AI infra: big capital bets
What happened
Microsoft plans a $10 billion, four‑year AI investment in Japan while Alphabet leans on tax incentives and data‑centre builds, and demand for Nvidia’s Blackwell systems is accelerating across cloud providers. ( ) Those moves show executives are making explicit capital-allocation bets under uncertainty—and the same portfolio-thinking should be used when proposing analytics investments. (markets.financialcontent.com)
Why it matters
Microsoft said the $10 billion package announced in Tokyo will be deployed from 2026 through 2029 and will include partnerships with Sakura Internet and SoftBank to supply computing capacity, plus a pledge to train one million AI engineers by 2029; Sakura’s stock jumped about 20% on the news. (bloomberg.com) Alphabet has outlined a roughly $175–$185 billion capital plan for 2026 and is explicitly front‑loading spending because a 2025 tax change restored 100% bonus depreciation, which lets companies deduct the full cost of qualifying equipment in the year it’s placed in service; Alphabet’s plan includes activating more than 20 new data‑center campuses and a roughly $40 billion commitment in Texas called the Goodnight campus. (markets.financialcontent.com) NVIDIA’s Blackwell platform is driving large cloud orders and system deployments: one industry analysis estimates about $1 trillion of Blackwell and Rubin orders across the market, and NVIDIA’s GB300 NVL72 systems are being rolled out by major cloud providers for low‑latency, long‑context inference workloads. (ainvest.com)(blogs.nvidia.com) “100% bonus depreciation” means companies can immediately write off the full purchase price of servers, cooling and other equipment against that year’s taxable income, creating a large near‑term tax shield that lowers the effective upfront cost of CAPEX; that accounting treatment is the explicit fiscal engine behind Alphabet’s accelerated build‑out. (markets.financialcontent.com) The Blackwell shift also changes data‑center engineering: newer Blackwell racks increase compute per rack so much that providers are redesigning power distribution — for example, moving toward higher‑voltage direct‑current (around 800 volts DC) to support denser, more power‑hungry racks without overheating or excessive losses. (ainvest.com) Those capital allocations are being made against an industry backdrop where the largest cloud and AI providers are planning roughly $650 billion of data‑center spending in the current build cycle, and governments are matching parts of the push (Japan earmarked about ¥1.23 trillion to support chips and AI this fiscal year), creating local incentives and also straining power availability in energy‑importing markets. (bloomberg.com) Public markets reacted unevenly to the scale and speed of these bets — Alphabet saw an initial price dip and analysts flagged questions about whether the near‑term tax benefit fully offsets the multi‑year capital commitment — underscoring that these are explicit, concentrated allocation decisions that hinge on tax rules, energy access, supplier capacity, and near‑term unit economics for advanced GPUs and racks. (markets.financialcontent.com)
Key numbers
- Microsoft plans a $10 billion, four‑year AI investment in Japan while Alphabet leans on tax incentives and data‑centre builds, and demand for Nvidia’s Blackwell systems is accelerating across cloud providers.
What happens next
- (markets.financialcontent.com) Microsoft plans a $10 billion, four‑year AI investment in Japan while Alphabet leans on tax incentives and data‑centre builds, and demand for Nvidia’s Blackwell systems is accelerating across cloud providers.
Quick answers
What happened in AI infra: big capital bets?
Microsoft plans a $10 billion, four‑year AI investment in Japan while Alphabet leans on tax incentives and data‑centre builds, and demand for Nvidia’s Blackwell systems is accelerating across cloud providers. ( ) Those moves show executives are making explicit capital-allocation bets under uncertainty—and the same portfolio-thinking should be used when proposing analytics investments. (markets.financialcontent.com)
Why does AI infra: big capital bets matter?
Microsoft said the $10 billion package announced in Tokyo will be deployed from 2026 through 2029 and will include partnerships with Sakura Internet and SoftBank to supply computing capacity, plus a pledge to train one million AI engineers by 2029; Sakura’s stock jumped about 20% on the news. (bloomberg.com) Alphabet has outlined a roughly $175–$185 billion capital plan for 2026 and is explicitly front‑loading spending because a 2025 tax change restored 100% bonus depreciation, which lets companies deduct the full cost of qualifying equipment in the year it’s placed in service; Alphabet’s plan includes activating more than 20 new data‑center campuses and a roughly $40 billion commitment in Texas called the Goodnight campus. (markets.financialcontent.com) NVIDIA’s Blackwell platform is driving large cloud orders and system deployments: one industry analysis estimates about $1 trillion of Blackwell and Rubin orders across the market, and NVIDIA’s GB300 NVL72 systems are being rolled out by major cloud providers for low‑latency, long‑context inference workloads. (ainvest.com)(blogs.nvidia.com) “100% bonus depreciation” means companies can immediately write off the full purchase price of servers, cooling and other equipment against that year’s taxable income, creating a large near‑term tax shield that lowers the effective upfront cost of CAPEX; that accounting treatment is the explicit fiscal engine behind Alphabet’s accelerated build‑out. (markets.financialcontent.com) The Blackwell shift also changes data‑center engineering: newer Blackwell racks increase compute per rack so much that providers are redesigning power distribution — for example, moving toward higher‑voltage direct‑current (around 800 volts DC) to support denser, more power‑hungry racks without overheating or excessive losses. (ainvest.com) Those capital allocations are being made against an industry backdrop where the largest cloud and AI providers are planning roughly $650 billion of data‑center spending in the current build cycle, and governments are matching parts of the push (Japan earmarked about ¥1.23 trillion to support chips and AI this fiscal year), creating local incentives and also straining power availability in energy‑importing markets. (bloomberg.com) Public markets reacted unevenly to the scale and speed of these bets — Alphabet saw an initial price dip and analysts flagged questions about whether the near‑term tax benefit fully offsets the multi‑year capital commitment — underscoring that these are explicit, concentrated allocation decisions that hinge on tax rules, energy access, supplier capacity, and near‑term unit economics for advanced GPUs and racks. (markets.financialcontent.com)
