31 May 2026: SoftBank bets on French AI compute (AM)

This morning’s AI news is a reminder that the visible tools only work because a much larger infrastructure race is happening underneath them. SoftBank is putting serious money behind European compute, browser makers are trying to make AI the new front door to the web, and developer tools are starting to show users the real cost of heavy agent use.

SoftBank says it plans to invest up to EUR75 billion in French AI data centres, making infrastructure the lead story rather than another model launch. According to SoftBank’s official announcement, the group wants to develop and operate 5 gigawatts of AI data centre capacity in France. The first phase is an initial EUR45 billion plan for 3.1 gigawatts in the Hauts-de-France region, with sites named at Dunkirk, Bosquel and Bouchain.

The practical point is that AI capability is becoming a power, land and supply chain question. A faster chatbot is not much use if the compute behind it is scarce, expensive or concentrated in too few regions. For UK readers and small businesses, this matters because European AI hosting capacity could affect latency, resilience, procurement rules and where sensitive workloads are allowed to run.

SoftBank says the project will also involve Schneider Electric and local manufacturing capacity. That makes the story bigger than one investor’s data centre spending. The thing to watch is whether European governments keep treating AI infrastructure as industrial policy, because that could decide which tools feel affordable and reliable over the next few years.

AI browsers are turning from a novelty into a serious product category, but the trade-off is how much context users are willing to hand over. A TechCrunch roundup points to Perplexity Comet, The Browser Company’s Dia, Opera Neon and OpenAI’s Atlas as part of a broader shift away from the browser as a passive window.

The appeal is obvious. If a browser can summarise pages, compare tabs, search your history and perform basic actions, it becomes more useful than a conventional search box. The cost is also obvious. A browser sits close to your accounts, documents, shopping, messages and work systems, so an AI layer inside it needs a higher trust bar than a standalone chatbot.

That makes browser choice a practical AI decision, not just a preference for Chrome, Safari or something smaller. Readers who already use AI for research may find this useful, but it is worth checking permissions, data settings and paid plan limits before making an AI browser your default. Cristoniq’s guide to AI note-taking and productivity tools is a useful companion here because the same privacy and workflow questions apply.

GitHub Copilot’s move towards usage-based billing shows that agentic coding is leaving the cheap trial phase. TechCrunch reported developer frustration around Copilot’s new token-based charging model, while GitHub’s own announcement says Copilot plans will move to GitHub AI Credits from 1 June 2026. The exact effect will depend on plan type and usage, but the direction is clear: heavy AI coding sessions are becoming metered infrastructure, not just a flat subscription perk.

For solo developers and small teams, the question is no longer simply whether AI coding tools are useful. It is whether the work they save is worth the variable cost they introduce. Long agent runs, repeated retries and broad codebase scans may feel effortless in the editor, but they still consume model time and compute somewhere.

The sensible response is not to abandon AI coding tools. It is to use them with clearer boundaries: smaller tasks, tighter prompts, reviewed diffs and cost alerts where the provider offers them. That is also why Cristoniq’s explainer on free AI tools versus paid AI tools matters. The best tool is not always the one with the longest feature list, it is the one whose limits you understand before the bill arrives.

Google’s Gemini Spark testing shows agent assistants are getting more useful, but still need human checking around details. In a hands-on TechCrunch test, Gemini Spark helped with everyday tasks such as local planning, inbox summaries and shopping research. The same test also found gaps, including an invalid promo code and missing support for Google Keep in one workflow.

That is probably the right level of expectation for consumer agents in 2026. They can reduce the blank-page work of planning, comparing and organising, but they are not yet a clean replacement for checking dates, costs, links and final actions. The useful framing is assistant first, autopilot second.

The bigger question is why Google has separated Spark from the wider Gemini brand. Separate names can make products easier to market, but they can also confuse users who just want to know which assistant can touch Gmail, Calendar, Docs or Sheets. Watch whether Google folds Spark deeper into Workspace, because that would make the product more relevant to small businesses than a standalone consumer experiment.

At a glance. Here is everything else worth knowing from today’s AI news.

• OpenRouter raised a large Series B. OpenRouter said it raised $113 million and now serves developers building across hundreds of models. The useful signal is not the valuation chatter, it is that multi-model routing is becoming a product category of its own.
• Meta is reportedly working on an AI pendant. TechCrunch reported that Meta is developing a wearable AI pendant, citing a memo viewed by The Information. Treat this as early hardware reporting, not a product you can buy today.
• Local MoE model work is still pushing into constrained hardware. A new arXiv paper on Rotary GPU explores local execution for large mixture-of-experts models under limited VRAM. That is technical, but it points to the wider effort to make bigger models run outside the largest cloud setups.

What to watch next. The next useful signal is whether these stories converge: more European compute, more AI browsers, more metered coding agents and more consumer assistants. If they do, the real divide will not be between people who use AI and people who do not. It will be between users who understand the cost, permission and reliability trade-offs, and users who simply accept the defaults.

This is a daily news update for informational purposes only. AI products and policies change rapidly. Verify details directly with providers before making decisions. Nothing here is financial or legal advice.

AI Daily is Cristoniq’s daily guide to developments in artificial intelligence, published every morning.