AI Infrastructure | Sovereign AI | Investment
HCLTech and Sarvam to establish ₹142.57B sovereign AI data centre in Odisha
24 July 2026 | India | HCLTech, Sarvam and Government of Odisha
HCLTech announced plans to establish its first AI data centre in the upcoming Odisha Sovereign AI Park in Bhubaneswar, in partnership with Sarvam and the Government of Odisha. The planned capital outlay is ₹142.57 billion (₹14,257 crore), including financial assistance from the state government. The facility will combine high-performance compute, HCLTech’s full-stack AI capabilities and Sarvam’s foundation models to support sector-specific applications, multilingual services and sovereign workloads for public- and private-sector organizations. HCLTech separately announced a Global Technology Center in Bhubaneswar designed for more than 5,000 employees, expanding the wider technology and talent ecosystem around the investment.
Source: HCLTech – AI Data Center announcement • Reuters
Strategic Watch
Track the project’s implementation structure, government assistance, power and water arrangements, GPU procurement, construction milestones and the conversion of sovereign-AI demand into committed workloads. The ₹142.57 billion outlay materially exceeds HCLTech’s earlier ₹35 billion company-level AI data-centre investment plan, making financing and phasing central execution questions.
GreyRadius Insight
This project moves India’s sovereign-AI agenda from model development toward integrated compute, applications and public-sector deployment. The competitive advantage will depend on connecting infrastructure capacity with local models, regulated workloads, developer ecosystems and dependable last-mile delivery—not on data-centre scale alone.
Digital Realty raises 2026 outlook as AI and hyperscale leasing accelerates
23 July 2026 | Global | Digital Realty
Digital Realty raised its 2026 Core FFO per-share outlook, excluding net promote income, to US$8.15–US$8.20 after reporting strong second-quarter leasing and financial performance. Q2 revenue reached US$1.9 billion, up 29% year on year, while FFO was US$982 million, or US$2.73 per share. The company signed bookings representing US$307 million of annualized GAAP base rent at 100% share and ended the quarter with a record US$1.9 billion leasing backlog. Two additional hyperscale leases signed in July represented US$410 million of annualized base rent at 100% share, reinforcing continued demand for large-scale cloud and AI capacity.
Source: Digital Realty – Q2 2026 results • Reuters
Strategic Watch
Executives should monitor the timing of backlog conversion, the nine-month weighted-average lag between lease signing and commencement, customer concentration and the capital required to deliver hyperscale capacity. Leasing momentum is strong, but realized returns depend on bringing powered capacity online on schedule and preserving pricing discipline.
GreyRadius Insight
The results show that AI infrastructure demand is translating into contracted revenue, not only development announcements. A record backlog and large July hyperscale leases strengthen visibility, while also increasing the importance of execution capacity, powered-land availability and disciplined joint-venture capital structures.
Google named sole operator of 8M sq ft Georgia data centre campus
24 July 2026 | United States | Google
Google was named the sole end user and operator of the planned Columbia County Data Center development in Georgia. The project is designed to span approximately 8 million sq ft (743,224 sq m), with site plans indicating 23 data-centre buildings across roughly 1,900 acres northwest of White Oak Business Park. The scale would make it one of the largest hyperscale campuses announced in the U.S. Southeast and signals continued geographic expansion of cloud and AI infrastructure beyond established Northern Virginia and major metro hubs.
Source: DatacenterDynamics
Strategic Watch
Leadership teams should track final approvals, utility interconnection, generation and transmission requirements, construction phasing, tax arrangements and community commitments. At this scale, delivery risk will be determined by power availability and infrastructure sequencing as much as by land and building development.
GreyRadius Insight
Google’s role as sole operator gives the campus an anchor tenant with long-term workload visibility. For regional governments and utilities, the strategic issue shifts from attracting a project to allocating grid costs, protecting ratepayers and ensuring that local infrastructure and economic benefits keep pace with hyperscale demand.
Up to US$5B to Anthropic Alongside 2GW GPU Deployment
22 July 2026 | United States | AMD and Anthropic
AMD and Anthropic announced a strategic partnership under which Anthropic plans to deploy up to 2 GW of AMD Instinct MI450 Series GPUs using AMD Helios rack-scale systems. Deployment of the first gigawatt is expected to begin in the first half of 2027. The infrastructure will use MI455X GPUs, AMD EPYC “Venice” CPUs, Pensando networking and ROCm software for next-generation AI training and inference. AMD also committed to make a future strategic equity investment of up to US$5 billion in Anthropic, while the companies will use Claude to optimize AMD workloads and accelerate ROCm software development.
Source: AMD – strategic partnership announcement
Strategic Watch
Executives should track the first-gigawatt deployment schedule, manufacturing and supply-chain readiness, power procurement, Helios system performance and the maturity of ROCm for frontier-model workloads. The commercial significance will depend on translating the headline capacity into reliable, economically competitive production compute.
GreyRadius Insight
The partnership combines capital, hardware capacity and software co-development, illustrating how AI infrastructure competition is shifting toward vertically coordinated ecosystems. Anthropic gains supplier diversification and dedicated capacity, while AMD gains a flagship frontier-model customer capable of validating its full-stack platform at gigawatt scale.
From Recycled Water to Waterless Cooling
22 July 2026 | Australia | NEXTDC and OpenAI infrastructure project
Developers of the planned 612 MW S7 AI data centre in Western Sydney abandoned a recycled-water cooling plan because the pipeline infrastructure and planning permissions required to deliver treated water were unavailable. NEXTDC said the facility will instead use a closed-loop approach that circulates liquid coolant around chips and rejects heat through air, eliminating operational water use but potentially increasing energy consumption. The project illustrates how utility infrastructure constraints can force material design changes in hyperscale developments and shift sustainability trade-offs between water and power.
Source: Reuters • DatacenterDynamics
Strategic Watch
Track the revised facility design, power-usage implications, grid connection requirements, environmental approvals and whether off-site utility upgrades are funded by the developer. The case demonstrates that sustainability commitments must be tested against the availability and delivery timelines of local water, transmission and pipeline infrastructure.
GreyRadius Insight
Waterless cooling reduces exposure to water scarcity but can transfer the constraint to electricity supply. For AI campus developers, resilient site strategy now requires integrated optimization across compute density, cooling architecture, grid capacity, water availability and permitting rather than treating each utility as a separate workstream.
