Digital Colliers Daily Briefing — May 31, 2026
The AI sector's capital intensity and its unit economics came into sharper focus today as three separate stories illustrated the same underlying tension: enormous sums are flowing into model labs and compute, while the costs of running these systems are being passed downstream to customers. Anthropic surged past OpenAI in private market valuation on a $65 billion round, GitHub Copilot prepares to flip from flat-rate to token-metered billing tomorrow, and SoftBank committed up to €75 billion to a 5-gigawatt French data center buildout. Together, they sketch the contours of an industry recalibrating who pays — and how much — for generative AI.
1. Anthropic's $65B Series H Pushes Valuation Toward $1 Trillion, Eclipsing OpenAI
What happened. Anthropic closed a $65 billion Series H led by Altimeter Capital, Dragoneer, Greenoaks, and Sequoia Capital, with previously committed capital — including $5 billion from Amazon — folded into the package. The round values the Claude developer at roughly $1 trillion, nearly tripling the $380 billion mark set in February and pushing it past OpenAI's $852 billion valuation from its $122 billion March round. Anthropic disclosed annual run-rate revenue of $47 billion, up from roughly $10 billion a year ago, and used the moment to introduce Claude Opus 4.8 and a closed enterprise security system branded Claude Mythos Preview. CFO Krishna Rao attributed the growth to accelerating global demand for Claude products.
That ARR figure warrants scrutiny. As Simon Willison highlighted, citing Karen Kwok for Reuters Breakingviews, Anthropic's "run-rate revenue" definition multiplies the last 28 days of consumption-based sales by 13, then adds monthly subscription revenue multiplied by 12 — a methodology that captures peak momentum rather than trailing performance.
Why it matters. The Anthropic-OpenAI duopoly has formally inverted at the top of the valuation table, and the bidding war for frontier-model equity has reached a tier where even hyperscaler balance sheets are stretched. Bloomberg's Yazhou Sun, via Techmeme, also reported that Anthropic trimmed its list of unauthorized secondary-market sellers from eight to four after the original notice rattled investors — a small but telling signal of how aggressively private holders are trying to monetize exposure.
Who is affected. Enterprise buyers now negotiate against a vendor with materially more pricing power; Amazon's strategic position as both investor and primary cloud partner strengthens; and Microsoft's OpenAI alignment looks comparatively less dominant than it did six months ago. Secondary-market investors and employees holding Anthropic equity face tighter liquidity controls.
What to watch next. CNBC has flagged a potential OpenAI IPO filing within weeks, and Anthropic is reportedly weighing the same. Whether public market investors accept the "×13" run-rate convention will determine how the headline ARR translates into a defensible float.
Sources:
- [HN · 404↑] Anthropic surpasses OpenAI to become most valuable AI startup — Hacker News](https://qazinform.com/news/anthropic-surpasses-openai-to-become-worlds-most-valuable-ai-startup)
- Quoting Karen Kwok for Reuters Breakingviews — Simon Willison
- Anthropic cuts its list of unauthorized secondary market sellers from eight to four after the initial notice caused panic and pushback from investors (Yazhou Sun/Bloomberg) — Techmeme
2. GitHub Copilot's Token Billing Goes Live June 1, and Power Users Are Doing the Math
What happened. Microsoft's GitHub Copilot transitions tomorrow from a flat-rate subscription to a token-usage billing model. According to TechCrunch, developers posting on Reddit and X have shared projected bill increases ranging from a $29 plan rising to roughly $750 a month, to one screenshot showing $50 jumping to approximately $3,000. Microsoft did not respond to TechCrunch's request for comment.
Reaction inside developer communities has split. Heavy agentic users — particularly those running long-running "vibe coding" sessions that spawn dozens of sub-agents — face the steepest jumps. Other developers argue the new pricing remains affordable for disciplined use and that the old flat rate was economically incoherent. As one Reddit commenter put it bluntly, "Holy fuck how much money was Copilot losing." A third camp lays the blame on Microsoft for actively encouraging the high-consumption workflows it is now repricing.
Why it matters. Copilot was the proof point that flat-rate AI coding subscriptions could scale. Its retreat to metered billing concedes that subsidized inference is unsustainable even for Microsoft, and it lands on the same day Hacker News surfaced a Wall Street Journal report that "Corporate America Is Starting to Ration AI as Cost Skyrockets." Together, the two stories point to a broader pricing reset across the stack — one where AI assistants begin to look less like SaaS seats and more like cloud compute.
Who is affected. Independent developers, startups, and small engineering teams bear the brunt; large enterprises with negotiated agreements may absorb the change more easily. Competing assistants — Cursor, Cognition, Anthropic's Claude Code, and Google's Gemini Code Assist — gain an opening to court defecting Copilot subscribers, though their underlying economics are no kinder.
What to watch next. Cancellation and migration data over the next 30–60 days, Microsoft's response on tiered plans or token allowances, and whether competitors hold flat-rate pricing or follow Copilot's lead.
Sources:
- 'What a joke': Github Copilot's new token-based billing spurs consternation among devs — TechCrunch AI
- With Microsoft's GitHub Copilot shifting to token-usage billing on June 1, many developers bemoan massive cost increases and the end of flat-rate subscriptions (Lucas Ropek/TechCrunch) — Techmeme
- [HN · 168↑] Corporate America Is Starting to Ration AI as Cost Skyrockets — Hacker News
3. SoftBank Commits Up to €75B to 5GW of French AI Data Centers
What happened. SoftBank Group announced plans to invest up to €75 billion (roughly $87 billion) to develop and operate up to 5 gigawatts of data center capacity in France, its largest European AI infrastructure commitment to date. The first phase, a €45 billion tranche reported by the Financial Times, will build 3.1 GW of capacity in the Hauts-de-France region — across Dunkirk (Loon-Plage), Bosquel, and Bouchain — targeted for completion by 2031. French Economy Minister Roland Lescure framed the announcement as validation of President Macron's strategy to position France across the AI value chain.
Why it matters. Europe has lagged the U.S. on frontier compute, and a 5 GW commitment from a single sponsor — one that is simultaneously an investor in and customer of OpenAI — materially shifts that picture. The choice of France leans on the country's nuclear baseload, a marked contrast with SoftBank's previously announced Ohio facility, which TechCrunch notes will be powered by a new 9.2 GW natural gas plant. With U.S. opposition to data center construction intensifying around grid strain and utility prices, France's regulatory and energy posture becomes a competitive asset.
Who is affected. European AI startups and sovereign-AI initiatives gain in-region capacity; Hauts-de-France benefits from industrial investment in formerly heavy-industry sites; U.S. hyperscalers face a more credible European alternative. Grid operator RTE and France's nuclear fleet inherit substantial new demand.
What to watch next. Site permitting timelines, power purchase agreements with EDF, the identity of anchor tenants (OpenAI being the obvious candidate), and whether other EU member states respond with competing incentive packages.
Sources:
- SoftBank says it will invest up to €75 billion to build French data centers — TechCrunch AI
- SoftBank pledges to invest up to €75B in AI computing clusters in France, first leading a €45B investment to build 3.1GW of capacity by 2031 in Hauts-de-France (Financial Times) — Techmeme
The three stories trace a single arc through the AI economy: capital flooding into model labs at trillion-dollar valuations, gigawatts of compute being poured into the ground to serve them, and end users — from independent developers to Fortune 500 buyers — increasingly asked to pay metered rates that reflect the true cost. The era of subsidized inference is closing just as the era of trillion-dollar private valuations begins, and the gap between those two facts is what the next several quarters will test.