00Opening read
The AI infrastructure boom is still finding money. Today made clear what that money now costs.
SoftBank raised $11.1 billion in the largest high-yield corporate bond sale on record, paying as much as 9.75% on its dollar debt as it funds a much larger bet on OpenAI and AI infrastructure.
At the same time, a new study estimates the U.S. buildout could require more than $10 trillion through 2032. Data centers are also becoming institutional infrastructure assets: a BlackRock and IFM-backed group is reportedly discussing a deal worth up to $25 billion for Stack Infrastructure's Asia-Pacific portfolio.
The signal is not that AI has run out of capital. Quite the opposite.
The signal is that the physical AI economy is becoming large enough to create its own credit market.
Power, chips and land still matter. Increasingly, so do leverage, yields, utilization assumptions and who ultimately carries the risk.
01What moved
AI debt reaches the high-yield market
SoftBank's new bond sale raised $11.1 billion.
The dollar tranches pay between 8.625% and 9.75%, a dramatic step up from the yields on SoftBank's 2021 issuance. Fitch told Reuters the sale shows AI-driven borrowing has now reached the high-yield market at scale.
That matters because the AI buildout is no longer being funded only from giant technology-company cash piles.
Capital markets are becoming part of the infrastructure.
The buildout gets a $10 trillion frame
A Columbia Business School study estimates U.S. AI infrastructure could require more than $10 trillion through 2032.
The estimate includes 183 GW of new data-center capacity over seven years.
The number is less important than the financing shift underneath it. Banks, private credit, real estate vehicles and other investors are increasingly helping fund a buildout whose future revenue remains uncertain.
The compute race is becoming a balance-sheet race.
Five million robots are already working in factories
The International Federation of Robotics says the global industrial-robot population reached 5 million last year.
Factories installed more than 600,000 new units in 2025. China alone installed 354,000, or 59% of the global total. The United States moved into second place with almost 38,500 installations.
Humanoids may dominate the conversation. Conventional industrial automation is where the installed machine economy already lives.
Rare earths remain a processing problem
The U.S. is investing heavily in domestic rare-earth supply, but Reuters reports it is still expected to depend on China into the 2030s.
The hard part is not simply mining. Rare earths have to be separated, refined and converted into magnets and other usable materials.
Yttrium shows how narrow the bottleneck can become. U.S. and Japanese officials recently met specifically to discuss shortages of the metal, which is used in aerospace, power equipment and semiconductor manufacturing tools.
The Gulf buildout gets a resilience layer
Microsoft plans to invest more than $10 billion across the UAE, Saudi Arabia, Qatar and Kuwait through 2030.
The company is also planning more than $400 million in regional connectivity.
The important word in Microsoft's strategy is resilience. After attacks on regional infrastructure, cloud capacity in the Gulf is being designed not just around cheap energy and sovereign capital, but around continuity, security and redundant networks.
02The throughline
Physical AI is entering a phase where the constraints no longer sit neatly in separate boxes.
A data center can have chips but lack power. It can have power but lack permission. It can clear both and still depend on expensive debt. A robot can have excellent hardware but no commercial workload. A mine can produce ore while the processing layer remains abroad.
That changes what deserves attention.
The next useful RCW question is not only who owns the constraint?
It is also who financed it, at what cost, and against what assumption?
The machine economy is becoming a balance-sheet economy.
03What to watch
- The Trump-Xi meeting today for movement on rare earths, chips and AI policy.
- September 25 for the full presentation of the study estimating more than $10 trillion of U.S. AI infrastructure investment.
- September 29 for settlement of SoftBank's record high-yield bond issue and its secondary-market reception.
- Whether the reported BlackRock-IFM bid for Stack Infrastructure turns into a definitive transaction.
04Source trail
Reuters, SoftBank record bond sale
https://www.reuters.com/business/media-telecom/softbank-issues-111-billion-bonds-openai-financing-push-2026-09-24/
Reuters, AI infrastructure financing risk
https://www.reuters.com/business/finance/financing-historic-ai-buildout-raises-systemic-risks-us-researcher-says-2026-09-24/
Reuters, BlackRock and IFM / Stack Infrastructure
https://www.reuters.com/legal/transactional/blackrock-ifm-close-25-billion-stack-data-center-deal-bloomberg-news-reports-2026-09-24/
International Federation of Robotics, World Robotics 2026
https://ifr.org/ifr-press-releases/news/five-million-robots-now-operate-in-factories-globally
Reuters, U.S.-China rare earths
https://www.reuters.com/world/china/chinas-rare-earths-dominance-overshadow-trumps-talks-with-xi-2026-09-24/
Reuters, U.S.-Japan yttrium discussions
https://www.reuters.com/world/china/us-japanese-officials-discussed-chinas-yttrium-curbs-ahead-trump-xi-meeting-2026-09-24/
Reuters, Microsoft Gulf investment
https://www.reuters.com/business/retail-consumer/microsoft-plans-10-billion-plus-gulf-investment-with-focus-resilience-2026-09-23/