Microsoft

Microsoft and G42 stall Kenya data centre talks over capacity guarantees amid rising demand risk in AI infrastructure.

The UK competition authority’s proposed Microsoft designation groups five business-software activities. Its own purchase data shows why an ecosystem view can matter—and why the much larger population of single-activity buyers needs its own evidence in any later remedy.

Microsoft’s 20-year Indiana power agreement adds about 1,100 MW of generation resources to NiSource’s GenCo pool, while a related 600 MW battery will serve multiple large-load customers. NiSource’s larger savings estimate covers existing NIPSCO electric customers across data-centre contracts; it does not disclose a Microsoft-specific customer credit or allocation formula.

A temporary pause ended; a large group of gateway updates restarted within a short period. At the same time, a separate change had made configuration requests slower. Microsoft’s incident review shows how the two conditions met in a shared regional control plane—and why a gateway should not count as healthy before its routes and configuration are back.

Microsoft plans multi-gigawatt Maia deployments from 2027 and says Maia 300 could raise tokens per megawatt by 40%. The commercial test is whether that engineering claim translates into useful, well-utilized capacity; the company has not disclosed Maia’s capital cost, fleet share or return.

A year-long software commitment can still be paid in smaller monthly amounts. From 1 October, Microsoft’s CSP software price list adds a 5% uplift when partners choose that payment cadence. The change reaches a different product scope from the earlier online-services pricing rule, while leaving the annual obligation intact.

Microsoft’s entry into two Nevada utility cases puts a tax-accounting rule beside a large-load investment decision. The proposed 10.7% and 10.6% generation figures apply to a defined construction contribution, not to a campus budget or household electricity bill.

The integration gives telecom operators a common data layer for network automation, reducing the need to build separate pipelines for each AI use case.

ICTA can appoint up to 28 providers across two lots, then compete individual backbone, metro and cross-border requirements through later call-off contracts.

New York recommends at least $1 million per MW for data-centre community investment, giving municipalities a common starting point for negotiations.

The Philippines plans to expand AI computing capacity to 1.5GW by 2033, with gas providing nearer-term power and renewables targeted for 40% of demand.

University of Utah researchers will test Creekstone's exhaust-water recovery process, with its projected 4,000 acre-feet annual output still unproven at scale.

Eligible Indigenous applicants now have first access to unused cellular spectrum, but licences still require a separate application.

Telstra is prioritising Microsoft Teams traffic on a dedicated 5G slice for business sites that need more predictable video-call performance.

The planned 200km cable will extend Telecom Egypt's Red Sea network to Taba, adding another route for traffic crossing Egypt between the Red Sea and Mediterranean.

At Metro Connect Fall, Andrew Lipman linked AI demand to practical limits in grid access, fibre deployment, permitting and project delivery.

Thai officials have asked developers of 49 data-centre projects under construction to pause voluntarily while new national rules are prepared.

Two Central US incidents disrupted Azure Databricks cluster starts, notebooks and jobs on 5 September, with both services later restored.

Germany plans a Cyberdome sensor network to detect cyberattacks, but has not set out deployment dates, funding or operator responsibilities.

A BGP disruption reduced visibility of Nigerian prefixes across 5–6 September, while the cause and affected operators remained unconfirmed.

Kentik’s network-intelligence platform depends on a proprietary analytical engine designed to retain large volumes of telemetry, add routing and business context, and return historical answers quickly enough to guide operations. Kentik Data Engine gives network teams a searchable memory, but it cannot improve evidence that was never collected: sampling, missing fields, stale metadata and incomplete routing context remain part of every result.

KT used a 5G SA slice to carry real-time CCTV video and emergency communications at Seoul's fireworks festival, extending an earlier large-event trial.

EXA and euNetworks have filed separate applications for new Irish Sea fibre cables, with both projects still at the regulatory stage.

The 264-acre site is planned for phased development, while first-phase capacity, power supply and construction timing remain undisclosed.

A Johnson County judge dismissed Edgerton’s case against residents, while their data-centre petition remains under separate court review.

China’s industrial 5G pilot lets approved enterprises build dedicated private networks around production needs.

KIDZ AI can take capacity in stages at an operational Phnom Penh data centre while pursuing a minority investment in HUADING Cambodia.

4iG will start with Hungary and Albania, then extend a later 2GHz 5G NTN service to Montenegro and North Macedonia.

An unnamed data-centre operator has paid to reserve manufacturing capacity for a planned 75MW FuelCell Energy project in Texas.

New federal powers could affect how US data-centre projects procure covered transformers, batteries, generators and control equipment before installation.

MasOrange has tested 6GHz spectrum with 256TR active antennas and beamforming as it evaluates radio technology for future 6G deployment.

A 41-deal count gives the US fibre consolidation debate a measurable base as investors weigh existing networks against further construction.

MSSA's Version 2.0 adds regenerative payload guidance and shared 5G NR and NB-IoT design options for non-terrestrial networks.

LS Electric will work with KT Cloud from the design stage on power systems for modular AI data centres, with equipment supply planned for future projects.

Sparkle and Hellas Sat have extended Sparkle's QSI service from a terrestrial data-centre link to a GEO satellite connection between Greece and Cyprus.

Nscale will use the facilities to fund GPU systems and supporting infrastructure at its Ward County and Madison AI sites.

The acquisition brings Dome's private 5G core and its developers into Orange Business for defence and critical-infrastructure deployments.

Boldyn is using 5G RedCap to connect a livestock sensor to its West Sussex private network, testing support for simpler, lower-power IoT devices.

Long transformer backlogs mean data-centre developers may need to secure electrical equipment years before a site is ready to energise.

Rystad says cooling choices could cut 2030 water demand sharply, but some lower-water systems require more electricity.

The reported deal depends on Hut 8 delivering Texas capacity before Lambda can turn Anthropic’s commitment into usable compute.

The systems are designed to smooth volatile AI loads before they reach grids or generators, with an 800VDC option planned for 2027.

A week-long outage linked to the wrong cooling-loop fill shows why coolant verification matters before direct-to-chip systems enter operation.

Compu Dynamics is expanding factory-built delivery for AI projects, shifting more power, cooling and white-space work away from the construction site.

Nexans has qualified a 525kV HVDC cable and repair joint for 3,000-metre depths on the planned Crete-Cyprus interconnector.

The new service gives subsea contractors temporary access to 4,000- and 6,000-tonne cable carousels, with a first European contract already secured.

Alberta’s NDP wants new AI data-centre approvals paused until the province sets broader rules for assessing project impacts.

Council has ordered a bylaw for 21 September that would pause applications above 50MW while the city writes rules for large data centres.

The Malaysian private company spans data-centre capacity, cloud, managed security, networks and IT outsourcing. Its partnerships are increasingly visible, but its ownership names, absolute financial scale and the economics behind some of its infrastructure commitments remain private.

Researchers linked an 18km free-space quantum channel to single-mode fibre, showing how adaptive optics can bridge atmospheric and terrestrial quantum links.

FreeCast is pitching its streaming platform to telecom and broadband providers, with Assist Wireless and enTouch Wireless already signed to use the service.

Students and teachers spent two weeks inside Mauritius Telecom, seeing how AI tools are used across the operator’s business.

Mavenir and Sanas will test real-time speech processing inside Mavenir’s mobile-core voice platform before making the functions commercially available to operators.

Chris Malone’s exit follows an earlier reorganisation that placed OpenAI’s data-centre engineering, compute and build-and-delivery work under separate leaders.

A replacement certificate changes tomorrow's build, but a valid timestamp can preserve yesterday's binary beyond the signer's expiry. The real cutover is therefore an inventory of artifacts, timestamp tokens, revocation time and verifier results—not a closed renewal ticket.

Itochu is reportedly planning around 10 50MW data centres in Japan by 2030, with leasing and eventual asset sales central to the model.

linuxptp turns a Linux host, hardware clock and network timing source into a precision-time system. Its daemons coordinate IEEE 1588 state, packet timestamps, servos and system clocks, but software alone cannot manufacture accuracy: the result still depends on NICs, oscillators, profiles, topology, calibration and the integrity of the reference source.

Colleferro has approved Cobra Green Hyperscale's planned AI data centre, but national clearance is still required before construction can begin.

Laurent Vanbever’s work treats network configuration as executable software whose failures can emerge before, during or after deployment. From safe routing migrations and configuration synthesis to runtime BGP detection and energy-aware operations, his research at ETH Zurich asks how operators can preserve intent while the network is continuously changing.

Texas is requiring more project information as ERCOT tries to assess a data-centre connection queue that has grown to more than 474GW.

Vapat has proposed a 160MW data-centre campus in Valladolid beside its wind assets and La Mudarra substation, with the project under environmental review.

OpenAI, Meta and Microsoft are putting community funding, construction jobs and greater disclosure into data-centre development as local opposition grows.

Katerina Argyraki’s research follows a problem that becomes harder as networks become more programmable: a packet-processing system may be fast and flexible, yet operators and users often have little evidence that it behaved correctly. Her work at EPFL connects software-router scale, code verification, performance contracts, packet receipts and external measurement into a layered approach to network accountability.

Starcloud will use the new funding to expand spacecraft manufacturing and Nvidia engineering work as it develops larger orbital computing systems.

A network login may be decided in a few packets, but the trust behind it can span certificates, directories, access devices, roaming partners and accounting systems. FreeRADIUS makes that policy inspectable and programmable, while leaving operators responsible for the legacy clients, dependencies and failure paths that turn a small authentication exchange into critical infrastructure.

Canberra will keep assessing new data-centre projects under existing rules while the federal government develops national standards for large facilities.

The Reeves County project would start with 1.5GW of behind-the-meter generation using pipeline natural gas and on-site ethane, with first power targeted for 2027.

Dave Maltz’s role at Microsoft is best understood through the breadth of the organisation he leads. Azure Networking spans customer-facing services, DNS, security, distributed control, host offload, switch software, data-centre fabrics, wide-area links and optical systems. That remit makes Maltz an accountable integrator across layers, not the sole inventor of the many systems beneath Azure.

The BIRD Internet Routing Daemon grew from a Czech university project into a control-plane system used in demanding routing environments, including internet exchanges where policy is concentrated and operational mistakes can spread quickly. Its history shows how open routing software can reduce dependence on proprietary control planes while shifting more responsibility for configuration, branch management, testing and incident response onto the operator.

Amphora Connect's 18km fibre route from Colchester to West Mersea is live, putting around 4,000 homes and businesses within reach.

TVA is making large data-centre projects commit to power capacity earlier, while offering interruptible service where permanent supply is not yet ready.

Colombia has cancelled a planned Amazonas and Putumayo fibre network after its estimated cost rose 67.1% and officials identified delivery risks.

China is placing more computing capacity in western regions where renewable power, land and cooler climates are easier to secure.

Kubernetes, Docker and managed cloud platforms make containers look like applications, pods and services. Beneath those interfaces, containerd moves images, prepares filesystems, preserves runtime state and supervises processes. Its value comes from being deliberately incomplete: a common, privileged middle layer that many products can share while leaving orchestration, networking, policy and most user experience elsewhere.

Latos's compact edge model now has to clear planning with its power, plant, noise controls and access built into the site design.

Hyperscale orders are pushing Generac to expand US generator production ahead of 2027 deliveries, while supplier capacity and tariff exposure remain unresolved.

The regulator found Synapse's proposed gas plant too close to homes, forcing a rethink of the power plan behind its 1GW Olds data centre.

Slurm allocates scarce compute across HPC and AI clusters, turning queue policy into a practical decision about who gets GPUs and when. NVIDIA’s 2025 acquisition of SchedMD gave the project more resources while making vendor-neutral stewardship a question that now has to be judged through releases, testing and contribution behaviour.

DNS can show that shop.example.co.uk sits beneath co.uk; it cannot tell a browser where independent registration begins. The Public Suffix List supplies that missing policy map. A volunteer-maintained text file now informs cookies, site grouping, certificates and service limits across products whose decisions the maintainers neither control nor guarantee.

Projects meeting the new NSW guidelines can seek a 75-day state assessment while committing to additional renewables, storage and flexible electricity demand.

DDC1 pairs a new Danish colocation facility with grid-responsive power management, testing whether flexible electricity use can work alongside normal data-centre operations.

Insurance limits cover only part of Sopaipilla's potential losses, leaving lenders to rely more heavily on Meta's lease and contractual support.

The trial sends GNSS correction data through 5G core and radio networks, putting precise outdoor positioning onto standard mobile-network infrastructure.

Sage Weil’s most durable contribution to internet infrastructure was not a storage appliance but an architecture. Ceph used CRUSH to calculate where data belongs and RADOS to spread storage, repair and movement across a cluster. The design helped commodity machines behave like one storage system—and created an operational project that ultimately had to outgrow its founder.

Satellite lifts the FCC's reported 100/20 broadband availability to 99.7%, but terrestrial coverage remains lower and satellite performance can vary.

UK Space Agency funding covers optical links, ground stations, onboard processing, rail connectivity, and terrestrial-network integration across 16 satellite communications projects.

Prince George will add a second Northern Health fibre link before hospital construction severs an existing route, using revenue from its municipal network.

Prince George will add a second Northern Health fibre link before hospital construction severs an existing route, using revenue from its municipal network.

Ontario's draft framework would assess large data-centre projects on data sovereignty, economic contribution, and community benefit alongside grid requirements.

Alberta is opening public discussions on AI data centres as communities weigh power, water and infrastructure requirements alongside new investment.

Telecom Plus is continuing its £40m buyback as investment rises, making capital allocation the key issue in its next growth phase.

The Space Force will test cross-network links and orbital routers as it tries to connect commercial satellite systems through standard interfaces.

BUZZ will deploy 2,016 GPUs at Bell's Merritt facility under a five-year customer contract, with service targeted to begin in Q4 2026.

Katalyst will take DIU's deorbit-as-a-service programme through preliminary design, with any flight phase using NEXUS still conditional.

An AI server may bring together processors, accelerators, network adapters, storage controllers, switches, retimers and security devices from companies that compete everywhere else. PCI-SIG maintains the shared interface that lets those parts find one another and exchange data, then runs the compliance system used to test whether implementations follow the same rules. Its influence is substantial, but bounded: the organisation governs a common path into the machine, not the complete machine or every outcome once the parts are assembled.

Steven M. Bellovin began by testing the claims computers accepted from packets: source addresses, routes, sequence numbers and names. His later work applied the same suspicion to institutions, asking what systems make the public trust when governments demand access, companies infer identity and regulators turn technical possibility into policy.

Modern servers contain several chips that boot code, hold keys and enforce policy before the host operating system is awake. Caliptra gives those components a common open root of trust, but factories, provisioning systems, physical implementations and fleet verifiers still decide whether its public design becomes dependable infrastructure.

Nebius is adding another UK location for its AI cloud infrastructure through leased data-centre capacity at Vantage's Newport campus in South Wales.

Distributed systems fail when software assumes machines are interchangeable. Ion Stoica’s work repeatedly moves complexity behind a narrower abstraction—from packet-carried state and Chord to Mesos, Spark, Ray and cross-cloud placement—then tests how much operational, economic and governance burden that abstraction can actually hide.

The Domain Name System answers current questions and preserves little of its own history. OpenINTEL has built a longitudinal record by repeatedly measuring a defined set of domains and address space, creating unusual research value while leaving user demand, resolver behaviour and unmeasured names outside the archive.

Richard Barnes has worked on the protocols that make trust routine enough to disappear into operations: automated certificates, recipient encryption, changing group membership and protected media. Their value lies in open, composable mechanisms; their limits appear when services, endpoints and institutions decide how those mechanisms are deployed.

RWE is moving two data centre sites towards customer agreements as it tests how existing power infrastructure can support new European capacity.

A 10-year equipment recovery programme with TXO shows how refurbished network hardware can reduce waste and the need for newly manufactured replacements.

Microsoft’s account of QNBQ-5W8 shows how a temporarily disabled safeguard let bad control-plane metadata reach edge resources—and why recovery promises need operational proof.

Optus has sent the hardware for extending D3's life into orbit, but the propulsion pod still has to reach and attach to the satellite.

The review will assess economic benefits alongside power, water and environmental impacts without changing the local planning approval process.

Oakville has paused new data-centre development for one year while it studies planning rules, with no application filed for the proposed 320MW project.

The Galaxy Campus is designed to scale beyond 2GW, but Envision has not disclosed how much IT capacity is currently operating.

Sage Weil’s most enduring contribution was not a storage device but an architecture. CRUSH calculates data placement, while RADOS distributes storage, repair and movement across the cluster. Ceph enabled ordinary machines to operate as one system, then had to learn to continue without its founder.

UCIe has defined a die-to-die interface that reaches 64 GT/s and now covers 3D packaging, raw data traffic and package-wide management. The deciding question is whether these rules can support multi-vendor products when power delivery, thermals, manufacturing yield, software, security and liability still have to be resolved outside the standard.

UCIe has defined a die-to-die interface reaching 64 GT/s and expanded its scope to 3D packaging, raw data traffic and whole-package management. The decisive test is whether these rules can support multi-vendor products while power, thermal design, manufacturing yield, software, security and responsibility remain matters resolved outside the standard.

Microsoft Threat Intelligence’s new analysis of DeadLock focuses on an extortion architecture that separates encryption from the channels used to communicate and deliver recovery resources. Session and blockchain-backed services may let the operation recover from the loss of one site, forcing defenders to map a system rather than chase a single domain.

Nuclea Energy and New Mining will study behind-the-meter microreactors for an existing 30MWe load, without committing either side to a power project.

Hyderabad gives Azure customers another Indian region for running workloads and disaster recovery, although Microsoft says more services will be added over time.

Microsoft has opened its largest data-centre hub in India, but the live Azure service table still labels India South Central as available only to early-access customers. The distinction matters: an operational region with selected users is a real infrastructure milestone, yet it is not general availability across Azure’s customer base or product catalogue.

Prosimo raised at least US$55 million to coordinate application, network, and security policies across multiple clouds. By early 2025, public records indicated the company was no longer operating independently; a cofounder later stated that the technology had been integrated into Palo Alto Networks products. It remains unclear who controls the topology, cloud credentials, and production routes, or whether customers can transfer this operational model to another platform.

Prosimo raised at least USD 55 million to co‑ordinate application, network and security policies across multiple clouds. By early 2025 its independent activity had disappeared from public records, and its co‑founder stated that the technology had been integrated into Palo Alto Networks products. The remaining questions are who controls the topology, cloud credentials and production paths, and whether customers can migrate that operating model to a different environment.

Prosimo raised at least $55 million to coordinate application, network and security policies across multiple clouds. By early 2025, public records indicated the company was no longer operating independently; a co-founder later stated the technology had been integrated into Palo Alto Networks products. It remains unclear who now controls topology, cloud credentials and production routes, and whether customers can transfer this operating model to another provider.

Prosimo raised at least $55 million to orchestrate application, network, and security policies across multiple clouds. By early 2025 public records no longer showed it as an independent entity, and a co-founder said its technology had been integrated into Palo Alto Networks; the question remains who controls the topology, cloud credentials, and production paths, and whether customers can migrate this operational model.

The Ultra Ethernet Consortium aims to expand the most familiar network in the data centre into a complete fabric for artificial intelligence and high performance computing — from libfabric APIs and RDMA semantics through packet spraying, congestion control and link recovery to security and 200 Gb/s lanes. The specifications are public; the harder test is whether patents, products and multi-vendor implementations converge into reliable, running infrastructure.

The Ultra Ethernet Consortium aims to transform the most common data centre network into an integrated connectivity fabric for AI and high-performance computing, from libfabric interfaces and RDMA semantics to packet spraying, congestion control, link recovery, security, and 200 Gbps per-lane signalling. The specification is now public; the tougher test is whether patents, products, and multi-vendor implementations will converge into dependable operational infrastructure.

The Ultra Ethernet Consortium seeks to transform the data centre's most familiar network into a complete infrastructure for artificial intelligence and high-performance computing, from libfabric APIs and RDMA semantics to packet spraying, congestion control, link-level recovery, security and 200 Gbps lanes. The specifications are now public; the real test is whether patents, products and multi-vendor implementations will converge into reliable infrastructure in operation.

Linux Traffic Control began as a queueing framework and now carries classification, redirection and hardware offload. Salim’s work across `tc`, Netlink, ForCES and P4TC shows how hard it is to add programmability without breaking the interfaces operators already trust.

Velox has started construction, but the first 2MW phase and 1,800 planned RNGD accelerators are not expected to enter service until early 2027.

Microsoft, Meta, Oracle, Amazon and Alphabet have signed leases for facilities that are not yet available for use, mostly data centres.

The 2025 ToolShell exploitation record shows that keeping collaboration infrastructure on premises also means owning inventory accuracy, supported-version planning, emergency isolation, patch deployment, key rotation, compromise assessment, and executive escalation.

Microsoft’s review focuses on a harder task than annual procurement: matching zero-carbon power to data-centre demand on the same grid, hour by hour.

Microsoft adopts an optical connector designed to simplify fiber installation and maintenance in dense data center environments.

Microsoft and Chevron sign a 20-year power deal for Project Kilby, linking West Texas generation with AI data centre capacity.

Microsoft is developing MicroLED cabling and Hollow Core Fibre to cut AI data centre power use, distance limits and latency.

Microsoft is a major cloud and AI infrastructure provider whose quantum hardware progress could shape future enterprise computing, cybersecurity and scientific workloads.

UK court allows class action alleging Microsoft cloud licensing inflates rival platform costs, raising concerns over cloud market competition.

The Microsoft-NVIDIA-Anthropic partnership matters because it ties model access, cloud capacity, accelerator architecture and capital into one public announcement. Anthropic committed to purchase $30 billion of Microsoft Azure compute capacity and to contract additional capacity up to one gigawatt, while NVIDIA and Microsoft committed to invest up to $10 billion and $5 billion respectively in Anthropic. The useful signal is not a generic AI alliance; it is a three-sided capacity bargain in which Claude gains Azure distribution, Microsoft gains a major non-OpenAI frontier model in its enterprise stack, and NVIDIA gains a direct architecture-optimization channel with one of the most compute-hungry model labs.

What happened: Microsoft and ValueLicensing prepare for high-stakes court showdown Microsoft is set to face UK-based ValueLicensing in court over a dispute regarding the resale of second-hand software licenses. ValueLicensing accuses Microsoft of unfairly restricting the resale of perpetual software…

Nokia's November 2024 expansion with Microsoft matters because it puts a named network supplier inside Azure's datacenter growth path for another five years. Nokia will supply datacenter routers and switches, including the 7250 IXR-10e platform and custom top-of-rack management switches, while Microsoft continues to scale Azure capacity for compute demand. The useful signal is not simply a vendor renewal; it is the hardware and SONiC operating layer behind Azure's move from 100GE toward 400GE capacity across new and existing sites.

The lawsuit, filed by a small technology company, alleged that Microsoft’s Cortana infringed upon patents held by the plaintiff.

Microsoft's February 2024 partnership with Mistral AI is best read as a cloud-distribution and model-choice signal, not as a simple investment headline. Microsoft put Mistral's premium models into the Azure AI route to enterprise customers while Mistral gained access to Azure AI infrastructure and a global sales channel. The strategic question is whether a European frontier-model company can scale through a hyperscale cloud partner without losing the independence that made it valuable to customers and regulators.

Microsoft spent a small fortune on a pumpkin farm, preparing to build a massive data center in Mount Pleasant, Wisconsin.

Water rights are becoming a deal-level constraint as data centre projects face tighter permits and community opposition.

Dell’Oro sees 2026 data centre capex topping $1tn as AI buildout accelerates, while Gartner warns power limits are constraining capacity.

Jefferies' capacity analysis gives BTW readers a quantified signal on the gap between AI infrastructure demand and deployable data centre supply.

Morgan Stanley warns AI-driven memory chip prices are spreading cost pressure from data centres to PCs and smartphones.

SOTI releases ONE 2026.1 enhancing shared device security, automation, and frontline operations globally.

Nvidia brings RTX Spark to PCs, pushing local AI agents into laptops and desktops with MediaTek and Microsoft.

NPO matters because a configuration error, security incident, backup failure, or registry mistake in its managed-services or LIR roles could disrupt enterprise customers that depend on its cloud, network, and recovery infrastructure. The February 2026 private-equity acquisition introduces a new governance layer whose strategic decisions may alter the company’s operational risk profile.

Changes in AS210852’s BGP announcements, upstream policy, or PeeringDB records can directly affect the reachability of managed IT, voice, and payment services for the region’s SMEs. A routing leak or facility outage could cascade across point-of-sale terminals, cloud workloads, and office telephony, making public monitoring a practical tool for anticipating regional internet disruptions.

Registry changes to GR11509-RIPE can affect external observability of Lyon’s network boundary. Commune de Lyon’s open-source office-suite migration represents a high-stakes public-sector digital-sovereignty experiment that influences data control, supplier dependence, and service access for over 525,000 residents.

How IPv4 scarcity and secondary markets are turning IP addresses into recognised digital capital in the modern internet economy.

AI demand pushed Q1 cloud spend to $128.6bn with 35% growth, led by AWS, Microsoft and Google Cloud dominance.

EU expands Digital Markets Act to cloud and AI, targeting Big Tech competition in infrastructure and AI services.

BT launched the UK’s first sovereign portfolio combining cloud, AI, connectivity and voice for regulated sectors.

SAP reported stronger-than-expected first-quarter profit growth, driven by rising cloud demand and improving revenue visibility from long-term enterprise contracts.

Microsoft says it brought 31 new data centres online across five continents in its final fiscal quarter and 88 across the year. The larger revelation is not the pin count. In the same quarter it added another gigawatt of capacity, while cutting the time needed to move new GPUs from delivery dock to live service by nearly half over the year. Those are separate measurements of an AI infrastructure system: places, power, equipment velocity and capital. Reading them together shows why the bottleneck has shifted from announcing buildings to repeatedly turning expensive components into sellable cloud capacity.

Microsoft has disclosed $329.1 billion of additional leases, primarily for datacentres, that had not commenced at 30 June. The number maps a future infrastructure commitment, not live computing capacity or cash already spent. A simultaneous accounting change makes the distinction sharper: reported 2026 capex falls from roughly $190 billion to roughly $175 billion even though management says its underlying investment expectation has not.

Britain’s competition regulator is investigating whether Microsoft 365 Personal and Family subscribers saw a cheaper continuity option clearly enough before renewal. The annual difference was £25, but the economic question is larger than the price: when a supplier adds Copilot, changes the default plan and leaves the customer to find an alternative, the renewal journey can determine which product is bought.

OpenAI and Oracle did not cancel their operating Abilene campus. They dropped a separate 600 MW expansion option; Microsoft later became the customer for two adjacent buildings.