AI Infrastructure Watch · November 18, 2026
AI Data Centers, Dedicated Compute Capacity, Liquid Cooling, Interconnects and the Financing Behind the AI Buildout
AI Infrastructure Is Being Built Around Power, Cooling, Memory and Capital
Welcome to this edition of the AI Infrastructure Watch, The CODEW’s intelligence stream covering the companies building the data centers, accelerators, networks, cooling systems and financing structures behind artificial intelligence. And here are the big stories: Giga Energy signed a 10-year, approximately $1 billion lease for its 60 MW Project Clay AI data center campus in Hunt County, Texas, with service expected in spring 2027. Nokia, Nscale, Skeleton, Telia and Verda formed AI Gigafactory Northern, a proposed Nordic-Baltic consortium seeking to build federated European AI infrastructure. Ecolab launched an AI infrastructure ecosystem with Siemens, Alfa Laval and Dassault Systèmes focused on liquid cooling, power and water efficiency.
TensorWave agreed to supply OpenAI with 457 MW of dedicated AMD-powered AI capacity across multiple U.S. sites. NVIDIA expanded its Vera CPU platform for AI and scientific computing. Marvell highlighted interconnect and memory technologies intended to address bandwidth and memory bottlenecks in larger AI systems. New U.S. restrictions on advanced AI chips and remote compute access put Southeast Asian data center expansion under greater scrutiny. And a reported Blackstone-backed transaction involving AWS data center campuses underscores the growing role of private capital in AI infrastructure financing.
The through-line: AI infrastructure is no longer defined only by GPU supply. It is increasingly shaped by access to power, water-efficient cooling, high-speed networking, memory capacity, customer contracts, and capital. The companies that can secure and integrate those inputs fastest will have an advantage as AI workloads scale beyond individual racks and data centers.
Editorial Note: This edition covers developments from November 11–18, 2026. Announced capacity, proposed projects, company claims, and reported transactions are distinguished according to their confirmed status. All sources are English-language. This article does not provide investment advice.
AI Data Centers & Capacity Expansion
10-Year Lease Covers 60 MW Project Clay Campus in Hunt County, Texas; Service Expected Spring 2027
Announced November 17, 2026 · Status: lease signed; campus under development.
What happened: Giga Energy signed a 10-year lease worth about $1 billion with an undisclosed U.S.-based AI cloud provider for Project Clay, a 60 MW AI data center campus in Hunt County, Texas. Giga broke ground in July and expects the campus to be ready for service in spring 2027.
Key players: Giga Energy, an undisclosed U.S.-based AI cloud provider, Hunt County stakeholders and the broader AI data center development market.
Why it matters: The deal converts a development-stage project into contracted demand before completion. It reflects the shortage of dedicated AI capacity and the importance of securing power, land, and construction capacity early. Giga’s GigaBase model uses off-site manufacturing and modular assembly to reduce construction time. The company says Project Clay will use dedicated power infrastructure and a closed-loop cooling system that does not rely on municipal or groundwater supplies for compute cooling. Those are company claims that should be assessed against operating performance once the facility is online.
Strategic implications: Specialized AI data center developers are competing with hyperscalers and established operators to deliver capacity quickly. Giga’s advantage is integration across land, power, manufacturing and operations; its challenge is executing a 60 MW campus on schedule. The spring 2027 service date will be the key test.
Source: Giga Energy; Reuters.
AI Gigafactory Northern Proposes Federated Sites Across Finland, Estonia, Latvia and Sweden
Announced November 16, 2026 · Status: proposed consortium bid; not an awarded or operating project.
What happened: Nokia, Nscale, Skeleton, Telia and Verda formed AI Gigafactory Northern, a consortium seeking to build a federated, multi-site AI infrastructure network across the Nordic and Baltic region. The group plans to bid under the European High-Performance Computing Joint Undertaking’s call for tenders for AI Gigafactory consortia.
Key players: Nokia, Nscale, Skeleton, Telia, Verda, VTT Technical Research Centre of Finland and KTH Royal Institute of Technology.
Why it matters: The proposed network would include facilities in Oulu, Finland; Iru, Estonia; and Valmiera, Latvia, with a Swedish site still to be determined. Rather than relying on one centralized campus, the consortium is proposing distributed AI capacity linked through secure connectivity and cloud infrastructure. Nokia would provide networking, Nscale and Verda AI cloud infrastructure, Telia connectivity and operations, and Skeleton high-power systems.
Strategic implications: The project aligns with Europe’s push for sovereign AI capacity and reduced dependence on non-European infrastructure. It could appeal to public-sector, research and regulated enterprise customers with data-residency and resilience requirements. But it remains a proposal: funding, sites, capacity targets, customers and construction schedules have not been finalized.
Source: Nokia and partners; Finnish AI Region.
Cooling, Compute & Networking
Partnerships With Siemens, Alfa Laval and Dassault Systèmes Target Liquid Cooling, Power and Water Efficiency
Announced November 17, 2026 · Status: ecosystem launched; commercial deployments to be monitored.
What happened: Ecolab launched an open AI infrastructure ecosystem at SC26, combining direct liquid cooling, smart thermal management, power systems, heat exchange and simulation. The initiative includes partnerships with Siemens, Alfa Laval and Dassault Systèmes.
Key players: Ecolab, Siemens, Alfa Laval, Dassault Systèmes, AI data center developers and operators.
Why it matters: Ecolab and Siemens are combining Ecolab’s liquid-cooling and coolant distribution unit technology with Siemens’ power distribution, automation, software and digital-twin capabilities. The companies say the partnership is intended to reduce deployment complexity and accelerate the time from design to first token. Ecolab’s CoolIT CHx 3000 coolant distribution unit initiative also includes Alfa Laval heat exchangers, pumps and valves.
Strategic implications: Cooling and power are becoming strategic constraints as AI racks become denser. Ecolab is positioning itself as an integrator rather than a component supplier. Its Dassault Systèmes virtual twin is designed to model trade-offs across water use, coolant performance, power efficiency and coolant distribution unit redundancy. The value of the ecosystem will depend on whether production deployments demonstrate measurable improvements in uptime, energy use, water use and deployment speed.
Source: Ecolab.
Six-Year Agreement Covers Dedicated AI Capacity Across Multiple U.S. Sites Using AMD Instinct MI455X GPUs
Announced November 16, 2026 · Status: agreement announced; capacity expected to begin coming online in H1 2027.
What happened: TensorWave announced a six-year agreement to provide OpenAI with 457 MW of dedicated AI infrastructure capacity across multiple U.S. data center sites. The deployment will use AMD Instinct MI455X GPUs and AMD Helios rack-scale systems.
Key players: TensorWave, OpenAI, AMD, data center partners, and the broader AI cloud market.
Why it matters: The agreement is part of the previously announced OpenAI-AMD strategic partnership. TensorWave will operate dedicated, single-tenant capacity for OpenAI rather than offering shared cloud resources. The 457 MW figure refers to IT capacity, not total facility power, which will be higher once cooling, power conversion, and other infrastructure are included.
Strategic implications: The deal gives AMD a major customer-facing deployment for its Instinct accelerator business and gives OpenAI another compute delivery channel. For TensorWave, the agreement supports financing but concentrates execution risk: the company must deliver power, cooling, networking, AMD systems and operational capacity on schedule. TensorWave says capacity is expected to begin coming online in the first half of 2027, with full deployment expected by early 2028.
Source: TensorWave / Business Wire.
Rack-Scale Platform Combines Vera CPUs, Quantum-X800 InfiniBand and High-Bandwidth Memory for AI and HPC
Announced November 16, 2026 · Status: platform announced; Vera CPU racks expected from partners in early 2027.
What happened: NVIDIA expanded its Vera CPU rack-scale platform, combining 192 Vera CPUs with Quantum-X800 InfiniBand networking. NVIDIA says the liquid-cooled system delivers more than 200 TB/s of aggregate peak memory bandwidth and includes approximately 300 TB of memory and 1 PB of storage.
Key players: NVIDIA, DKRZ, Bull, Dell Technologies, HPE, Lenovo, Supermicro and scientific-computing customers.
Why it matters: NVIDIA is extending beyond GPU acceleration into CPU-based high-performance computing. Many scientific workloads, including climate modeling and molecular dynamics, remain dependent on CPU performance, memory bandwidth and low-latency networking. The Vera platform is designed to address those requirements alongside AI workloads.
Strategic implications: DKRZ has selected Vera CPUs for its HLRE-5 climate-science supercomputer, which will also use NVIDIA GB200 Grace Blackwell systems and Quantum-2 InfiniBand networking. Dell, HPE, Lenovo and Supermicro are expected to offer Vera CPU racks beginning in early 2027. NVIDIA’s reported performance and efficiency gains are vendor-reported benchmarks, not independently verified results.
Source: NVIDIA.
SC26 Demonstrations Target Bandwidth and Memory Constraints Across AI Clusters
Announced November 13, 2026 · Status: technology demonstrations; commercial adoption to be monitored.
What happened: Marvell outlined technologies intended to address connectivity and memory bottlenecks in AI infrastructure. Its SC26 demonstrations include photonic interconnects, CXL memory expansion, Teralynx switching silicon, optical modules, active electrical cables and storage technologies.
Key players: Marvell, AI cloud providers, hyperscalers, data center operators, semiconductor partners and networking suppliers.
Why it matters: As AI workloads scale beyond single racks, accelerators can become underutilized if data cannot move quickly enough between compute, memory and storage. Marvell describes three scaling domains: scale-up within a growing compute-and-memory domain, scale-out between racks, and scale-across between data centers.
Strategic implications: Marvell’s portfolio spans switching, optics, interconnects, CXL memory and storage. That breadth positions it to address multiple bottlenecks in AI data centers. Its challenge is proving that these technologies improve real-world accelerator utilization, training time and cost per token rather than simply increasing component-level bandwidth.
Source: Marvell.
Policy & Capital
Restrictions Target Advanced Chip Shipments and Remote Access to Compute in Selected Countries
Reported November 17, 2026 · Status: policy reported; compliance and licensing details require monitoring.
What happened: The Trump administration said it would restrict shipments of advanced chips from companies including NVIDIA to Southeast Asian data centers, citing the risk that Chinese companies could use overseas facilities to access U.S. computing power. Singapore, Thailand, Malaysia, Indonesia and Brazil were identified as countries where advanced chips face a higher risk of diversion to China.
Key players: U.S. government, NVIDIA, cloud service providers, Southeast Asian data center operators, AI cloud providers and Chinese AI developers.
Why it matters: The reported restrictions extend beyond physical chip sales to remote access to computing capacity. That could require stronger customer verification, workload screening and licensing processes from cloud providers and data center operators serving international customers.
Strategic implications: The immediate effect may fall on GPU deployment and capacity activation rather than construction demand. Operators serving established U.S. hyperscalers may be better positioned to meet compliance requirements, while facilities dependent on opaque GPU-as-a-service arrangements or Chinese-linked demand could face delayed activation or weaker utilization. The policy could reshape customer mix, investment decisions and the geography of AI compute.
Source: Reuters.
Reported Acquisition of AWS Campuses With Leaseback Highlights Private Capital’s Role in AI Infrastructure
Reported November 18, 2026 · Status: reported transaction; original reporting and transaction terms require verification before publication.
What happened: A Blackstone-backed joint venture reportedly acquired AWS data center campuses that Amazon will lease back. The reported sale-and-leaseback structure would move the real estate off Amazon’s balance sheet while Amazon continues operating the facilities.
Key players: Blackstone, AWS / Amazon, data center real estate investors and the broader AI infrastructure financing market.
Why it matters: Hyperscalers face substantial capital requirements as they expand AI data centers and procure GPUs. A sale-and-leaseback can free capital tied up in real estate while preserving operational control of the facilities. The structure separates ownership of the physical asset from operation of the cloud infrastructure inside it.
Strategic implications: Private capital is becoming a larger participant in AI infrastructure because data center leases can provide long-duration, contracted cash flows. For Amazon, the reported deal would preserve service control while shifting real estate exposure to a financial investor. For Blackstone, it would provide exposure to hyperscale data center income. The transaction’s parties, campus locations, value, lease terms and closing status must be verified against the original reporting before treating it as confirmed.
Source: Slicast; The Register reporting to be verified.
AI Infrastructure Intelligence — The Trends Behind the News
Dedicated capacity is being contracted before it is built. Giga Energy’s $1 billion lease and TensorWave’s 457 MW agreement show that AI cloud providers are securing future capacity to reduce the risk of compute shortages. The trade-off is execution risk: announced capacity only becomes valuable when power, cooling, networking and accelerators arrive on schedule.
Cooling and water are becoming strategic infrastructure constraints. Ecolab’s ecosystem reflects the industry’s shift toward liquid cooling, integrated power management and water-efficient design. As rack density rises, cooling decisions increasingly affect deployment speed, energy cost, reliability and community acceptance.
Memory and interconnects are the next utilization bottlenecks. Marvell’s focus on photonic interconnects, CXL memory and switching reflects a broader recognition that GPU performance depends on data movement. Larger AI systems need bandwidth across racks and data centers, not just faster individual accelerators.
AI infrastructure is becoming a capital-markets business. The reported Blackstone-AWS transaction illustrates how sale-and-leaseback structures and private capital can fund the physical layer of AI expansion while cloud operators retain operational control.
The AI infrastructure market is shifting from a race for GPUs to a race for integrated capacity.
Giga Energy, TensorWave, Ecolab, NVIDIA, Marvell and the reported Blackstone transaction all point to the same constraint: AI compute is only useful when it is powered, cooled, connected, financed and operated at scale. The companies winning the next phase will not simply supply the fastest chip. They will control or coordinate the systems that turn electricity, land, memory, networking and capital into dependable AI capacity.
The most important story in this edition is the widening gap between announced and operational capacity. Giga Energy has a signed lease for a campus expected in spring 2027. TensorWave has a six-year agreement with capacity expected to begin coming online in 2027. AI Gigafactory Northern remains a proposed consortium. These are important commitments, but they are not yet operating infrastructure.
The caveat worth stating plainly: this edition rests on company announcements, Reuters reporting and credible industry coverage. Giga Energy’s construction-speed, cooling and water-use claims are company statements. NVIDIA’s performance benchmarks are vendor-reported. The Blackstone-AWS transaction is reported and requires verification of the original reporting and transaction terms. This article does not provide investment advice.
The CODEW Stat
457 MW. The dedicated IT capacity TensorWave says it will provide to OpenAI under a six-year agreement using AMD Instinct MI455X GPUs. The number illustrates the scale at which AI cloud providers are now contracting capacity—but it is not yet operating capacity. The decisive measure will be how much of that capacity comes online in the first half of 2027 and whether it can be delivered reliably at scale.
Sources & References
- Giga Energy — Project Clay 10-year, $1 billion lease announcement (November 17, 2026)
- Reuters — Giga Energy signs $1 billion Texas AI data center lease (November 17, 2026)
- Nokia and partners — AI Gigafactory Northern consortium announcement (November 16, 2026)
- Finnish AI Region — AI Gigafactory Northern consortium and proposed sites (November 17, 2026)
- Ecolab — AI infrastructure liquid cooling ecosystem announcement (November 17, 2026)
- TensorWave / Business Wire — 457 MW OpenAI agreement using AMD accelerators (November 16, 2026)
- NVIDIA — Vera CPU platform expansion announcement (November 16, 2026)
- Marvell — SC26 AI infrastructure connectivity announcement (November 13, 2026)
- Reuters — U.S. AI chip curbs and Southeast Asia data center expansion (November 17, 2026)
- Slicast — November 18 AI infrastructure news roundup; Blackstone-AWS transaction requires verification against The Register (November 18, 2026)
This edition covers developments from November 11–18, 2026. Announced capacity is not operational capacity. Proposed projects are not awarded or completed projects. Company-reported performance, efficiency and sustainability claims are attributed to the companies making them. The Blackstone-AWS transaction requires verification of the original reporting and transaction terms. All sources are English-language. This article does not provide investment advice.
Related Coverage
Reviewed by Erwin Castro
on
Wednesday, November 18, 2026
Rating:

No comments: