ramat Gan positions

    • Learning & Growth

  • Continuous learning through difficult expert-level problems

  • Opportunities to work across AI, optimization, mathematics, hardware, and scientific computing

  • Exposure to advanced systems including CPUs, GPUs, distributed infrastructure, and formal verification