Earnings call transcript: SoftBank Corp Q4 2025 misses EPS, revenue beats

SoftBank reported FY2025 revenue of JPY 7.04 trillion (+8% YoY) and net income of JPY 550.8 billion (+5%), both record highs, but EPS badly missed expectations at 1.31 yen versus 7.7 yen forecast. The stock fell 0.5% after hours as investors weighed the earnings miss against a modest revenue beat, strong cash flow, and management’s FY2030 targets for JPY 1.7 trillion operating income and JPY 700 billion net income. The company also outlined a large AI infrastructure push, including AI data centers, cloud services, and battery manufacturing tied to its long-term “Activate AI for Society” strategy.

Analysis

This is less a clean fundamental beat than a roadmap for where incremental capital will migrate over the next 12-36 months. The key signal is that SoftBank is turning itself into a demand aggregator for the AI stack: compute, data center shell, sovereign cloud, and power infrastructure. That should be constructive for NVDA, ARM, MSFT, GOOGL, ORCL, and V because the spend is shifting from speculative model training to monetizable inference and enterprise deployment, which is stickier and more utility-like.

The second-order winner is not just chip demand but the entire attachment ecosystem around power, networking, security, and payment rails. A buildout anchored in AI data centers and enterprise software implies recurring demand for networking, identity, and transaction services; ORCL and MSFT benefit from sovereign/private-cloud positioning, while V gains if AI-driven commerce and embedded payments expand across SoftBank’s consumer and financial layers. The more interesting read-through is to CRWV: if Japanese hyperscale demand is increasingly satisfied through modular, off-balance-sheet or customer-owned GPU structures, the market may be underestimating how quickly “neocloud” economics become commoditized unless differentiated by software and access to power.

The main risk is timing mismatch: management is effectively monetizing a long-duration option while funding it with near-term operating cash, which can look elegant until power, GPU, and memory costs re-rate faster than customer pre-commitments. Any delay in power availability, permitting, or customer off-take would push the story from monetization to capital intensity, and that is where the equity can de-rate quickly over a 1-2 quarter window. Conversely, the upside catalyst is visible: if the first AI data-center phases launch on schedule and pre-sold capacity converts into revenue, the market will likely re-rate this as a real infrastructure earnings compounder rather than a vague AI adjacency story.

Consensus is probably underestimating how much of the value accrual sits outside the telecom P&L. The market may focus on the EPS miss and miss that the strategic shift raises the optionality of enterprise cloud, sovereign AI, and power equipment manufacturing — a mix that can support multiple expansion if execution is even moderately credible. The contrarian view is that the dividend and capex narrative can coexist only if off-balance-sheet funding remains available; if not, the stock could stall despite good top-line optics.

Key Decisions for Investors

• Long NVDA vs short CRWV over 3-6 months: express the view that customer-precommitted AI infrastructure demand will keep chip leverage superior to bare-metal/neocloud exposure; target ~15-20% relative outperformance if capacity remains tight.
• Initiate a basket long in MSFT + ORCL on any 3-5% pullback over the next 2-4 weeks: sovereign/private cloud demand should benefit from enterprise data-residency concerns and the shift from training to inference; risk/reward favors 2:1 if AI deployment budgets reaccelerate.
• Add ARM on weakness for a 6-12 month horizon: the market is still underpricing the embedded value of edge/inference workloads if AI moves from centralized training to distributed deployment; use a 10-15% trailing stop.
• Buy V as a low-volatility AI-adjacent beneficiary over 6-12 months: if SoftBank’s financial/consumer ecosystem deepens AI-enabled commerce, payments should capture transaction growth without direct capex risk; look for 8-12% upside with limited drawdown.
• Avoid chasing the broad AI infrastructure basket until power/off-take milestones are validated: if data-center permitting or grid access slips, reassess longs in infrastructure names and rotate toward software beneficiaries.