11/09/2025 What I Read Last Week
#DigitalInfrastructure news: Google TPUs for inference and space, Microsoft signs deal with Iren, VAST Data signs deal with CoreWeave, quantum computing, and much more.
Weekly edition of curated news about Digital Infrastructure
[Link] Google TPUs to power inference. Google announced the availability of three new products built on custom silicon that deliver exceptional performance, lower costs, and enable new capabilities for inference and agentic workloads. The 7th generation TPU Ironwood is purpose-built for the most demanding workloads: from large-scale model training and complex reinforcement learning to high-volume, low-latency AI inference and model serving. The new Arm-based Axion instances offer up to 2x better price-performance than comparable current-generation x86-based VMs. Google also announced C4A Metal, the first Arm-based bare metal instance, will be available in preview soon. Lots of good data center pictures in this article - like a part of an Ironwood superpod, directly connecting 9,216 Ironwood TPUs in a single domain.
[Link] Google TPUs to power space. Google announced Project Suncatcher, a new research moonshot to one day scale machine learning in space. Working backward from this potential future, Google is exploring how an interconnected network of solar-powered satellites, equipped with Tensor Processing Unit (TPU) AI chips, could harness the full power of the Sun. Google has demonstrated 1.6 Tbps satellite communication, confirmed TPUs can handle space radiation, and projects that falling launch costs could make it economically viable by the mid-2030s.
[Link] StarCloud CEO Philip Johnston was on NBC News last week, talking about the race to build data centers in space.
[Link] Microsoft has signed a 5-year, $9.7 billion deal with Iren to access its Nvidia GB300 GPUs. The GPUs will be deployed in Childress, Texas, in phases through 2026 as liquid-cooled data centers are delivered. Iren also announced an agreement with Dell to purchase the GPUs and ancillary equipment for approximately $5.8 billion.
[Link] Fermi America, in partnership with the Texas Tech University System, announced the Texas Commission on Environmental Quality (TCEQ)’s preliminary approval of 6 GW of clean natural gas-based power generation, of the projected 11 GW campus, which will make Project Matador one of the largest clean natural gas power generation facilities in the world.
[Link] VAST Data has signed a $1.17 billion commercial agreement with CoreWeave, solidifying VAST’s AI Operating System as the primary data foundation for CoreWeave’s AI cloud. The expanded partnership enables scalable, high-performance infrastructure for AI training and inference, with seamless deployment across data centers.
[Link] Vantage Data Centers is investing $2 billion to develop a 192MW hyperscale campus (VA4) in Fredericksburg, Virginia, creating 1,100 construction jobs and 50 permanent roles while expanding its statewide footprint to 782MW across four campuses.
[Link] Energy provider Calpine and CyrusOne have expanded their power supply agreement to 400 MW for a hyperscale data center at the Thad Hill Energy Center in Texas, with the new 210 MW phase building on the initial 190 MW to ensure reliable power, grid connectivity, and operational coordination by Q4 2026.
[Link] Akamai is seeing strong early demand for its Akamai Inference Cloud platform, which brings AI inference processing to the edge rather than centralized data centers. The platform, announced at NVIDIA’s GTC Conference, is being used across industries for applications including 8K video workflows, live sports production, personalized shopping recommendations, AI-powered chatbots in gaming, and interactive toys.
[Link] London’s data centre boom accelerates with three major schemes worth over £10 billion announced in one week, driven by AI demand. Colt Data Centre Services (£2.5bn, Hayes): Expanding campus with 3 hyperscale centres + innovation hub; adds 97MW (total 160MW); Ark Data Centres (£2bn, near Watford): Up to 6 data centres on former hotel site; up to 200MW; potential Europe’s largest campus. Equinix (£3.9bn, South Mimms, Herts): 250MW colocation campus on 85-acre site near M25; acquired from JV.
[Link] BrainChip announced its AKD1500 neuromorphic AI co-processor chip. The chip delivers 800 GOPS while consuming under 300 milliwatts, making it highly efficient for battery-powered and heat-constrained edge AI applications. It integrates with x86, ARM, and RISC-V platforms and supports on-chip learning—a key advantage over conventional AI accelerators.
[Link] PsiQuantum and Lockheed Martin have signed a memorandum of understanding to advance quantum computing applications in aerospace and defense, focusing on developing algorithms for U.S. government and allied needs using PsiQuantum’s fault-tolerant systems and Construct software platform. This collaboration aims to leverage utility-scale quantum computing to enhance simulations in areas like fluid dynamics, propulsion, and materials science, building on PsiQuantum’s recent $1 billion Series E funding.
[Link] DARPA (Defense Advanced Research Projects Agency)has selected 11 companies to advance to Stage B of its Quantum Benchmarking Initiative, which aims to determine if any quantum computing approach can achieve utility-scale operation by 2033. The 11 companies taking part in Stage B are:
[Link] IonQ
[Link] IBM
[Link] Quantinuum
[Link] QuEra
[Link] Atom Computing
[Link] Nord Quantique
Diraq, Photonic, Quantum Motion, Silicon Quantum Computing, and Xanadu.
[Link] Good article: How Virtual Power Plants Can Help the United States Win the AI Race. Virtual power plants (VPPs)—networks of distributed energy resources like batteries, EVs, and smart thermostats—can rapidly provide grid capacity to support data center growth while traditional power infrastructure faces years-long delays from supply chain bottlenecks and interconnection queues. The article proposes novel commercial models where data centers directly fund VPP development in exchange for expedited grid connection, offering a faster and more cost-effective solution than conventional generation and transmission upgrades.
Disclosure: I own incredibly small stock positions in Microsoft, BrainChip, and Fermi.
Follow me on X for more links and news.
Coinbase is the easiest (and best) place to buy and sell cryptocurrency. Use my referral link to sign up.
Couldn't agree more, reading about TPUs going into space with Project Suncatcher really got me thinking, what if this literally elevates AI capabilities beyond our current terestrial limits, creating a truly global brain for humanity's biggest problems?