Welcome to this edition of the InfraOps Digest, in which we’ll unpack detours around restrictions, the spike in compute demand, the scramble for data center real estate, and what all of this means for your multinational business.
DeepSeek’s Circuitous Chip Path. U.S. officials say DeepSeek used Southeast Asian shell companies to sneak Nvidia H100 GPUs around export bans—then handed that firepower to China’s military and intelligence branches. Meanwhile, Nvidia’s Singapore-billing revenue ballooned from $0.5 billion (9%) in Q3 2023 to $7.7 billion (22%) in Q3 2025. As I explained, the intricate web of export controls only invites creative workarounds. Business entities are routing this hardware from Singapore through a labyrinth of different shell companies and routes into China. I don’t know why people think they can stop the flow of chips. Export evasion exposed
Microsoft’s Sovereign Chip Challenge. Designing in-house AI silicon isn’t easy. Microsoft’s Maia chip is six months delayed and trails Nvidia’s Blackwell—but the gamble underscores Microsoft’s commitment to owning its infrastructure and driving sovereignty. Of course, this isn’t just Microsoft—every hypercloud (e.g., hyperscalers, high performance computing firms, leading AI companies) depends on sovereignty for survival. Microsoft’s sovereignty has paid dividends: when OpenAI needed compute capacity, Microsoft offered its means of production in exchange for 20% of its revenue. No pain, no gain
Meta Taps Private Credit for $29 Billion AI Build-Out. Meta is raising $3 billion in equity and $26 billion in debt from Apollo, KKR, Brookfield and others to fund its U.S. data center expansion, sharing risk to secure compute capacity. Meta wants more resources for more sovereignty. For the fastest players, this is going to be a CapEx race. The fact that Meta is even leveraging at all—it reported marketable securities of $77.81 billion in 2024, and debt is not cheap right now—says a lot. If you buy into the notion that every business is becoming a technology business, your technical spend should probably reflect this. You don’t need to raise money like Meta did—but start considering how to deploy a greater CapEx towards your infrastructure and improve its efficiency. Debt-funded sovereignty
Compute is Going Bananas. A report suggests AWS, Azure and GCP now own 44% of global capacity—on track for 60% by 2030—while colocation vacancy hits all-time lows and rents jump 15%. If your business is a massive consumer of these hyperscaler companies, it makes sense for you to start managing your own workloads. Keep sensitive data and latency-critical workloads on-prem or in regional colos, while using hyperscale for burst compute to retain control over data sovereignty, performance, and compliance. Remember, if you’re renting, nothing is stopping your vendors from raising the rent on you. This should be a very scary stat if you’re running an enterprise business. The cloud squeeze
Apple Considers Outsourced AI for Siri. Apple is exploring Anthropic’s Claude and OpenAI’s ChatGPT to power its next Siri release, putting in-house models on hold to fast-track feature parity. This would be a big mistake. While its rivals spend tons of money to bet on their own sovereignty regarding AI—and the world is effectively saying that it’s going to be the new internet layer—Apple seems to be tapping out. It’s playing right into the trope that Apple can’t think of anything better to do other than stock buybacks ($656 billion in the past decade). Surrendering sovereignty