If you think tech companies are spending too much money on servers right now, you aren't paying attention to Masayoshi Son.
The SoftBank Group founder stood before executives in Tokyo and threw out a number so massive it sounds like a typo. By 2040, Son says the world will need to spend $5 trillion every single year to build out artificial intelligence infrastructure. That is roughly 800 trillion yen. Naturally, Wall Street is currently panicking about an AI bubble, wondering when the massive capital expenditure from tech giants will actually pay off. Son thinks those worries are completely ridiculous.
"To ask whether AI is a bubble is a foolish question," he told the crowd. He didn't stop there. He claimed that those who reject the technology are "closing down their world" and basically "spitting upward".
You have to admire the sheer audacity. While retail investors sweat over Nvidia's daily stock fluctuations, Son is looking decades into the future at a reality that requires rewriting the rules of global energy and finance. It is an incredibly bold bet, even for a guy who made his fortune by writing early checks to Alibaba.
The Math Behind the Madness
How do you even get to a $5 trillion annual price tag? Son didn't actually show his math, which is classic for him. Instead, he justified the figure by looking at the scale of the future global economy.
His core thesis relies on a massive assumption. By 2040, Son expects AI-related industries to account for roughly 20% of global GDP. In that context, spending $5 trillion a year on data centers, advanced silicon, and energy grids isn't a crisis. It's just a rounding error.
Think about what that infrastructure actually looks like. We aren't just talking about buying more microchips. To make this vision work, the world needs a physical overhaul:
- Massive data center complexes built at a scale never seen before.
- Advanced semiconductor manufacturing facilities capable of pumping out millions of next-generation processors.
- Entirely new power grids dedicated solely to keeping the machines running.
SoftBank isn't just talking, either. They're spending with terrifying urgency. The company's total investment in ChatGPT-maker OpenAI is on track to cross $60 billion before 2026 wraps up. They even launched a dedicated battery business in Japan to start building the electric power infrastructure needed to keep up with surging energy demand.
The Three Terawatt Energy Trap
Software is easy. Hardware and power are hard. That's the real bottleneck nobody wants to talk about, but Son isn't dodging it.
He predicts that by 2040, AI data centers will require 3 terawatts of electricity. To put that in perspective, that is nearly 1.8 times the entire world's current global power consumption. You can't run that on a few extra solar panels.
Son believes the transition will happen in two stages. In the short term, the tech sector will lean heavily on natural gas to keep the lights on. But the long-term savior? Nuclear fusion.
He even took a subtle dig at Elon Musk's ideas about beaming solar energy down from space. Son firmly believes that building fusion reactors right here on Earth will end up being the cheaper, cleaner, and more practical solution. Keep in mind, commercial nuclear fusion doesn't actually exist yet. Betting the future of global computing on an unproven energy source is incredibly risky, but Son has never been one to play it safe.
Moving From Humans to 100 Trillion Agents
Why do we need all this power and cash? Because Son expects the internet to get crowded. Very crowded.
He envisions a 2040 society populated by 100 trillion autonomous AI agents. These wouldn't be simple chatbots waiting for you to type a prompt. We're talking about software entities that make their own decisions, execute complex tasks, and talk directly to other AI agents without a human ever getting involved.
"We will go from a human-centric world to an agent-centric world," Son warned. He believes the era of humans being the highest life form on Earth will effectively end. Whether you think that sounds like a utopia or a sci-fi nightmare, Son insists it's unstoppable.
Should You Trust the Vision?
It's easy to get swept up in the grand rhetoric, but we have to look at the track record. Son is famous for his massive wins, like turning a $20 million bet on Alibaba into a historic fortune. But his hyper-aggressive style has also led to legendary disasters.
Remember WeWork? SoftBank poured billions into the shared-office company, buying into the hype that it was a revolutionary tech platform, only to watch it collapse into bankruptcy.
Right now, the market is deeply divided. Tech companies are burning through cash to buy chips, and Wall Street keeps asking a very simple question: where are the revenues that justify this spending? Son calls the question absurd because he's operating on a twenty-year timeline. If you're managing a quarterly pension fund, you don't have the luxury of waiting for nuclear fusion to power your portfolio.
If you are trying to navigate this landscape as a business leader or investor, you shouldn't blindly swallow the $5 trillion figure. Instead, focus on the structural shifts it highlights. The real money over the next decade won't just be in the flashiest new software apps. It will be in the unsexy physical infrastructure—the copper, the transformers, the power plants, and the real estate required to keep the algorithms running. Pay attention to the energy grid bottlenecks in your region, keep an eye on capital expenditure reports from the major cloud providers, and start experimenting with agentic workflows in your own operations today. The future might not hit $5 trillion by 2040, but the physical constraints of computing are going to dominate the economy regardless.
