Recommended Articles
Fundraising

doubleAI Raises $200M Series A for a New Category of AI That Optimizes Running Code

By Angeled · October 20, 2026 · Fundraising
doubleAI Raises $200M Series A for a New Category of AI That Optimizes Running Code
doubleAI Raises $200M Series A for a New Category of AI That Optimizes Running Code

The round in brief

doubleAI (officially AA-I Technologies Ltd.), a Tel Aviv frontier AI lab, announced a $200 million Series A on October 19, 2026, co-led by Lightspeed Venture Partners and Bessemer Venture Partners, with participation from NVIDIA, Dell Technologies Capital, Greenfield, Pitango, and BRM. The round values the company at more than $1 billion, bringing total capital raised to roughly $221M. The Wall Street Journal reported that doubleAI will use half the round — $100M — to purchase 1,000 NVIDIA B300 Blackwell GPUs as it competes with OpenAI and Anthropic on automated research tasks.

The founder and the new category

doubleAI is the latest venture of Amnon Shashua, who stepped down as CEO of Mobileye (the autonomous-driving company he co-founded, acquired by Intel and later re-IPO'd) and left his roles at AI21. The company has operated largely in stealth and calls its technology "Code Execution Models" (CEMs) — a new category of AI model that operates "between the code and the silicon."

The distinction matters. Coding agents and LLMs help developers write software. CEMs do something different: they work inside running software, continuously analyzing a live program, the hardware it runs on, and the data it processes — then optimizing the underlying machine instructions to fit those conditions in real time.

How Maxwell works

The company's first model, Maxwell:

  • Watches code as it runs, rewrites the compiled binary underneath (not the source code), and fits it to exactly what the machine is doing;
  • Formally proves every rewrite correct with mathematical certainty before applying it;
  • Runs on both CPUs and GPUs, optimizing complex workloads;
  • Is self-improving: each verified, faster version becomes the starting point for the next optimization, so programs keep getting faster the longer they run.

Early adopters report up to 30% speed gains without manual refactoring, according to UXC News. The business model is novel, too: doubleAI plans to charge customers based on the computing-cost savings it delivers — a "you keep part of what you save" pricing model.

The receipts: cuGraph, pandas, chess, and math

doubleAI has been building credibility through public demonstrations:

  • In March, Maxwell rewrote NVIDIA's own cuGraph library, speeding up 55% of its algorithms by 2x or more;
  • It produced an optimized version of pandas (the ubiquitous Python data-analysis library), improving performance on the PDS-H query-speed benchmark by 38x;
  • It doubled the video-generation speed of the LTX-2.5 world model;
  • Earlier in October, Maxwell broke a nearly 50-year chess record by finding a board position with 63 moves giving check (the prior record was 60), while a doubleAI agent made a new graph-theory discovery with a formal Lean verification file.

Why it matters for founders

doubleAI's rise shows where venture capital is moving beyond the coding-assistant land grab: toward infrastructure-level AI that reduces compute cost — the single biggest expense line of the AI era. If Maxwell can reliably deliver 20–30% efficiency gains in production, its savings-based pricing could spread adoption faster than license-based models, and every enterprise with a cloud bill becomes a prospect. Founders building AI infrastructure should note the go-to-market angle: sell the outcome (lower compute cost), not the technology.

Why it matters for investors

A $1B+ Series A valuation with only ~40 employees is the 2026 market in a sentence — but here the pedigree and the proof points are unusual. The participation of NVIDIA and Dell is strategic, not just financial: NVIDIA's own libraries run faster under Maxwell, which aligns the incumbent with the disruptor. The diligence questions to carry into this category: do formally-proven rewrites hold up across exotic production environments, and does the company capture enough of the savings to justify the valuation before copycats emerge?

Angeled.com connects founders with investors.

Sources: UXC News (October 19, 2026), SiliconANGLE (October 19, 2026), CTech/Calcalist (October 19, 2026), company announcement via Business Wire (October 19, 2026).

Read Next