Thomas sohmers He is a co-founder & CTO of Positron-Redefining AI with memory optimized material for the next AI.
The next five years, as and 90% of the software could be created AI. Depending on who you are asking, this can even be a conservative estimate. Anthropic’s Dario Amodei believes that AI could handle the entire code until next year. GitHub’s CEO thinks Copilot could write 80% of the new code.
But this is not just about replacing engineers. It is a fundamental shift in the way the work itself works. AI does not only automate work – redefines the very concept of human production itself. This changes the equation for productivity, economics and where and how value is created in the 21st century.
And here is the place that most people lose: every AI-created discreet, every work automated by an agent and every hour that is preserved by the machine intelligence-all run into silicon. The conclusion is the multiplier. And the chips that make this conclusion? Almost all this is done abroad.
This is a problem.
The conclusion is congestion – and the opportunity
Most people believe that the training of large linguistic models (LLMS) is the expensive place. Is not. Training is a lump sum event. The conclusion-the LLM “Thinking” models in real-time, to respond to users’ prompts, decision-making or registration code-is the cost center that escalates with any product launch, user interaction and integrated work.
And the conclusion is massive. In just a few years, AI computational demand It is expected to consume as much power as countries as Switzerland. If the ability to produce material is doubled this year, demand will probably escalate equally quickly and could represent almost half of all world data consumption by the end of the year, a trajectory that is already evident in the provisions of the data center and in the projections of energy. This is not excessive. There we direct. And yet, most of the chips The execution of these workloads comes from an island on the other side of the planet.
If you are not interested in, you should not, not because of nationalism, but because no serious executive will start their activities completely dependent on foreign infrastructure during a geopolitical knife race.
Why Made-in-America AI is not just patriotic-is strategic
Following Chips actThe construction of US semiconductors ultimately achieves the investment it needs. Intel spends $ 100 billion for new Fabs. Taiwan Semiconductor Manufacturing Company (TSMC) is trying to get its domestic internet functions. And specialized companies – including Positron – build the AI material in conclusions entirely on American soil.
Why does this matter?
Because the next generation of AI agents creating copilots, strategy assistants and multimodal reasoning tools-do not run yesterday’s infrastructure. They are very committed, very heavy and very fast. The future will not only be determined by who has the largest GPU clusters. It will be who can offer the best performance per watt, per dollar or shelf – without waiting for chips stuck in a Taiwan export queue.
We already see it. US companies are escalating faster because they have a local calculation. Federal contracts lean towards domestic infrastructure. Businesses choose stability in relation to the supplier locking. The companies that win at this time will be the ones that do not need to ask for a chain of supply for escalation.
AI work changes the mathematics of capitalism
Adam Smith He wrote about 250 years ago that “the labor was the first price, the original purchase money paid for all things”. That was true then, and it’s still true now. What is changing is that we are no longer limited by how much human work we can hire.
Today, a single AI server can overcome the dozens of lower engineers on a scale, at request and 24/7. This is not a point of speech. This is a basic financial. And the consequences are huge. For the first time in history, we introduce an exponential scale to a production agent that is always linear. The work now escalates like computational power.
This does not mean that people are outdated. It means we are free to work on higher leverage problems. And that means that computing power becomes Gatekeeper in productivity. Just as fossil fuels supplied the latest industrial revolutions, silicon authorizes this.
America has a plan to do it right
Let’s be honest. At the moment, about, about 80% of the AI conclusion It runs three companies – Nvidia, Advanced Micro Devices (AMD) and Intel. This is a recipe for fragility. If computational power is the new job and only a handful of companies control it, what happens to everyone else?
AI is the economical engine of the future. But without enough domestic material, it runs with borrowed time. If we want to stay in front of China, Europe or anyone else, we need a full stack of AI infrastructure, from fab to firmware and from chips to growth.
The good news? We build it – in Ohio, Arizona and in garages and cleaners across the country. If the latest industrial revolutions taught us something, this is: the winners were not the ones who were waiting for permission. They were the ones who built early and escalate quickly.
Let’s not lose our moment.
Forbes Technology Council It is a community only for an invitation for CIOS, CTOS and world -class technology. Do I qualify?
