Two things moved this week: money and law. Microsoft is spending billions to build AI capacity in Europe. A US court signed off the largest copyright settlement on record. Both make AI easier for a small business to buy with confidence. Plus one image tool worth watching, and two older releases that still matter more than most headlines.
1. Microsoft funds Mistral's European build-out
Microsoft agreed this week to spend billions on Mistral's computing infrastructure in Europe, including data centres in France. Azure customers get wider access to Mistral's models, and Mistral's Medium 3.5 and OCR 4 join Microsoft's Foundry catalogue. No equity changed hands. Exact terms are undisclosed.
Why it matters for SMEs: if your contracts or your GDPR position require data to stay in Europe, you can run capable models on European infrastructure through the Azure account you already have. More models in the Azure catalogue also means more price competition.
2. Court approves Anthropic's $1.5bn copyright settlement
A federal judge gave final approval on Monday to Anthropic's $1.5bn settlement with book authors — about $3,000 for each of roughly 500,000 books copied from pirate libraries to train Claude. It is the largest copyright class settlement on record. Note the scope: an earlier ruling in the same case found that training on lawfully bought books was fair use. The payout is for the pirated copies.
Why it matters for SMEs: this does not change your risk when using AI tools. The case was about how a model was trained, not what customers do with it. The lesson is data provenance. If you train or fine-tune on content you do not have rights to, there is now a price on that.
3. Meta's Muse Image is live — but not in the UK yet
Meta launched Muse Image on 7 July, the first image model from its Superintelligence Labs division. It runs inside the Meta AI app, WhatsApp, and Instagram Stories, free for everyday use. Rollout is US-first. The UK is not included yet.
Why it matters for SMEs: free image generation inside the apps your customers already use is genuinely useful for social content — when it arrives here. Nothing to do today except know it is coming.
Catch-up: two releases you may have missed
Not from this week, but they shape what we recommend, so a paragraph each.
GPT-5.4 mini and nano (released March). OpenAI's small models. Mini sits in ChatGPT's free tier; nano is API-only, built for embedded use. For routine work — sorting email, pulling data out of documents, drafting standard replies — small models do the job at a fraction of the cost of the flagships.
Claude's 1M-token context (generally available since March). The long-context surcharge is gone; standard pricing applies across the full window. In practice: a whole contract pack, tender pack, or codebase in a single request, instead of chopping documents up and losing the thread.
Kontinuity builds AI automation for small firms. If you want to know which of these actually fits your business, book a call.