Sunday, May 31 May 31, 2026
Title: Billion-Dollar Bets & Billing Blowups: AI's Infrastructure Reckoning
Good morning. It's Sunday, May 31st, 2026, and today's briefing is brought to you by an industry spending money like it's infinite, billing developers like it's a slot machine, and stuffing AI into every wearable it can find.
Let's get into it.
---
STORY ONE: SOFTBANK DROPS €75 BILLION ON FRENCH DATA CENTERS
SoftBank announced yesterday that it will invest up to 75 billion euros — roughly 87 billion US dollars — to build AI data centers in France. That's not a typo. Eighty-seven billion dollars. The plan involves three initial sites in the Hauts-de-France region, targeting 3.1 gigawatts of capacity by 2031, with a total buildout reaching 5 gigawatts. French economic minister Roland Lescure called it a "testament to President Macron's ambition to position France as a leading destination in the AI value chain."
Context matters here: in the United States, data center construction is facing fierce opposition over power grid impacts and environmental concerns. SoftBank, which is both an investor in and a customer of OpenAI, is clearly betting that Europe — particularly France — will be more accommodating terrain for the physical layer of AI. This is the single biggest AI infrastructure commitment on the continent. Ever.
---
STORY TWO: GITHUB COPILOT'S BILLING CHANGE IGNITES DEVELOPER REVOLT
If you're a developer using GitHub Copilot, brace yourself. Microsoft is switching from flat monthly subscriptions to token-based billing starting June 1st — which is tomorrow. And developers are furious.
One Reddit user claimed their costs will jump from 29 dollars a month to nearly 750 dollars. Another shared a screenshot showing their bill going from 50 dollars to 3,000 dollars. The outcry is real and it's loud.
Microsoft's position is essentially: heavy "vibe-coders" who let agents churn for hours were getting subsidized. The new model charges for what you actually burn. But critics — including plenty of professional developers — argue that Microsoft built Copilot to work the way people were using it, and now they're pulling the rug. TechCrunch reached out to Microsoft for comment and didn't hear back by publication time. One thing's clear: this is a defining moment for AI developer tooling economics. Free-tier and subsidized-tier AI has an expiration date.
---
STORY THREE: META IS BUILDING AN AI PENDANT
Meta is reportedly developing an AI-powered pendant wearable. According to an internal memo viewed by The Information, the company plans to begin testing within a year. This builds on Meta's late-2025 acquisition of Limitless — a startup that made a clip-on pendant for recording conversations. Meta says it plans to expand its AI glasses lineup and launch something called "Wearables for Work," a business subscription tier. The goal appears to be reversing the fortunes of Reality Labs, which lost 4 billion dollars in just Q1 of this year.
Earlier AI wearables — including the Humane AI Pin — crashed hard with consumers. But companies like OpenAI, Meta, and others clearly aren't giving up on the ambient AI vision. The question is whether 2026-era hardware plus foundation models finally makes the value proposition click.
---
STORY FOUR: GOOGLE'S GEMINI SPARK — THE 24/7 AI ASSISTANT THAT MOSTLY WORKS
TechCrunch's Sarah Perez took Google's new Gemini Spark assistant for a real-world test drive this weekend, and the verdict is nuanced. Spark is Google's answer to always-on agentic AI — it runs in the cloud, integrates with Gmail, Calendar, Docs, and Sheets, and is designed to handle ongoing digital to-dos without you keeping your laptop open.
In testing, Spark performed well on some tasks — finding drugstore coupons, stacking savings — but fell flat on others. Most notably, Spark can't use Google Keep, which is a Google product. Pichai introduced Spark at Google I/O by joking it means "you can close your laptop," but Perez's takeaway is that it's genuinely useful for work-adjacent tasks, though perhaps not deserving of its own brand just yet.
That said, 24/7 cloud-native agentic assistants are the direction everything is headed. Spark is Google's early, real-world implementation.
---
STORY FIVE: XCENA RAISES $135 MILLION TO SOLVE AI'S MEMORY BOTTLENECK
Here's one for the infrastructure obsessives. XCENA, a chip startup founded by Samsung and SK Hynix veterans, just raised 135 million dollars in a Series B at a 570 million dollar valuation. Their thesis: AI's real bottleneck isn't compute — it's memory.
Every time you run an AI inference, data shuttles between memory, CPU, and GPU in a wasteful relay race. XCENA's MX1 chip processes data directly within the memory module using RISC-V cores connected via CXL — eliminating those costly round trips. The company claims what used to need 10 servers could potentially run on one. Mass production is slated for end-of-2026 off Samsung's foundry lines, with revenue expected in 2027. The timing is sharp: Samsung, SK Hynix, and Micron all crossed trillion-dollar valuations this month.
---
STORY SIX: THE BUZZ — AI TOOLS CONSOLIDATION AND THE SUPER APP RACE
Two quick signal items worth tracking together.
First: Microsoft is reportedly working on a unified AI super app that combines GitHub Copilot, the Copilot chatbot, Copilot Cowork, and a new agentic workflow tool internally called "Autopilot." Fortune broke the story, and the timing lines up with Microsoft Build next week. OpenAI has been pursuing its own super app ambitions. This is the AI tooling arms race condensing into single interfaces.
Second: OpenAI quietly sunset ChatGPT's Canvas interface. Canvas — the side-by-side code and text editor — won't work with GPT-5.5 Instant or GPT-5.5 Thinking. Subscribers can use it a "limited time" through legacy models. OpenAI is also actively trimming GPT-5.5 Instant responses to be shorter and less bullet-heavy. The AI interface is consolidating and simplifying.
---
BUSINESS IDEA OF THE DAY: AI COST INTELLIGENCE FOR DEVELOPER TEAMS
Here's today's gem.
The GitHub Copilot billing explosion isn't a one-off. It's the leading edge of a category crisis: organizations adopting AI tooling under flat-rate pricing that is now converting to usage-based billing — across Copilot, Cursor, Claude Code, OpenAI, and a dozen other tools simultaneously.
The business: an AI spend intelligence platform specifically for engineering and product teams. Think finops, but for AI tooling instead of cloud compute.
What it does: ingests token usage data across every AI development tool a company uses — from IDE assistants to API calls to autonomous agent runs — normalizes the data, surfaces anomalies and runaway cost patterns, forecasts monthly burns, and suggests configuration changes to reduce waste without killing productivity.
Who pays: CTOs, engineering managers, and procurement leaders at companies with 20 to 500 engineers who are now realizing their AI tooling budgets were built around flat-rate assumptions that no longer exist. This is real pain, today, not theoretical.
Why now: Microsoft's billing change goes live tomorrow. Every Copilot customer with heavy agent usage is about to get a bill shock. That's a forcing function. Companies need visibility before the invoice arrives.
What makes it defensible: deep integrations with the specific billing APIs of AI dev tools — Copilot, Claude Code, Cursor, OpenAI API — create switching costs. The normalized cost attribution model across tools is genuinely hard to build. And the early customer relationships with engineering-forward teams create distribution through word of mouth among technical buyers.
The wedge is simple: offer a free 30-day retrospective audit — "here's what your team actually spent last month, broken down by tool, team, and task type." That sells the subscription.
---
That's your AI Morning Briefing for May 31st. Trillion-dollar infrastructure bets, angry developers staring at thousand-dollar bills, wearables that won't quit, and the slow consolidation of AI interfaces into single apps. The race to own attention — and your compute budget — is fully on.
Have a great Sunday. See you tomorrow.
---
SOURCES: - TechCrunch: SoftBank €75B French data centers (May 30, 2026) - TechCrunch: GitHub Copilot token billing revolt (May 30, 2026) - TechCrunch: Meta AI pendant report (May 30, 2026) - TechCrunch: Google Gemini Spark review (May 30, 2026) - TechCrunch: XCENA $135M Series B (May 29, 2026) - The Verge: Microsoft AI super app (May 29, 2026) - The Verge: OpenAI Canvas sunset (May 29, 2026) - Anthropic Newsroom: Claude Opus 4.8 + $65B Series H (May 28, 2026)