Atombeam has raised USD $18.3 million through a Regulation A+ offering on StartEngine, bringing its total capital raised to more than USD $55 million.
The California-based company will use the funding to support the commercial rollout of Neurpac, its data compaction software, and further develop its Persistent Cognitive Machine, or PCM, an AI architecture designed for continuous learning and persistent memory.
Neurpac is already in commercial production. The software is designed to reduce and secure machine data, and Atombeam says it can cut the amount transmitted by 50% to 90% before reconstructing the original at the receiving end.
The technology is intended to ease pressure on wired, cellular and satellite networks as AI workloads and machine-generated data volumes grow. Neurpac is also aimed at traffic types that conventional compression tools struggle to shrink, particularly small packets.
Commercial push
The fundraising comes as Atombeam seeks wider adoption of Neurpac and prepares PCM for external testing and customer trials. Its patent portfolio in this field includes 678 filings, with 308 issued or allowed and 370 pending.
Charles Yeomans, Co-Founder, Chairman and Chief Executive Officer, linked the new capital to both products.
"Closing this round marks an inflection point for Atombeam. We now have the capital, the team, and the technology to accelerate Neurpac adoption while bringing PCM through independent benchmarking and into customer pilots," said Charles Yeomans, Co-Founder, Chairman and Chief Executive Officer of Atombeam.
Yeomans also outlined the company's position in the market for AI infrastructure tools and newer AI models.
"Neurpac is in production with customers today, measured on their own traffic. PCM is something genuinely new: an AI that keeps learning after it ships and becomes an expert in the organisation it serves with no data centre behind it. In November we will put it beside conventional language models in independent benchmarks," he said.
AI infrastructure
Atombeam is positioning itself around a problem that has become more visible as use of large language models and connected devices rises: pressure on networks, storage systems and compute resources from growing volumes of data. It argues that both transmission efficiency and AI design must change if organisations are to manage those loads without sharply increasing infrastructure spending.
Neurpac is the first part of that strategy. According to Atombeam, the system stores intelligence in learned codebooks rather than embedding it in every message, allowing it to shrink data flows that standard approaches often leave largely untouched. The company says results have been measured on satellite, defence and commercial links, including validation on a major satellite operator's traffic.
PCM is the second part. Atombeam describes it as a separate AI architecture designed for continuous learning and real-time decision-making, with persistent memory and lower demands on compute, memory and connectivity than stateless large language models.
That places PCM in a category of systems intended for devices and other environments where conventional AI models can be difficult to use because of hardware or network limits. The design is meant to support AI processing closer to where data is generated rather than relying on a remote data centre for every task.
Broader claim
Yeomans said the company sees the two products as addressing linked parts of the same problem.
"With Neurpac we are solving the challenges large language models exacerbate - wired, cellular and satellite networks and data centres pushed to the brink by heavy AI workloads and the barrage of data from connected machines," he said.
"At the same time, with PCM we are creating a form of AI that addresses the weaknesses of LLMs that contribute to unprecedented network congestion and data centres that use as much power as small cities. In these ways we are using AI to address the challenges of AI adoption today while creating the innovations needed to realise AI's full potential in the future," he said.