doubleAI's Maxwell AI Model Slashes Compute Costs by 67% Through Runtime Code Optimization

doubleAI Introduces Maxwell to Combat Rising Compute Costs

Tel Aviv-based startup doubleAI has launched Maxwell, pioneering what it calls code execution models (CEMs) that autonomously improve code performance to reduce computing costs. Co-founded by Amnon Shashua, who also leads Mobileye, the company announced Maxwell on Monday and opened it for early access1

1

2

2

. The timing addresses a critical pain point: hourly prices for Nvidia H100 chip capacity on one-year contracts have surged 60% over the past year, according to SemiAnalysis2

2

.

Maxwell operates fundamentally differently from large language models that assist developers in writing software. Instead of generating source code, Maxwell sits within running programs and rewrites the machine code underneath to optimize hardware utilization on both GPUs and CPUs1

1

. The original source code remains unchanged while Maxwell adapts the execution layer to specific hardware and data patterns it observes during runtime optimization1

1

.

How Code Execution Models Autonomously Improve Code Performance

The breakthrough behind Maxwell lies in its ability to formally verify each code modification mathematically, ensuring program behavior remains identical despite the changes1

1

2

2

. Shashua told The Wall Street Journal that verifying code was the main obstacle, and doubleAI's formal verification method solved this challenge2

2

. When Maxwell optimizes code based on certain conditions, it implements guards that return control to the original code if those conditions no longer hold1

1

.

This AI-driven software optimization operates without human review, placing the AI model between code and silicon to better utilization of GPUs and CPUs3

3

. "AI has changed how software is written, but not how it runs," Shashua stated1

1

. Maxwell addresses this gap by rewrites code as it runs, delivering measurable performance improvements across different workload types.

Dramatic Performance Gains and Cost Reductions

Early results demonstrate Maxwell's impact on compute costs. One customer achieved a 67% reduction in computing expenses1

1

2

2

. LTX, which builds open-weight multimodal models, found that Maxwell doubled its model inference speed. "Halving the GPU time immediately translates to your serving costs, your margins," said Zeev Farbman, LTX's co-founder and CEO1

1

2

2

.

On real-world workloads, Maxwell made data pipelines more than 38 times faster and accelerated AI inference kernels by over 4.5 times1

1

. These performance improvements directly address the challenge Shashua identified: "More hardware is being built, but the need for compute is rising faster than the ability to build GPUs and CPUs. If you can utilize the hardware better, then you can achieve something incredible in terms of savings"2

2

3

3

.

Revenue Model Tied to Customer Savings

doubleAI plans to charge customers a share of what they save on compute, though Shashua did not disclose the exact percentage1

1

2

2

. He explained that if Maxwell reduced a customer's compute spending from $10 million to $6 million, doubleAI would collect a portion of the $4 million saved2

2

. This model aligns the startup's revenue directly with customer value.

Shashua expects AI labs with massive training and inference bills, such as OpenAI and Anthropic, to benefit most from Maxwell. For these organizations, even a 10% savings could translate to hundreds of millions of dollars1

1

2

2

. This value proposition becomes increasingly compelling as AI spending continues to escalate across the industry.

Strategic Investment and Infrastructure

The two-year-old company has raised $220 million in total, with a $200 million Series A co-led by Lightspeed and Bessemer Venture Partners1

1

2

2

. Nvidia, Dell Technologies Capital, Greenfield, Pitango, and BRM also invested1

1

2

2

. The startup has approximately 40 employees1

1

2

2

.

DoubleAI allocated roughly half its funding to purchase 1,000 Nvidia B300 chips for its own computing needs1

1

2

2

. "You cannot rely just on renting compute. Prices are going up," Shashua told The Wall Street Journal1

1

2

2

. This strategic decision reflects the company's understanding of the compute scarcity challenge it aims to solve for customers.

Implications for Enterprise AI Adoption

Maxwell's launch comes as finance teams increasingly seek solutions to manage AI spending. Research from PYMNTS Intelligence shows that 43% of CFOs expect agents that continuously scan spending patterns and reallocate budgets to have major impact3

3

. Dynamic budget reallocation using real-time cost data ranks as the highest-priority use case for agentic AI among surveyed CFOs3

3

.

The broader release of Maxwell is slated before month's end2

2

. Watch for adoption patterns among major AI labs and how Maxwell's formally verifies approach to machine code optimization influences the development of similar tools. The success of code execution models could reshape how organizations approach hardware utilization and infrastructure planning as compute demand continues outpacing supply.