Google Starts Training Gemini 4 — 2026 Investment Forecast Raised to Up to $205 Billion

Alphabet CEO Sundar Pichai said the next leap forward for Gemini requires a "bigger base model," revealing that training runs for Gemini 4 have already begun. The company's 2026 investment forecast has been raised to as much as $205 billion (roughly 30 trillion yen), with demand now outpacing the pace of spending.

Google Cloud posted 82% year-over-year growth in the second quarter, a sign that its AI infrastructure investments are starting to pay off. Google also disclosed that it holds SpaceX shares worth $94.1 billion (roughly 14 trillion yen) following SpaceX's IPO — a return earned as an early investor, with $80 billion of that amount subject to lockup restrictions on short-term sales.

Kimi K3 Isn't "Just Distillation" — Experts Push Back

The Kimi K3 model, developed by China's Moonshot AI, has long been dogged by speculation that it reached its performance level purely through distillation from Anthropic's test model "Fable." But several experts told TechCrunch that "you can't build a model this powerful, this quickly, strictly by distilling from Fable."

While the full technical details of how Kimi K3 was actually trained have not been disclosed, the prevailing view is that distillation alone doesn't explain it.

Poolside's "Laguna S 2.1" — A Small but Powerful Coding Model

Poolside has released "Laguna S 2.1," its third consecutive monthly coding model release. What sets it apart is that, rather than relying on scale, it has been trained to "repeatedly double-check its own work, correct failed approaches, and not give up too easily during long agent sessions."

On benchmarks, it posted scores beating several larger rival models. It also reportedly solved a math problem that had remained unsolved since 1975 for less than 10 cents, drawing attention for its cost-performance as well.

AMD Unveils Next-Generation AI Infrastructure to Take On Nvidia

AMD has announced the launch of next-generation AI infrastructure aimed at challenging Nvidia's dominance in AI chips. Reported by Reuters, the move signals that competition in the AI data center market is set to intensify further. Detailed specs and product names are expected to be revealed at the launch event, and Nvidia's monopoly position could be in for a change.

The Battle Over Open-Weight AI Regulation — Big Tech vs. Startups

OpenAI and Anthropic have taken the position of calling for regulation, citing the risks of Chinese open-weight (open source) AI models, while the startup industry is pushing back.

The "Little Tech Association," whose roughly 200 member startups include Y Combinator, has petitioned the Trump administration to hold off on banning Chinese open-weight AI models. Arcee AI and others have likewise voiced opposition to banning open Chinese models.

The standoff has become a complex fight that involves not only "safety" versus "innovation," but also the business interests of established incumbents.

"x402," a Payment Rail for AI Agents — HTTP 402 Revived After 35 Years

The status code "402 Payment Required," reserved in the HTTP specification back in 1991 and never used by anyone, has been brought back to life as a payment standard for the AI agent economy.

More than 40 companies, including Visa and Mastercard, have joined the "x402 Foundation" led by the Linux Foundation. The vision is to serve as infrastructure for an era in which AI agents transact autonomously, enabling payment processing at the HTTP protocol level. AI agent payments have so far relied on bespoke implementations by each company, and standardization is expected to mature the ecosystem as a whole.

Other Headlines

Google introduces selfie video login**: The company has added a new authentication method that lets users sign in to their accounts with a video of their face, as a safeguard against lost devices and forgotten passwords.

AI-driven drug design accelerates**: MIT Technology Review reports that AI is beginning to play a central role in the drug discovery process, particularly in the development of biologic drugs. AstraZeneca and others are actively expanding their teams.

Generative AI floods the book market**: A study published on arXiv has documented how books created with generative AI are diluting the market.

---