AI inference use accounted for 60% of new orders at KDDI’s connectivity data centres, the Japanese telecom group’s results presentation for April to September 2026 shows. The unit’s operating revenue rose from 48.8 billion yen to 57.7 billion yen, up 18.2% by our arithmetic, while group revenue grew 4.7%.
KDDI’s AI Integration and cybersecurity line grew from 80.4 billion yen to 95.3 billion yen, up 18.5% by our arithmetic, or about 3.9 times the group’s growth rate. The company describes the data-centre unit as “Business growth capturing AI inference demand”.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on pricing by use rather than by seat. It teaches the general idea only and does not discuss KDDI.
The key insight: KDDI says AI inference use made up 60% of new orders at its connectivity data centres, a unit whose revenue rose 18.2% to 57.7 billion yen in April to September 2026, against 4.7% growth for the group, by our arithmetic.
KDDI’s operating revenue for the first half of the fiscal year ending March 2027 was 3,052.8 billion yen, up 4.7% from 2,916.0 billion yen, and adjusted operating income rose 13.9% to 633.1 billion yen, the presentation shows. The company called the results “Solid progress, ahead of the full-year forecast”.
The connectivity data-centre business, which the presentation labels Telehouse, lifted EBITDA from 21.4 billion yen to 25.7 billion yen. That is up 20.1%, and a margin of about 44.5% on the half’s revenue, by our arithmetic. The presentation’s chart of new orders labels 60% as AI inference use and leaves the other 40% unlabelled.
Operating revenue for April to September 2026 against the same half a year earlier, indexed to 100, for KDDI’s AI Integration and cybersecurity line, its connectivity data centres and the group, from KDDI’s results presentation of 6 November 2026. Growth rates other than the group’s 4.7% are FourWeekMBA’s arithmetic.
Where the AI Revenue Sits
Within AI Integration, the presentation gives an index of 129 against 100 a year earlier, AI development sales up “approx. 20%” and cloud infrastructure sales up “over 30%”. It says the number of AI development projects rose about threefold between March and September 2026.
KDDI splits its data-centre strategy into a “Connectivity Model” for AI operators and connectivity providers near cities, and an “AI Factory Model” at its Osaka Sakai site for AI service providers, enterprise users and sovereign cloud. For Sakai, the presentation says KDDI is “Introducing next-generation GPU ‘Rubin’ as part of the leading group”, without giving a number of chips.
The London Expansion
Under the heading “Addressing AI demand through London DC expansion”, KDDI says it plans to open a sixth building at its London campus by the fiscal year ending March 2028. The existing campus has 78MVA of power capacity on 27,178 square metres, and newly acquired adjacent land adds 65MVA on 29,000 square metres, for 143MVA in total by our arithmetic. KDDI labels the second step “Double power capacity (MVA)” from the fiscal year ending March 2029.
As we read the presentation, KDDI’s fastest-growing lines are the AI-linked ones: AI Integration and cybersecurity grew 18.5% and the connectivity data centres 18.2%, both about four times the group’s 4.7%, by our arithmetic.
As we read the order chart, inference rather than training is what KDDI is selling space and power for at Telehouse, since that is the only use the chart labels.
As we read the London plan, the constraint is power: KDDI describes the newly acquired land by its 65MVA of capacity, next to a 78MVA campus.
FOR KDDI’S DATA-CENTRE MARGINS The unit’s EBITDA rose from 21.4 billion yen to 25.7 billion yen in the half, the presentation shows.
FOR LONDON CAPACITY KDDI says it plans a sixth building by the year to March 2028 and labels the next step “Double power capacity (MVA)”.
FOR JAPAN’S GPU SUPPLY KDDI says its Osaka Sakai site is introducing the next-generation “Rubin” GPU as part of the leading group, without giving a number.
The framework says: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, the data centres that host those queries collect part of that recurring spend, which is the demand KDDI says it is capturing. The framework predates KDDI’s results and does not discuss KDDI.
What Is Not Established
The presentation does not say how the 60% is measured, whether by contract value, power or count, or over what period; the other 40% carries no label. It gives no size for the new orders, no number of Rubin GPUs for Sakai and no investment figure for the London expansion. KDDI’s results were published on 6 November; Telecompaper reported them on 9 November. The growth rates, margin and 143MVA total are our arithmetic. We did not contact KDDI.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
KDDI says AI inference use made up 60% of new orders at its connectivity data centres, a unit whose revenue grew 18.2% in April to September 2026 against 4.7% for the group, by our arithmetic. It plans a sixth London building by the year to March 2028 and is adding newly acquired land with 65MVA of power capacity next to a 78MVA campus.
A note on sourcing. KDDI’s figures, quotes and plans are from its results presentation for April to September 2026, published on 6 November 2026. The growth rates, margin, 3.9x and 143MVA figures are our arithmetic. We did not contact KDDI.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
