AI News Daily 2026-07-14
00Executive summary
- Masayoshi Son used the opening keynote of SoftBank World 2026 to put a number on the AI buildout: worldwide AI infrastructure investment running at 5 trillion dollars (roughly 800 trillion yen) a year by 2040.
- The same keynote framed the demand side — 100 trillion AI agents and 1 billion humanoids in operation — and the supply side, with natural gas powering AI data centres in the near term and nuclear fusion positioned as the generation source after that.
- SoftBank's AI cyber defence business has drawn interest from 137 companies, an early read on how quickly the security layer of the AI stack is being commercialised.
- China's interim measures for anthropomorphic AI interaction services take effect on July 15, and ByteDance's Doubao and Alibaba's Qwen have already notified users that agent features will be switched off.
- The two vendors are handling user data differently: Qwen permanently deletes agent settings and conversation history, while Doubao keeps them readable until October 15 — a divergence that matters for anyone building personality-bearing AI products for the Chinese market.
01Masayoshi Son: “AI infrastructure investment will be 800 trillion yen a year in 2040” — SoftBank World 2026 keynote
Published: 2026-07-14 | Category: Corporate developments | Source tier: Tier 2
Facts
SoftBank World 2026, SoftBank's largest corporate event, opened on July 14. In the opening keynote, Chairman and President Masayoshi Son set out his view that global AI infrastructure investment will reach a scale of 5 trillion dollars — approximately 800 trillion yen — per year in 2040.
Alongside the investment figure, Son described the world that spending is meant to serve: a future in which 100 trillion AI agents and 1 billion humanoids are in operation. He also addressed the question that increasingly gates any AI infrastructure plan — where the electricity comes from. His answer was staged: natural gas as the power source for AI data centres for the time being, with nuclear fusion generation the promising option after that.
The keynote also touched on SoftBank's AI cyber defence business, which the company says has attracted interest from 137 companies.
Background
Three separate reports underpin this chapter — the Nihon Keizai Shimbun on the 800 trillion yen investment outlook, Bloomberg on fusion as the next-generation power source for AI, and a further Nihon Keizai Shimbun report on the 137 companies interested in the cyber defence business. Taken together they describe a single keynote argument rather than three unrelated announcements: demand at unprecedented scale, the capital required to serve it, the energy that capital has to buy, and the security layer that sits on top.
The framing also lines up with a domestic trend already covered in this morning's edition — the shift of “AX” from a slogan into an execution platform. The keynote is one of the more concrete pieces of evidence for that shift so far.
Implications
The value of the number is not that it is precise but that it is public and specific. A forecast of this kind, from one of Japan's largest AI investors, becomes a reference point that other people's plans are measured against: it gives anyone sizing an opportunity in AI infrastructure, power generation or security a stated benchmark to agree or disagree with, rather than an open-ended guess.
The energy sequencing deserves particular attention. Treating natural gas as the bridge and fusion as the destination is a statement about timelines as much as about technology, and it puts the power question — not the chip question — at the centre of the AI infrastructure conversation. The 137 companies circling the cyber defence business point the same way: the parts of the AI stack that are being commercialised fastest are increasingly the unglamorous ones.
For readers assessing exposure, the practical question is which of the four layers in the keynote — compute, power, agents and humanoids, security — their own business actually sits on, and whether the 2040 figure changes the size of the market they are planning against.
Sources
02China's AI companion rules take effect on July 15 — Doubao and Qwen switch off agent features
Published: 2026-07-06 | Category: Regulation and policy | Source tier: Tier 2
The reporting itself dates from July 6, which falls outside this edition's normal 48-hour window. It is carried here as a deliberate exception because the rules take effect tomorrow, July 15.
Facts
China's Interim Measures for the Administration of AI Anthropomorphic Interaction Services (AI擬人化インタラクションサービス管理暫定弁法), promulgated on April 10 by five government bodies including the Cyberspace Administration of China, take effect on July 15.
The measures cover AI companions that sustain an ongoing emotional relationship with the user. Providing virtual lover or virtual family services to minors is prohibited.
Two of China's largest consumer AI services have already moved. ByteDance's Doubao and Alibaba's Qwen have both notified users that agent features will be suspended. The handling of user data differs between them: Qwen will permanently delete users' agent configurations and conversation history, whereas Doubao will keep them viewable in read-only form until October 15.
Background
Bloomberg and Quartz both reported the vendor response on July 6, several weeks after the April 10 promulgation and roughly ten days before the effective date. That gap is the interesting part of the story: the companies are not reacting to enforcement, they are pre-emptively removing a product category ahead of it.
Implications
This is among the first regulations anywhere to draw a legal line between an agent that does work and an agent that keeps you company. Productivity bots fall outside its scope. Any AI service that has been given a personality and is offered into the Chinese market falls squarely inside it.
The divergence in data handling is the second thing worth watching. Two vendors facing the same rule have chosen opposite defaults — permanent deletion at Qwen, a read-only grace period until October 15 at Doubao — and from a user-protection standpoint those are very different outcomes for people who have accumulated months of conversation history. Where a rule specifies what a service may no longer do but not what happens to the data it already holds, vendor discretion fills the gap.
For product teams outside China the immediate takeaway is narrower than “AI regulation is tightening.” It is that the anthropomorphic layer — persona, emotional continuity, relationship framing — is now a regulated surface in its own right, separable from the underlying model and from the productivity use cases built on it.
Sources
03Editor's note: capital and rulemaking on the same calendar
Read side by side, the two stories in this edition are the two halves of the same question. One puts a price on how much AI capacity the world intends to build; the other shows a government deciding which uses of that capacity are acceptable and removing one of them from the market before the deadline arrives.
Regulatory clock: the FTC comment window
A third date belongs on the same calendar. The United States Federal Trade Commission's policy statement on the suppression of AI accuracy, carried in the Federal Register on July 7, has a public comment deadline of July 31. The framework treats ideological manipulation of AI output as consumer deception. Companies considering a submission have a little over two weeks left.
A week of dated events
The days immediately ahead are unusually dense with fixed dates:
- July 15 — China's AI companion regulation takes effect (chapter 02).
- July 17 — scheduled release date for Gemini 3.5 Pro.
- July 19 — the 50 percent increase in Claude Code weekly limits ends.
- July 31 — deadline for public comment on the FTC policy statement.
None of these is a surprise, which is precisely why they are useful: each is a point at which something concrete changes, and each is far enough away to plan around. The reporting behind this edition is set out in full in the references below, including Bloomberg's account of the SoftBank keynote and its report on the withdrawal of AI companion features in China.
If there is a single line to carry out of the day: the money is being sized for a decade, and the rules are being written for next Wednesday.