Microsoft Azure AI News, October 2026: Maia 300 Launch, $115.95B Capex, and UK Data Center Exit

Microsoft's FY2026 capital expenditure reached $115.95 billion — up 79.6% year-on-year — as the company launched its Maia 300 AI accelerator targeting multi-gigawatt deployment in 2027, while withdrawing from a UK data center deal with Nscale over grid-connection delays extending to the 2030s.

Microsoft's AI infrastructure push entered a distinctly new phase this month, defined by a milestone chip announcement, a record capital-expenditure trajectory, and an increasingly uneven geographic buildout. The company unveiled the Maia 300, its third-generation custom AI accelerator, alongside plans for a multi-gigawatt deployment beginning in 2027. According to multiple reports, the 96-chip rack configuration promises seven times the token throughput of its predecessor and a 40% improvement in token output per megawatt — advances Microsoft says will allow Azure to deliver more AI compute for each unit of electrical power consumed. That silicon ambition is backed by a financing commitment of extraordinary scale: Slicast's own compilation of SEC XBRL filings shows Microsoft's capital expenditure for fiscal year 2026 reached $115.95 billion, up 79.6% from $64.55 billion in FY2025, representing 35% of FY2026 revenue of $331.84 billion. By capex intensity, that places Microsoft second among the seven hyperscalers Slicast tracks, behind only Oracle; by absolute spending, second behind Amazon. The largest single quarter on record — $19.39 billion for the period ending September 30, 2025 — illustrates the pace at which those annual totals are being assembled.

The month's most widely covered single event was Microsoft's reported withdrawal from a data center project in Loughton, Essex, operated by Nscale. Multiple outlets, citing The Daily Telegraph, described the project as involving roughly 23,000 Nvidia chips with Microsoft serving as the anchor tenant. The reason cited consistently across sources is power: UK grid connections to the site are reportedly delayed until the 2030s, a timeline Microsoft chose not to absorb. Multiple reports — including from Computing UK and Reuters — linked the withdrawal to pressure on Nscale's planned IPO; Computing UK put that IPO at up to $35 billion. The episode is a pointed illustration of a structural constraint tightening across European AI infrastructure — demand is outrunning grid capacity, and at least one hyperscaler has demonstrated it will redeploy capital rather than wait. A counterpoint worth noting is that Nscale simultaneously announced an expanded deal with Microsoft covering capacity in Norway, suggesting the commercial relationship has been geographically repriced rather than terminated.

On the US side, Microsoft's land and power acquisition activity continued accelerating on several fronts. The company reportedly concluded a 1.1-gigawatt power agreement with utility NiSource, filed for a three-building data center campus in Floyd County, Georgia, and acquired 600 acres of land in Woodward, Iowa. Read alongside the Maia 300 announcement, these moves sketch a coherent strategic logic: secure scale power in electricity-friendly US markets while deploying a proprietary chip architecture designed to extract more inference per megawatt — a combination aimed at containing per-unit AI compute costs as aggregate spending escalates. TechTarget reported that Microsoft is explicitly positioning the Maia chips as a mechanism to stretch AI data center power budgets, and The Next Platform highlighted the chip's bandwidth characteristics as a key differentiator for inference workloads. Separately, reports indicate the Maia 200 is already in production powering Copilot workloads, providing a commercial baseline against which Maia 300 targets can be measured.

Two complicating signals deserve equal analytical weight. First, OpenAI — long positioned as a Microsoft-Azure-exclusive cloud customer — reportedly began making its models available on Google Cloud, which, if confirmed, would represent a meaningful erosion of one of Azure's most consequential competitive advantages. The report comes from a single outlet and has not been corroborated by multiple independent sources, so it should be treated with caution; the Microsoft-OpenAI partnership remains deep by any measure. Still, the development introduces structural ambiguity around exclusivity that analysts are likely to press on when Microsoft reports earnings on October 28. Second, reports circulated that Microsoft's cloud leadership had called for cost reductions of at least 20%; the company publicly denied those claims. Neither assertion is supported by SEC filings or multiple corroborating sources, so both warrant skepticism — but the denial itself confirms the topic is live at senior levels. EnergyNow.com separately reported that Microsoft may find its clean-energy-powered data center commitments difficult to sustain at the current pace of buildout, a reputational and regulatory risk that sits alongside the financial ones.

Heading into the October 28 earnings call, the picture is of a company allocating capital at a historically unusual rate — $115.95 billion in FY2026 against a $331.84 billion revenue base — while navigating a set of constraints that are shifting in character. The binding limit is no longer access to capital; it is increasingly access to power and, potentially, to the exclusive partner relationships that differentiate Azure's AI proposition. The infrastructure thesis is coherent and well-funded, but execution risk has migrated. Three concrete signals are worth watching: the capex guidance Microsoft provides on October 28 relative to the $115.95 billion FY2026 baseline; the pace of Maia 300 deployment confirmations against the multi-gigawatt 2027 target; and any further clarity on the terms of OpenAI's reported Google Cloud arrangement and what those terms imply for Azure's share of the most strategically significant AI workloads.