日本語
2026-08-03 Morning edition
Morning edition
AI News Daily

AI News Daily
2026-08-03

Models are getting faster and cheaper to release while the compute underneath them is contracted a decade ahead — and the EU has just started enforcing how all of it must be labelled.

Review edition — 2 new stories, 3 looking back
Today at a glance

Today's highlights

Model release

Qwen3.8-Max goes broadly available

Published: 2026-08-03
  • Alibaba's Qwen team made the 2.4-trillion-parameter multimodal model Qwen3.8-Max widely usable through Token Plan, Qoder and other surfaces.
  • An open-weight version is scheduled for release next week.
  • The team describes the model as strong at autonomous coding and long-running tasks.
Why it matters

A top-tier Chinese model claiming performance close to Western frontier systems while going open-weight on a short clock could accelerate both cost competition and commoditisation.

2.4 trillion
parameters in the new multimodal flagship
Source: https://qwen.ai/blog?id=qwen3.8-max-preview · https://x.com/Alibaba_Qwen/status/2084100707423289643
Regulation and policy

EU AI Act: Article 50 enters enforcement

Published: 2026-08-02
  • On 2 August 2026 the European Commission announced it had begun applying and enforcing the transparency obligations under Article 50 of the AI Act.
  • Conversational systems such as chatbots must disclose that they are AI.
  • AI-generated content must carry a machine-readable marking.
Why it matters

Labelling duties for generative AI content have reached the practical stage. Companies serving the EU market now face concrete work on chatbot disclosure and content marking.

Article 50
the transparency clause now being enforced
Source: https://ec.europa.eu/commission/presscorner/detail/en/ip_26_1714
Model release · looking back

Claude Sonnet 5 becomes the default model

Published: 2026-06-30
  • Anthropic announced Claude Sonnet 5 as the new default model for the Free and Pro plans.
  • It delivers agentic performance close to Opus at lower cost.
  • Also available in Claude Code and on the Max, Team and Enterprise plans.
Why it matters

Offering top-tier autonomous-execution performance in a lower price band points to falling cost barriers for enterprise AI agent adoption.

Sonnet 5
new default on the Free and Pro plans
Source: https://www.anthropic.com/news/claude-sonnet-5 · TechCrunch
Corporate developments · looking back

Anthropic and Amazon expand to $100 billion

Published: 2026-04-20
  • Anthropic will invest more than $100 billion in AWS technology over the next ten years.
  • The deal secures up to 5 gigawatts of new compute capacity for training and running Claude.
  • New Trainium2 and Trainium3 capacity comes online in stages during 2026.
Why it matters

Locking down multi-gigawatt compute on long-term contracts is now routine, raising the weight of power and semiconductor supply chains in AI infrastructure decisions.

5 gigawatts
new capacity secured for Claude
Source: https://www.anthropic.com/news/anthropic-amazon-compute
Corporate developments · looking back

TPU capacity locked in with Google and Broadcom

Published: 2026-04-07
  • Multi-year agreements secure roughly 3.5 gigawatts of next-generation TPU capacity from 2027.
  • Anthropic said Claude's annual recurring revenue grew from about $9 billion at the end of 2025 to more than $30 billion in 2026.
  • Together with Amazon's Trainium, this makes procurement explicitly dual-track.
Why it matters

Dual-sourcing at this scale symbolises how frontier labs fence off compute in advance while avoiding dependence on any single vendor.

3.5 gigawatts
next-generation TPU capacity from 2027
Source: https://www.anthropic.com/news/google-broadcom-partnership-compute · TechCrunch
Trend overview

Three currents shaping the day

Speed and cost join the race

With Qwen3.8-Max claiming performance near Western top models and going open-weight within a week, model competition now turns on publication speed and price, not capability alone.

Rules reach the shop floor

Article 50 of the EU AI Act is being enforced, so generative-content labelling and chatbot disclosure have become live engineering and product work.

Compute booked by the gigawatt

Led by Anthropic, major labs kept securing gigawatt-scale capacity on long-term, multi-vendor contracts throughout the first half of 2026.

Wrap-up

Three things to remember

What to watch next

The promised Qwen3.8-Max open-weight release, how enforcement of Article 50 lands on real products, and the staged arrival of Trainium capacity in 2026 ahead of TPU capacity in 2027.