1. AI as a strategic imperative: power, control, and dependencies
The AI race is no longer a contest between labs posting benchmarks. In the past few months, the story has become physical.
A data-center operator weighing a $100 billion valuation or sale. A $500 billion financing effort forming around the world’s dominant chipmaker. A utility reporting data-center power demand up 55%. A Chinese foundry raising wafer prices because AI wants more chips. A county in California voting to draft rules for humanoid robots.
Different headlines, different events, but they all tell the same story. They are the surface of a structural shift. AI has become a strategic asset, and the countries and companies that control its infrastructure control the technology.
The global AI race is intensifying, driven by critical infrastructure build-out and a reactive but increasingly strategic regulatory environment. That combination is creating new dependencies and new geopolitical fault lines, and we are slowly moving from the polite meetings to the cold, hard competition for power. But this time, that power is being built out of concrete, copper, and megawatts.
2. The power of computation behind AI infrastructure
The binding constraint on AI is not simply model design, or the end-user pricing. The name of the game is compute, and compute is physical. Every frontier model needs training clusters and inference capacity, and those need land, power, and cooling.
The scale of capital entering the sector tells the first half of the story. Vantage Data Centers, one of the largest private operators in the world, is exploring an IPO at a $100 billion valuation or a sale. $100 billion is not a niche infrastructure bet. The market is pricing data centers as core national infrastructure, the same way it prices ports and pipelines.
The second half is energy. Southern Company, a major U.S. utility, reported data-center power usage up 55% year over year, with a contracted large-load pipeline at 17 gigawatts. Europe is responding in kind. RWE says it is close to deals with data center operators at two sites, and Helios Nordic Energy is planning an 80-hectare data center in Finland.
Where power is cheap and land is abundant, the industry follows.
Cooling has quietly become its own industrial sector. Cooling is the dirty part of the operation. It is not sexy. It is not photogenic for press releases. It is not even interesting. But without cooling, no computer works, and this bare fact pushed cooling tech into one of the fastest-growing sub-sectors within AI infrastructure. Daikin is allocating $270.2 million for a cooling facility in Tijuana, projecting about 8,000 jobs by 2029. Data centers consume electricity, and in parallel, generate enormous heat, so managing that heat is now a manufacturing and employment story.
Even the direction of capital is shifting.
Hyperscale Data sold roughly 685 bitcoin to fund AI data-center expansion. A company that mined a cryptocurrency is selling its holdings to build AI infrastructure. After spending years building its mining setup and slowly piling up all that bitcoin as a reserve asset.
When miners repurpose their balance sheets and facilities toward compute, the signal is crystal clear. The demand for physical AI capacity now outbids even the most dedicated crypto assets. On the demand side, Asian operators are riding the same wave. GDS reported second-quarter revenue up 65% year over year on AI data-center demand, and Rowan Digital Infrastructure is proposing a $700 million sixth data center in Temple, Texas.
This matters to the average person. It is not abstract. The price of compute and its fluctuation show up in subscription prices, and data-center demand shows up in power bills, in grid strain, and in land prices too. When a utility’s power load jumps 55% in a year, somebody’s rates and somebody’s local economy feel it.
The countries that can produce cheap, reliable power become the hosts of the AI economy, just like they became host countries for bitcoin mining. The ones that cannot become its importers.
3. Semiconductor sovereignty
If compute is the machine, the chip is the choke point. Whoever controls advanced manufacturing controls the pace of AI, and governments now treat chip supply as national security rather than trade. Back in the day, trade restrictions were usually about certain weapons. Today there are talks about tightening the flow of chips harder (while export control is already in place), and who knows, maybe tomorrow the chips will be on the same list as nuclear materials.
AI being a strategic imperative is not a euphemism. Not anymore.
The evidence is concrete. SMIC, China’s leading foundry, is raising wafer prices on strong AI demand, with second-quarter revenue topping $3 billion. A foundry that can raise prices because AI wants its output is pricing power that did not exist a few years ago, and it is a direct consequence of the demand-pull in the previous chapter.
Countries are responding by building their own capacity. India Chip, a joint venture between HCL and Foxconn, has established a DDIC and OSAT facility in Jewar with a $500 million investment. Packaging and assembly are where a country enters the chip supply chain without building a full fab, and India is using that door. Larsen & Toubro is partnering with Together AI to deploy an NVIDIA B300 AI Factory in Chennai. NVIDIA, meanwhile, is seeding ecosystems globally, opening Indonesia’s first university AI center with Indosat and UGM.
The pattern is deliberate. The dominant chipmaker is building both demand and talent across the global south, not just selling chips. Some even say it is like colonization with glass pearls that are actually worth something.
The enabling silicon matters as much as the flagship GPU. Microchip launched a 160-lane PCIe Gen6 switch and retimer at FMS 2026. The enormous data flows inside an AI data center depend on these switches. A bottleneck here slows the whole machine, which is why even the “boring” silicon is now a strategic asset.
The fault line is the dependency.
The United States has export controls on advanced chips. China is scaling domestic foundries like SMIC. India and others are building packaging and assembly. The result is a supply chain that is deliberately fragmented along national lines, and in that fragmentation, “strategic dependency” has become the risk word of the decade. Every country that wants AI leadership must answer the same question.
Who makes your chips, and what happens to your AI ambitions if they stop?
4. Regulatory frameworks
Regulation is reacting to a moving target, and it is splitting into two logics. Safety standards on one side, industrial policy on the other.
The safety logic is visible at every level. San Mateo County, California, voted to draft a humanoid robot permit ordinance. This is local rulemaking for a technology most people have not touched, and it signals that regulators are not waiting for a federal framework. The security dimension is accelerating too. AI-driven cyberattacks surged against Spanish companies, and an Iranian regulator warned about AI-driven digital self-medication risks.
Regulators are waking up across the world, from cyber to health to labor, simply because people have already started to use these AI tools en masse, seemingly regardless of the risks.
The industrial-policy logic runs alongside. The same governments writing safety rules are also subsidizing fabs, data centers, and research. That is the tension at the heart of the regulatory race. Too little regulation risks runaway harm, but reactive bans and heavy rules can push activity offshore or stall the very investment the state is trying to attract.
The signal to watch is the direction.
Not a simple rule, but the full policy, because regulation is becoming a competitive instrument. A country that writes clear rules attracts capital. One that reacts with uncertainty pushes it away. This has happened with every innovation, in every field, everywhere in the world since the dawn of the industrial revolution.
The AI version is just bigger and faster, and it happens in front of our eyes. History in the making.
5. Interconnected risks and opportunities
The three axes of compute, chips, and regulation feed each other. A lead on one creates advantages on the others, but a dependency on one becomes a vulnerability on all three.
The Nvidia-OpenAI-Ohio story is a case study in how fragile these arrangements are. Nvidia scaled back a roughly $250 billion funding guarantee for the Ohio OpenAI data center, and is separately in talks to invest up to $3 billion in SB Energy as part of the same project. These are enormous numbers that keep moving.
Capital, compute, and power are being assembled in real time, and mega-deals are being renegotiated as they scale. The infrastructure stack is still being built, and nobody has a finished playbook.
There are also lead-lag signals worth watching. Power utilities and chipmakers may be the earliest, most honest indicators of AI’s real pace, because they are paid for physical demand, not for narrative. Regulators lag the farthest. When a utility reports 17 gigawatts of contracted load, that is harder to fake than a benchmark score. Investors and enterprises that want to separate signal from hype had better watch the physical layer first, because maybe that tells the real story.
The opportunity is equally real.
India’s chip and AI-factory build-out, Indonesia’s university AI center, Finland’s cheap-power data centers, and Texas data-center expansion all show capital and capability spreading beyond the usual hubs. The race is concentrating some power, but it is also dispersing the infrastructure of the technology.
6. National security, economic leadership, and cooperation
The race will probably not be settled by a single model, but by whoever can build, power, and supply the physical machine, and by whether the world can manage the dependencies it is creating.
The stakes are threefold.
For national security, compute is now a strategic asset, and the chip is its choke point, so the countries that control the supply chain hold advantages over the countries that do not.
For economic leadership, data centers and fabs are the new engines of jobs, investment, and trade. So the ones that host them capture a disproportionate share of the value.
For international cooperation, the danger is fragmentation. A world where each bloc builds its own rails and dependencies harden into rivalries.
For the average person, the consequence is concrete. The race will show up in electricity prices, in where the jobs land, in the cost of every digital service, and in the balance of power that decides what the technology is allowed to do. Who is even allowed to use that tech. The dependencies being built now will outlast any single company’s lead.
The race is real, and it is not predetermined.
But it will not be decided in a lab, or in political meetings, or by philosophical arguments. The reality will be the judge. The race will be decided by whoever can switch on the lights, keep the servers cool, and make the chips, and by whether the world chooses competition over cooperation, or finds a way to do both.
