Skip to main content
AI Briefing

Three layers of AI economics break on the same day

GitHub Copilot moves to token billing, SoftBank pledges €75B for French data centers, Google AI kills organic traffic — predictability ends across the stack.

LV EN

Summary

  • GitHub Copilot moves from flat $29/mo to token-based billing today — developers estimate 10-50× bill increases
  • SoftBank pledges €75B for French data centers — up to 5 GW capacity in Europe, first phase ready 2031
  • Niche websites grow while Google AI kills organic traffic — historical maps site “Pastmaps” as evidence

Core feel: one day, three layers, one trend — AI business predictability is ending. You can no longer assume tool costs, capacity location, or Google traffic will look like last year.


1. GitHub Copilot moves to token billing — bills can grow 10-50×

What changed. Today, June 1, GitHub Copilot ends the flat $29 per month plan and switches to token-based billing. Developers publicly estimate bills from $750 to $3000 a month for users who previously paid $29-50. Microsoft stops subsidising — reasoning model costs on Microsoft’s side run 10-50× higher than standard completion.

Why it matters. This is not an isolated price change. It is the first major signal that the AI tooling market is moving from SaaS (fixed annual price) to consumption (pay-per-use). Your CFO can no longer lock a 12-month AI budget — usage is variable, and the more the team uses the tool, the bigger the bill. The same will happen with Claude Code, Cursor, ChatGPT Business and every other AI tool over the next 6-12 months.

Do this month.

  • List which AI tools in your company are flat-rate vs metered
  • Set spending caps per user or team — before the bill arrives
  • Make sure the team understands “standard completion” vs “reasoning models” — the latter costs 10-50× more
  • Open alternatives: Claude Pro flat $20/mo, open-source models locally, DeepSeek API

What to expect.

  • Next 60 days: Cursor, Codeium and other competitors will market “predictable pricing” as their key advantage
  • Q3 2026: Microsoft publishes “average cost per active developer” data to calm the market
  • Q4 2026: an “AI token risk management” category emerges — tools that throttle team consumption in real time

2. SoftBank pledges €75B for French data centers — Europe finally builds capacity

What changed. SoftBank announced up to €75 billion investment in French data centers — up to 5 GW additional capacity across three sites in Hauts-de-France (Dunkirk, Bosquel, Bouchain). First phase, €45B, delivers 3.1 GW by 2031. SoftBank is both an OpenAI investor and customer; the partner is Schneider Electric.

Why it matters. For the first time we see a concrete number behind “European AI sovereignty.” Until now the EU depended on US cloud providers (AWS, Azure, GCP), and GDPR/EU AI Act created legal conflict between data law and actual infrastructure. Now there is a specific answer: 5 GW of capacity in EU jurisdiction by 2031. If your company processes EU customer data with AI, this capacity will change your supplier selection — and your pricing leverage against US providers.

Do this month.

  • Review jurisdiction clauses in your AI supplier contracts — where exactly is data processed
  • If you use OpenAI/Anthropic, ask whether and when an EU “data residency” option will exist
  • Start tracking Mistral, Aleph Alpha, OVHcloud — local competitors with European data centers
  • Plan a 2027-2028 budget option for “EU-only AI deployment” as risk mitigation

What to expect.

  • 6 months: Germany and Italy announce comparable data center investments to keep up
  • 12 months: “EU-sovereign AI” emerges as a marketing category — local models on local GPUs
  • 18-24 months: regulators (Germany’s BSI, France’s CNIL) start mandating EU jurisdiction for critical sectors

3. Historical maps site grows while Google AI drains traffic

Kas mainījās. The Verge reports on Pastmaps — a niche historical maps site with 185,000+ map archive, growing while large media sites lose traffic to Google AI Overviews (“Google Zero” effect). The founder’s strategy: stop relying on Google Search, build direct relationships with a specific audience.

Why it matters. Google AI Overviews have reduced clicks to publishers by 30-60% over the last 12 months depending on niche. That means the content marketing model “write a post → rank on Google → customer arrives” no longer works the way it did in 2024. If your company has spent the last 5 years investing in SEO and blog content as the main lead generation channel, this is the first moment to ask whether it still pays back.

Do this month.

  • Open Google Search Console — compare click counts 2025-Q4 vs 2026-Q2
  • Start measuring direct traffic, email subscribers and return-visit rate as primary metrics, not just Google traffic
  • If your SEO budget exceeds your email or community budget, reallocate
  • Experiment: how big is your email list and how often do you actually email it

What to expect.

  • 6 months: several Latvian B2B companies publicly announce cutting SEO budget in favour of email and partnership channels
  • 12 months: Google introduces a “publisher payments” compensation scheme (driven by European regulator pressure)
  • 24 months: niche “direct audience” media (newsletter, community, podcast) become more valuable than Google-ranked content

Today’s picture

Three stories that look unconnected — Microsoft changes Copilot billing, SoftBank builds data centers in France, a historical maps site grows — are in fact one predictability crisis signal. For three decades digital business has rested on assumptions: software tools cost a fixed monthly fee, capacity sits where it is cheapest, and Google sends traffic if you write good content. Today all three assumptions simultaneously change shape.

Summary table:

EventConsequence
Copilot moves to token billingAI tooling budget no longer predictable
€75B French data centersJurisdiction and supplier mix will shift
Google Zero effect goes publicContent marketing is no longer the default channel

Three questions for the CEO:

  • What is your new “AI tooling budget risk” plan if consumption grows 5×?
  • Does it matter to your company in which jurisdiction your AI data is processed?
  • If Google sent you 50% fewer clicks tomorrow, what is your next customer acquisition channel?

Sources