Even a digital breakthrough needs a place to run.
A model can appear on your screen in seconds. Behind that simple experience is a lot of real-world work: making chips, supplying electricity, moving data, and keeping equipment cool. People have to build and maintain all of it.
Buying processors is only part of the job. A data center also needs a suitable site, reliable power, good connections, and software that uses the hardware well. Those details may not get the launch-day applause, but they make the launch possible.
The builders are getting to work.
America’s infrastructure story has names, places, and serious commitments behind it. Here are four examples worth following, from operating computers to the factories and facilities being developed for what comes next.
SpaceXAI: Colossus today, Terafab for what comes next.
SpaceX’s June 5, 2026 prospectus reported roughly 1 gigawatt of computing power across Colossus and Colossus II. Those facilities put substantial computing resources behind the company’s SI ambitions.
Terafab takes the effort into chipmaking. The prospectus describes collaboration with Tesla and Intel to bring chip design, manufacturing, memory, and advanced packaging together. The aim is to expand the supply of chips for Tesla’s vehicles and robots and SpaceX’s planned orbital computing systems.
That is a bold step toward building more of the foundations in-house. Colossus is operating infrastructure; Terafab’s full manufacturing ambition is a development project, not finished capacity.
SpaceX prospectus · June 5, 2026 · computing and Terafab plans · Terafab project site
Stargate: bring the computers and the power together.
OpenAI’s Stargate effort brings together data centers, chips, energy, and partners including Oracle and SoftBank. On January 9, 2026, OpenAI and SoftBank announced a combined $1 billion investment in SB Energy, alongside OpenAI’s 1.2-gigawatt data-center lease for the initial buildout.
On April 29, OpenAI reported it had surpassed its original goal of securing 10 gigawatts of U.S. infrastructure. Securing capacity is not the same as having every facility online, but it shows the scale of the effort.
OpenAI / SB Energy · January 9, 2026 · OpenAI infrastructure update · April 29, 2026
Amazon: a major commitment to the next wave.
In his 2025 shareholder letter, published in 2026, Amazon CEO Andy Jassy outlined approximately $200 billion in planned capital spending for 2026. The letter connects that spending to AWS growth, customer demand, and the company’s chip and computing ambitions.
That figure is Amazon’s company-wide, global plan—not a U.S.-only SI budget. It is a useful illustration of how much physical investment sits behind the services people use on screen.
Amazon · 2025 shareholder letter, published in 2026
TSMC: make more advanced chips in America.
On March 4, 2025, TSMC announced plans to bring its total U.S. investment to $165 billion, including additional chip factories, advanced packaging, and a research center. Its first Arizona facility began volume production in late 2024.
This is a long-term investment plan, not money all spent at once. It also shows how trusted international partners can help strengthen the manufacturing base on American soil.
TSMC U.S. investment announcement · March 4, 2025
Different projects, different schedules, one practical goal: give more people the computing resources and chips to build useful things. The figures above describe different commitments and should not be added into a single spending total.
Four pieces that need to work together.
01PowerReliable electricity and a site that can use it.
02ComputersChips, cooling, and connections that keep things running.
03ResearchTools and access for people with ideas worth pursuing.
04Useful workResults people can check, trust, and put to use.
This is a way to think about the buildout, not a rating of individual projects. A good plan should explain who will use the capacity, what they need it for, and how the surrounding community will benefit.
Building more is one route forward. Getting more from what we already have is another. Better software, more efficient chips, and good engineering belong in the same conversation.
Bigger computers are a start, not the destination.
A larger bank of computers does not automatically deliver General Super Intelligence. It creates room to try more ideas. What matters is what the systems can actually do. The capability scale describes performance, not the size of the equipment behind it.
There is an exciting possibility here: better systems could help design better software and hardware, giving the next generation more to work with. The singularity essay explores that idea—and what would have to go right.
