Recently, Chrstopher Mims and Nate Rattner of The Wall Street Journal penned an article titled “When AI Hype Meets AI Reality: A Reckoning in 6 Charts,” that explored the “riskiness of the world’s collective investment in artificial intelligence.”
Available only to subscribers behind the WSJ’s paywall, it’s a notable read on numerous levels, but one particular theme, or perhaps more accurately one particular question, stands out for its inescapable relevance: that question, posed by the authors is “Can we even build all the necessary infrastructure?”
Regardless of where you stand on the answer, one thing is certain. AI, in its current form requires more of virtually everything – more land for data centers, more municipalities to approve them, more people to construct them, more components and hardware to build them out, and dramatically more power to keep them up and running.
If you follow our blog, you know that at Atombeam we have strong feelings about the limitations of the current approach to AI, particularly the inherent and fatal flaw in Large Language Models, namely that they start from scratch with every query and rely on brute force compute power to get the job done.
That’s never been the approach of effective computing innovations, which excel by doing more with less – fewer resources, lower costs, and less effort. With LLMs we have AI that requires more of everything. And the more we use them, the more resources they need. Yes, they are incredible for analyzing massive data sets, but as the foundation of AI they are woefully inadequate and inefficient.
In contrast, our solutions and innovations at Atombeam address and overcome those inherent weaknesses. Neurpac, and our Data-as-Codewords technology fundamentally and exponentially increase the available bandwidth in networks and enable enterprises to move more data, more securely and with less effort. With baseline available bandwidth gains of 4x and often significantly more as our pilots with the U.S. military reveal, data centers achieve the same throughput with smaller pipes while requiring less compute and less storage.
In other words, Neurpac utterly changes the economics not just of AI and data centers, but all infrastructure – including wired, cellular and satellite networks. But that’s just the beginning.
Our Persistent Cognitive Machine, or PCM, takes our disruption of AI even further. With PCM, we see AI that truly learns – not which simply pulls up a fact or two with RAG and other memory schemes, but actually gets to know you. And because of how it is architected and how it truly learns it is light enough to fit on a device, such as a cell phone.
The human brain only requires about 20 watts of power to provide true intelligence. In contrast, AI today, with its LLM underpinnings requires data centers that require as much power as a city, without getting closer to true learning, true intelligence or true answers. That is why at Atombeam, we believe the time for a new approach to AI has come. The future is now.
Thankyou
