Israeli AI Firm Two Delta Raises Tens of Millions From Top Investors
Translated & summarized from Calcalist by baba
Israeli AI startup Two Delta has raised tens of millions of dollars from investors including Bill Ackman and Eric Schmidt. The company's technology, DCOCO, reportedly doubles the throughput of AI models by improving efficiency. Two Delta's research also indicates current AI agents struggle to maintain application integrity over time as complexity increases.
The story in 5 lines · by baba
- Israeli AI firm Two Delta raised tens of millions from investors Bill Ackman and Eric Schmidt.
- The company's DCOCO technology reportedly doubles AI model throughput and improves efficiency.
- Two Delta's research shows AI agents struggle to maintain application integrity over time.
- The company was founded in 2025 by Dvir Zeltzer and Dr. Eli David.
- Nvidia collaboration accelerated the Nemotron 3.5 Super AI model.
Israeli artificial intelligence company Two Delta has secured tens of millions of dollars in a seed funding round, attracting prominent investors including Bill Ackman, founder and CEO of Pershing Square, and former Google Chairman Eric Schmidt. The company, established in 2025, has maintained a high level of secrecy, with limited public information available beyond its research publications.
Two Delta's core technology aims to enhance the speed and efficiency of running AI models, optimizing the computational resources required. The company is co-founded by Dvir Zeltzer, who brings a business and management background including a senior role at Google, and Dr. Eli David, a seasoned AI researcher and entrepreneur.
In October 2026, Two Delta published results from a collaboration with Nvidia, detailing a method called DCOCO designed to accelerate the Nemotron 3.5 Super AI model. DCOCO reportedly increases the output of text generated by the model per processing step without compromising quality. It builds upon Speculative Decoding by using contextual information to select more consistent word sequences, thereby increasing the number of tokens processed in each check.
According to the company's experimental results on ten datasets, DCOCO achieved an average throughput 2.03 times higher than standard decoding under deterministic processing. Compared to the DFlash method, the average throughput improvement was 19.8%, a benefit that reportedly persists under high concurrent workloads.
Additionally, Two Delta released research on EvoWeb, examining the long-term ability of AI agents to build and maintain applications amidst ongoing user requests for modifications. Unlike standard tests that assess single-prompt application building, EvoWeb evaluates cumulative development. The research found that none of the tested models, including leading ones like Fable 5 and Kimi K3, maintained full application integrity over time, with error rates increasing significantly as complexity grew. This highlights a gap between AI agents' code generation capabilities and their ability to sustain functional applications as they become more complex.