Anatoly Guin: Boris, I will begin this dialogue with a piece of news that is currently circulating among artificial‑intelligence specialists. Two global companies that are on everyone’s lips—OpenAI and NVIDIA—have agreed to jointly create a truly colossal data‑center. It will be an entire city with a power consumption of 10 gigawatts, enough for Marty McFly and Doc Brown to jump in time a whole 8 times. NVIDIA is investing $100 billion in the project.
Boris Zlotin: The data‑center’s energy is consumed by the very expensive “transformer” devices on which AI systems such as ChatGPT run. Our PANC™ requires no training, so those energy‑guzzling depths are unnecessary for it. In other words, when we compare energy efficiency on large data sets, PANC™ will consume thousands of times less power than today’s AI.
Anatoly Guin: Thus, the dangerous idea of small nuclear reactors on every corner to power AI—an idea championed by Sam Altman, who is raising $5 billion to start it—will become irrelevant once PANC™ is deployed.
Boris Zlotin: There is another side to the story. Data‑centers not only demand huge capital outlays and consume massive amounts of energy, they are also unreliable. The worst case is that, amid the mountain of equipment, a cable is severed, a lightning strike occurs, voltage spikes, a human makes a mistake, and the whole training (or inference) process must be started again from scratch.
Anatoly Guin: So Elon Musk and others, unaware of PANC™, are wasting money on hardware and energy? Will those enormous expenditures be lost to humanity?
Boris Zlotin: That is indeed a problem. Yes, money is being spent wastefully, but not entirely without purpose—that’s the paradoxical yet correct answer.
Why wasteful? A large‑language model such as Chat GPT must perform on the order of a trillion multiplications of enormous matrices, each containing roughly 150 million digits, merely to answer a question like “What are the circumference and diameter of the Earth?”. Those are unimaginable figures with massive costs hidden behind them.
In PANC™ there is also a matrix‑multiplication operation, but the matrices are thousands of times smaller, and there is only a single multiplication. A home computer can handle it without overloading the power grid. The electricity demand of a PANC™‑based AI is modest and does not require the construction of new power stations.
Why the spending is not entirely wasteful: the surplus energy will be useful to humanity, powering improvements that benefit a huge mass of people. Millions of transformers will still be needed. However, for good AI built on PANC™ ordinary network servers—costing hundreds of thousands of dollars rather than billions for supercomputers—are sufficient.
The transformers already built, and those yet to be produced, will not be lost. With modest upgrades and the comparative thinking approach embodied in PANC™, they will enable what was once only a dream. Thirty years ago we proposed the concept of a “second self”, an alter‑ego that constantly accompanies a person and assists in all practical activities.
These transformers will make it possible to provide every person on Earth with a personal computer companion. That would be a social transformation of the same magnitude as the advent of language, the mastery of fire, and the development of agriculture. Consequently, I view these computing centers positively. They will become centers of assistance for people.
Anatoly Guin: It is very encouraging to hear that. Combined with other technologies—such as planet‑wide Wi‑Fi envisaged by Musk through Starlink, the Internet of Things, and advancing robotics—humanity could become a “god‑like” entity, as Pierre Teilhard de Chardin once imagined.