The pricey debt behind the AI boom, infrastructure projects in space and more in this week’s Prototype. To get it in your inbox, sign up here .

My colleague Giacomo Tognini has an absolutely must-read article about the expensive infrastructure boom sparked by the growth of AI, where he highlights the fact that “estimates of the total cost of the buildout range from $2.8 trillion by 2030 to $10.3 trillion by 2032 in the U.S. alone.”

That’s a staggering figure, potentially representing a greater percentage of America’s GDP than the railroad boom of the 1880s and 1890s, which minted a great number of fortunes. A lot of that, as Tognini points out, is being financed by debt–at least $1.3 trillion so far. That debt is starting to run into the greater macroeconomic uncertainties of our time: higher interest rates on treasury yields are pushing up interest rates on bonds for data centers, with some reaching rates as high as 9.75%.

Paying for that debt puts more pressure on AI companies to actually deliver on promises of revenue. But some of that promise fell to Earth a little bit this week, when a leak of Anthropic’s prospectus seemed to indicate the company only took in $4.6 billion in revenue last year, with operating losses alone of more than $8 billion–a figure that does not account for the company’s hundreds of billions worth of present and future capital expenditures.

That said, Anthropic did claim to have hit $65 billion in annualized revenue this past July, but that’s not the same as it making $65 billion this year. That figure will likely be much lower. SpaceX, for example, is expected by analysts to reach about $14-16 billion in AI segment revenue this year–though a chunk of that comes from a deal it struck to provide compute power for Anthropic. Those figures are certainly nothing to sneeze at, but they’ll need a lot more before seeing any return on their unprecedented investment.

Complicating these efforts are growing concerns over AI safety. The leaders of AI companies are being more vocal about improving guardrails around their software, but high-profile security and safety incidents keep piling up–and are starting to attract regulatory scrutiny. Last week, the Federal Trade Commission opened an investigation into OpenAI and Anthropic and California appears to be following suit.

That matters because one of the biggest barriers to AI adoption by potential customers are concerns about security and safety, according to a report from McKinsey earlier this year. “Organizations are less constrained by experimentation capabilities and more by confidence in their ability to safely deploy autonomous systems at scale,” the report states.

If that continues to be the case, AI companies may find it increasingly difficult to pull in the money they need to keep the AI boom rolling.

A Big Week For Space Infrastructure

As I previewed in the last edition , SpaceX did launch its Starship rocket into orbit last week. In the process, it delivered 26 of its V3 Starlink satellites into orbit. However, it had issues with its engines, and so the orbital flight was cut short for a controlled splashdown into the Pacific Ocean. The next Starship test flight is tentatively scheduled for later this month, which will test the company’s ability to recover the upper-stage of the spacecraft and potentially the booster as well.

Success here is key for the company’s plans, especially around its internet service Starlink. The 26 satellites that were just deployed bolster the company’s bandwidth capacity and therefore its revenue opportunities. Still, there have been persistent problems with the spacecraft’s engines, including this time around, which I’m sure are going to be closely looked at before the next launch. As analysts from Clear Street wrote in a research note, “any meaningful slip in launch cadence or technical reliability creates cascading risk across the entire model.”

SpaceX had two other launches last week, too. In Florida on Thursday, the company sent four crew members to the International Space Station for NASA in a mission that marked the fastest time from launch to docking in the history of the orbital facility.

Also on Thursday, on the other side of the country in California, SpaceX launched its Transporter-18 mission, which launched 130 different payloads for various customers. Of particular interest to me:

  • Google’s Project Suncatcher satellite prototype, which aims to test the company’s design for orbital data centers. On board the small spacecraft are four AI chips and solar panels to power them.
  • Starfish Space’s first Otter spacecraft, this small spacecraft is designed to autonomously service its customers’ satellites. For this initial mission, the company will test its ability to approach and inspect different objects in space.
  • Star Catcher, which plans to demonstrate power beaming technology from its Protostar satellite. The company plans to build a constellation of satellites that aim to provide power to other spacecraft in orbit.
  • Also testing power beaming is Cowboy Space, whose first spacecraft Reason-1, aims to demonstrate beaming solar power from space to a ground station on Earth. I spoke to founder and CEO Baiju Bhatt about the company’s plans to build orbital data centers earlier this year .

There are many more missions that launched Thursday beyond these, ranging from communication to imaging satellites, highlighting the continuing growth of economic infrastructure in the final frontier.

Anthropic v The Pentagon, Week 31: A federal court last week rejected Anthropic’s challenge to the Pentagon labeling the company a “supply chain risk”–which means that the Defense Department can bar the use of Claude on its systems. That’s in contrast to a different ruling earlier this summer that found the designation illegal, but the issue is that different laws were applied in each case. Anthropic hasn’t filed an appeal of this decision yet but said it was weighing its options.

Grok At War: Speaking of AI and the military, Time reported this week that President Donald Trump consulted Grok, SpaceX’s AI chatbot, before making his decision to order the U.S. military to capture Venezuelan president Nicolas Maduro earlier this year. “Trump, according to officials, came away thinking Grok was ingenious,” the report said. It’s not clear whether the president also used the system to shape his thinking on the war he launched against Iran.

Science Funds Diverted: The Trump Administration is withholding funds allocated to the National Science Foundation by Congress for basic research, instead wanting to move $1.4 billion towards AI and quantum computing. This follows a move earlier this month to divert research funds Congress appropriated to the National Institute of Health for research grants to the Pentagon, instead.

Quantum Stuff: New Mexico’s Roadrunner Quantum Lab, a 35,000 square foot campus for quantum computing, has opened its doors . I spoke to New Mexico governor Michelle Lujan Grisham about this project last year . In other quantum news, the NSA announced it’s taking more steps to transition computer systems involved in national security to algorithms that are resistant to attempts at decryption by quantum computers.

Hospital Robots: California-based Relay Robotics announced last week that it has deployed two robots in the HCA Florida JFK Hospital. The robots are designed to autonomously deliver items within the hospital, such as equipment, lab specimens and medications, saving professionals time in an era where healthcare providers are seeing shortages in nurses and other staff.

What’s Entertaining Me This Week

I recently read Far Sector , a Green Lantern miniseries written by N.K. Jemisin, the Hugo award-winning author of the Broken Earth trilogy (which, if you haven’t read, you should). It introduces Sojourner “Jo” Mullein, a new Green Lantern who’s been assigned to the City Enduring, where there hasn’t been a murder in 500 years. Until the beginning of the story that is. What follows is a sharp, fun sci-fi tale that stands as one of the better Green Lanterns stories. It’s not groundbreaking, but it does bring together a lot of interesting ideas and Jemisin’s typically strong character work shines here.