Mega-cap Hyperscalers

Alphabet reportedly won two separate US antitrust cases in late September, easing immediate legal pressure on Google. Google is helping form a cross industry coalition focused on AI data center development ahead of the 2026 US midterm elections, expected to address questions around power use, data sourcing and community impact from large scale AI computing hubs. The antitrust decisions support the view that Alphabet can keep rolling out AI powered features across Search, YouTube and Google Cloud without immediate structural remedies that break its distribution. The AI data center coalition cuts both ways, aligning with the potential to monetise AI infrastructure and long duration power contracts while spotlighting the risk that regulators and communities could still cap how far Alphabet can push energy hungry AI workloads. Analysts have already flagged regulatory pressure and high capital intensity as two of the biggest swing factors for the company.

Anthropic's draft IPO prospectus shows roughly $110 billion owed to Amazon Web Services over about the next decade, part of $518 billion in total cloud, computing and infrastructure obligations the AI company has planned. Amazon's share traces back to a pledge of more than $100 billion over 10 years and, like most of that total, is owed even if Anthropic's usage falls short, giving Amazon revenue visibility that ordinary pay-as-you-go cloud usage does not. Spread evenly, the AWS portion runs above $10 billion a year, while Anthropic reportedly spent $7.33 billion on all of its suppliers combined in 2025. Amazon has also invested $18 billion in Anthropic through convertible notes and nonvoting preferred stock, with up to $15 billion more available, and those holdings are now carried at about $190 billion, up from about $61 billion at the end of 2025. Second-quarter net income more than tripled to $62.6 billion, with $53.4 billion of pre-tax, non-operating income coming mostly from the Anthropic investments. Separately, Synopsys and AWS announced a multi-year deal worth more than $1 billion under which Amazon's cloud unit will license chip design blueprints, and AWS now offers Claude Sonnet 5.5 through Amazon Bedrock.

Microsoft and other large US tech firms have formed a new coalition focused on the environmental, energy grid, and local community impacts of AI data center buildouts across the United States. Member companies are preparing shared sustainability and job creation commitments ahead of key US political events. The group is meant to address local pushback on power use, water consumption, and grid stress that has become a real friction point for large AI facilities, and Microsoft is using it to protect its license to keep building the AI infrastructure its cloud and Copilot products rely on. Microsoft, a US software heavyweight with a US$3.8 trillion market cap, faces a key checkpoint in how it breaks out AI driven data center costs and Azure usage under its new FY27 reporting structure, with the first segment numbers due under Agents and Infra. The coalition lines up with an existing risk around heavy AI and cloud investment that raises financial and execution pressure, and coordinated work on permitting, grid access, and environmental rules may help manage that capital intensity without removing the exposure to very large, long dated infrastructure bets.

Barclays said fresh data shared by Nvidia points to at least $30 billion of potential upside to the chipmaker's revenue from the top five cloud providers in both 2026 and 2027. Analyst Tom O'Malley noted that Nvidia disclosed its share of IT capital spending across the five largest cloud service providers in recent years, through the second quarter of 2026, along with an estimate of 44% for 2027. Using Barclays' internet team's assumption that cloud capital spending reaches about $1.1 trillion by 2027, with 80% going to IT, O'Malley calculated hyperscaler revenue of $237 billion in 2026 and $401 billion in 2027, above Barclays' current forecasts of $206 billion and $370 billion. In an upside scenario where historical ratios hold, he sees revenue reaching $246 billion and $417 billion in those years. O'Malley also noted that Nvidia's hyperscaler revenue includes SpaceX while the top five cloud provider figure does not, implying an even larger total, and said sentiment on the name can continue to improve amid solid numbers upside, increased buybacks and the Hugging Face acquisition.

Sen. Elizabeth Warren and fellow lawmakers sent letters this week to the CEOs of Amazon, Alphabet, Meta and Microsoft demanding details on their tax breaks and their lobbying for incentives in the One Big Beautiful Bill Act. Warren pointed to Amazon, which paid $7.8 billion less in tax last year, while Meta received nearly $7 billion in tax relief in 2025, paying $2.8 billion in federal tax compared to $9.6 billion in 2024. According to the Institute on Taxation and Economic Policy, Microsoft and Alphabet each qualified for tax cuts of nearly $19 billion last year. The breaks stem largely from the One Big Beautiful Bill Act's 100% bonus depreciation, which lets companies write off qualifying capital investments such as AI data centers in their first year; Microsoft claimed nearly $12 billion in depreciation tax breaks, Amazon $6.5 billion, Meta $4.9 billion and Alphabet over $3.3 billion. Corporate tax revenues are down 23% this year, amounting to $96 billion in lost federal revenue, according to Politico, as federal debt has topped $40 trillion. Warren and her colleagues want data centers disqualified from bonus depreciation tax breaks, while White House spokesperson Kush Desai said the law's pro-growth provisions were driving historic job, investment and wage growth.

NTT DOCOMO BUSINESS, Inc., formerly NTT Communications Corporation, has migrated its shared internal Oracle-based database platform to Oracle Autonomous AI Database on Oracle Cloud Infrastructure, moving tens of terabytes of data off its on-premises environment and cutting database license costs by 43 percent. The shared platform hosts critical internal systems including network management, security, and service operations, and currently supports 12 critical internal systems connected to approximately 400 servers. Akihiro Ito, senior vice president of the Integrated Service Operations Division at NTT DOCOMO BUSINESS, said the flexible consumption model delivered the 43 percent license-cost reduction versus on-premises, while automation of patching, backups, and routine maintenance has significantly reduced day-to-day administrative work. To strengthen business continuity, the company implemented a disaster recovery architecture across two Oracle Cloud regions in Japan, which it says supports stable operations and faster recovery from disruptions. The company added that automated backups, patching, and maintenance have cut the time and effort needed to address security vulnerabilities and enabled faster responses to AI-enabled cyberattacks and emerging vulnerabilities.

Anthropic's confidential initial public offering prospectus has revealed the AI lab's heavy dependence on a small group of big tech partners and clients, according to Reuters. The hyperscalers Amazon and Google benefited as Anthropic routed 47% of its sales to customers in 2025 through those two cloud partners, with sales through the pair accounting for $2.16B. Anthropic paid about $351M, or about 16 cents per dollar, in distribution fees back to those cloud partners in 2025. Sales through the two partners have surged every year, rising from 11% of revenue in 2023 to 32% in 2024 to nearly 50% in 2025, while both Amazon and Google have heavily invested in Anthropic in return. In addition, two unnamed customers accounted for 12% of Anthropic's total revenue in 2025.

Jefferies analyst Brent Thill reiterated a Buy rating and $575 price target on Microsoft, calling the company's revamped Copilot incrementally positive and supportive of the firm's Microsoft 365 thesis. The biggest change, according to Jefferies, is that Microsoft is pushing Copilot beyond simple AI assistance toward software that can perform longer, more complex work, with new features such as Cowork, Code and Autopilot potentially increasing both the number of tasks Copilot handles and how frequently customers pay for AI usage. Thill said Cowork bridges AI assistance to delegated, end-to-end work, while Autopilot extends this into proactive, persistent and recurring execution, a shift Jefferies believes could make AI consumption more tangible across Microsoft 365. Thill also compared the new Copilot app with Meta Platforms' Muse, arguing Microsoft may hold an important enterprise advantage because its AI tools are grounded in proprietary corporate data and workflows, and because it controls a much broader enterprise technology stack spanning productivity, collaboration, cloud infrastructure, data, identity, security, governance and compliance. Jefferies cautioned that better AI capabilities do not automatically guarantee stronger consumption, as enterprises closely monitor AI budgets and increasingly demand clear returns before expanding spending, meaning adoption could remain gradual; the key metric for investors will be whether these new agentic capabilities move from demonstrations into recurring production workflows.

NetApp and Oracle announced a new fully managed storage service that brings enterprise-grade NetApp storage natively to Oracle Cloud Infrastructure. The service, called Oracle Cloud Infrastructure NetApp Storage Service, will bring NetApp ONTAP data management capabilities natively to OCI to help organizations simplify migration and management of critical AI and enterprise workloads in the cloud. It is designed to help enterprises migrate, run, protect, and modernize databases, enterprise applications, virtualized environments, EDA/HPC, regulated applications, and AI data pipelines while preserving familiar storage operations, and customers can manage it through the OCI Console and SDKs, ONTAP APIs, and familiar operational workflows. Pravjit Tiwana, Senior Vice President and General Manager of Cloud Storage and Services at NetApp, said customers want the flexibility to move AI and enterprise workloads to the cloud without giving up operational simplicity, data management, and reliability, while Konstantinos Papamiltiadis, Senior Vice President of Partnerships at Oracle Cloud Infrastructure, said enterprises shouldn't have to rebuild their data infrastructure to move to the cloud. Alvaro Celis, Senior Vice President, Chief Commercial, Partner and Ecosystem Officer at NetApp, said the partnership combines OCI's scale and performance with NetApp's enterprise data capabilities to help customers run mission-critical workloads, especially in regulated industries. OCI NetApp Storage Service is planned for general availability within the next 12 months.

Microsoft launched a reimagined Copilot in September 2026 built around Home, Code and Autopilot, alongside a hybrid pricing model that keeps per-user pricing for everyday AI while moving advanced capabilities to consumption-based Copilot Credits. The company reported that Microsoft 365 Copilot paid seats topped 30 million in fourth-quarter fiscal 2026, with net seat additions more than doubling sequentially, while Agent 365 registered nearly 40 million agents within two months of launch and Foundry revenues more than doubled year over year. Commercial remaining performance obligation surged 84% year over year to $678 billion. For first-quarter fiscal 2027, Microsoft expects revenues of $89.85-$90.95 billion, implying 16-17% growth, with Microsoft 365 Commercial cloud revenues projected to grow about 16% in constant currency, Azure expected to grow roughly 45%, and Intelligent Cloud revenues forecast at $40.95-$41.25 billion, up 33-34%. Cost of revenues is expected to rise 23-24% in the fiscal first quarter, keeping operating margins roughly flat year over year, while capital expenditures are projected to exceed $50 billion, and for fiscal 2027 Microsoft expects operating margins to decline by less than one point as capex keeps rising. Microsoft faces competition from Salesforce, whose Agentforce annual recurring revenue topped $1.5 billion in second-quarter fiscal 2027, up more than 240% year over year, and from Alphabet, whose Gemini Enterprise is now used by nearly 90% of Fortune 100 companies and whose Google Cloud backlog reached $514 billion in the second quarter of 2026.

ClickHouse announced a significant expansion of its strategic collaboration with Microsoft at The European Microsoft Fabric + SQL Community Conference in Barcelona, Spain, encompassing four major milestones. The company launched a native ClickHouse workload for Microsoft Fabric, now in public preview through the Fabric workload hub, which brings the ClickHouse query engine directly into the Fabric experience and accelerates OneLake queries to sub-second response times. ClickHouse support for reading Apache Iceberg tables from OneLake using OneLake Table APIs is now generally available, while support for writing Iceberg tables to OneLake with credential vending is in public preview. ClickHouse Bring Your Own Cloud is now available to Azure customers through the Microsoft Marketplace, letting organizations run ClickHouse Cloud inside their own Azure tenant while ClickHouse manages provisioning, upgrades, monitoring, and scaling. Yury Izrailevsky, President of Product and Engineering at ClickHouse, said the engineering teams worked closely together so these capabilities are seamlessly integrated into the Microsoft data platform rather than layered on top of it, and Dipti Borkar, Vice President of Microsoft IQ and OneLake at Microsoft, said joint customers have more best-in-class analytical engines to choose from without moving or copying data between platforms.

Anthropic reported a net loss of $42 billion in 2025 and plans to spend $518 billion on cloud, computing and infrastructure obligations in coming years, according to its IPO prospectus seen by Reuters. Revenue grew 12-fold in 2025 to nearly $4.6 billion, even as the company lost more than $8 billion on an operating basis, excluding writedowns of various liabilities mostly tied to previous fundraising. The AI lab spent $7.33 billion on compute and infrastructure last year, a threefold surge from 2024, accounting for more than half of its $12.65 billion in total operating expenses. Anthropic had cash, cash equivalents and short-term investments totaling $20.28 billion as of December 31. The debut would follow SpaceX's IPO that valued Elon Musk's company at $1.77 trillion, while Anthropic's biggest challenger OpenAI confidentially filed for an IPO in June and is expected to list by early 2027.

Alibaba Group Holding is expected to secure Chinese approval to purchase Nvidia RTX PRO 5500 chips under current export rules. Regulators in Beijing are reviewing the RTX PRO 5500 orders as a category of AI hardware seen as compliant with US curbs. The prospective approval would expand Alibaba's pool of Nvidia processors for use in its cloud services and in-house AI projects, supporting the company's RMB 380 billion AI and cloud buildout over three years alongside its Zhenwu V900 chips. The likely green light comes as Alibaba competes with Tencent and Huawei to serve enterprise AI workloads at scale, while analysts warn that relying on both in-house accelerators and imported Nvidia gear could keep capital intensity high and leave profitability under strain if usage lags capacity.

Crusoe announced it is the developer of Google's data center campus under construction in Armstrong County outside of Amarillo, Texas. The campus is engineered to provide capacity to power advanced AI and has a direct connection to Serena's adjacent Goodnight 1 wind farm, which at 265.5 MW is operational, and the Goodnight 2 wind farm, also 265.5 MW, which is under construction and was catalyzed by Crusoe's development at the site. When wind generation exceeds campus demand, surplus power is supplied back to the grid, benefiting local ratepayers and reducing curtailment of West Texas wind energy. Construction broke ground in June 2025 and is underway with more than 5,500 skilled laborers on-site daily, and the project will use a closed-loop, non-evaporative liquid cooling system designed to limit its water footprint primarily to domestic uses such as kitchens. The campus is part of Crusoe's growing West Texas portfolio, which includes two campuses under development in Abilene.

Nokia Oyj is extending its partnership with Microsoft Corporation to build an agentic, unified data foundation that combines Nokia Data Suite with Microsoft Fabric to enable AI-driven telecommunications network operations. The collaboration integrates network data products with Microsoft's cloud analytics and AI capabilities, allowing operators to access structured network data in minutes rather than weeks, and the solution is currently available, building on existing agreements across cloud infrastructure, security, and AI workloads. For Nokia, the integration accelerates monetization within its AI & Cloud segment, where order intake reached €2.8 billion in Q2 2026 and net sales to AI & Cloud customers surged 105% year-over-year, supported by €2.78 billion in net cash. For Microsoft, the move strengthens Azure's enterprise ecosystem, supporting Azure's 43% revenue growth in Q4 FY26 and expanding Microsoft Cloud revenue beyond its $59.3 billion quarterly base. Nokia's reported operating margin fell to (1.0)% in Q2 2026 on accelerated restructuring expenses, while Microsoft's FY26 total revenue reached $331.8 billion amid heavy AI infrastructure spending.

Nokia Oyj shares rose 3% to $10.98 on September 9 after Google announced a €13 billion investment to expand its AI infrastructure and cloud footprint in Hamina, Finland. Google said its Hamina site, part of a 43-region network, lets local entities like Nokia deliver high-performance, low-latency services and scale AI solutions, though the announcement is not a direct contract win for Nokia. In Q2 2026, Nokia recorded a 105% year-over-year surge in net sales to AI and Cloud customers, supported by €2.8 billion in AI and Cloud order intake, while its Network Infrastructure division grew 12%, led by a 20% jump in Optical Networks and a 16% rise in IP Networks. Nokia closed Q2 2026 with €2.8 billion in net cash and interest-bearing financial investments, but the quarter also brought a reported operating loss of €50 million, a negative 1.0% reported operating margin, and a 95% year-over-year decline in reported profit to €5 million. Management said supply remains the main industry constraint relative to strong demand, and Nokia must manage negative short-term free cash flow, restructuring costs, and optical supply limits to convert hyperscaler demand into sustainable earnings growth.

Nvidia disclosed that Anthropic now has more than $180 billion of contracted value across cloud providers and neocloud operators, a figure the chipmaker revealed during its latest Non-Deal Roadshow. The Dario Amodei-led AI developer has also contracted for 2.5 gigawatts of Nvidia AI infrastructure scheduled for delivery through 2028. Nvidia said its revenue opportunity per gigawatt, based on its reference designs, has risen from roughly $18 billion during the Hopper generation to $25 billion with Blackwell and $40 billion with Rubin. The company added that its share of information-technology capital spending among the five largest cloud service providers has climbed from 11% in the first quarter of 2023 and is projected to reach 44% by 2027, while Australian partners including IREN and Firmus expect to bring roughly 2 gigawatts of capacity online by 2027. Nvidia also expanded its share-repurchase authorization by $150 billion to $235 billion.

Meta Platforms is planning a C$13 billion data center in Alberta, its first in Canada, a 1-gigawatt facility that could eventually scale to 1.8 gigawatts to support its growing AI operations. Capital Power will initially supply 250 megawatts, while Meta has a long-term agreement with Pembina Pipeline's planned natural gas-fired Greenlight Electricity Centre, expected to enter service in late 2030. Capital Power CEO Avik Dey told Reuters that several other prospective data-center developers are discussing electricity supply with the company, and he expects more U.S. hyperscalers to eventually build large facilities in Alberta, where more than 100 data-center projects have been proposed. Meta's second-quarter revenue rose 28% year over year to $60.8 billion, while advertising revenue grew 27%, but capital expenditures reached $31.08 billion in the quarter alone and the company now expects $130 billion to $145 billion of capex for 2026, leaving free cash flow at just $784 million versus $8.55 billion a year earlier. A Pembina Institute report cited by Reuters warned that letting data centers use Alberta's grid before their own generating capacity is operating could strain supply and raise consumer costs, and a July Angus Reid poll found 68% of Canadians would oppose a large data center near their home.

Alibaba Group laid out plans for a five trillion to ten trillion parameter model and a new Zhenwu V900 chip at its September 22 conference. The V900 delivers three times the performance of its predecessor and can power clusters of up to 500,000 chips, while CEO Eddie Wu set a target for Alibaba Cloud to operate more than 20 gigawatts of data center capacity by 2032. Reuters reported that the announcement lifted Alibaba shares on September 22, but the U.S. shares slipped about 0.1% to $109.63 at 11.10am ET on September 28, showing that early excitement has since cooled. The investment case rests on turning that computing scale into paying cloud demand, since building its own models and chips could give Alibaba more control over costs but a 20 gigawatt buildout raises the stakes for spending and utilization. The shares sit $11.86 below the $121.49 GF Value estimate, a 9.76% gap that will matter only if the expansion earns an adequate return.

Piper Sandler analyst Thomas Champion is projecting that Google's newly disclosed external chip business will generate $104 billion in TPU hardware revenue by 2028, a forecast the firm says the broader Street has yet to price in. Champion published the note after Google disclosed external hardware sales for the first time in its second-quarter 2026 earnings report, a milestone that triggered a formal reassessment of the revenue opportunity. The firm's model has TPU sales climbing from approximately $8 billion in 2026, or roughly 240,000 chips, to $51 billion in 2027, or approximately 2.4 million chips, and then to $104 billion in 2028, anchored to a facilities and power-capacity build-out of 3 gigawatts of run-rate capacity by the end of 2027 and 6GW by the end of 2028. Piper Sandler raised its price target on Alphabet to $400 from $395 and reiterated its Overweight rating, with its 2027 and 2028 Cloud revenue estimates now 15% and 17% above consensus and its Cloud EBIT estimates 18-19% above consensus. Champion also flagged that even these estimates may prove conservative, noting that Broadcom management recently spoke to 5GW of Anthropic demand in 2027 and 10GW in 2028, which Piper Sandler estimates as $105 billion and $210 billion in revenue if that capacity is achieved.