Digital Colliers Daily Briefing — June 19, 2026
The infrastructure layer of the AI industry took on a distinctly political shape today. Federal regulators imposed structural change on how data centers connect to the US grid, the White House's week-long standoff with Anthropic hardened into a de facto licensing regime, and Google and Amazon moved separately to chip away at Nvidia's accelerator monopoly. Together, the three stories describe an industry where capital, compute, and regulatory access are now being negotiated in roughly equal measure.
1. FERC's unanimous fast-track order rewrites the queue for data center interconnections
What happened. The Federal Energy Regulatory Commission unanimously approved a package of orders directing six major grid operators to expedite interconnection requests from data centers and other large electricity users, with a target of resolving requests within 90 days, according to Bloomberg's account flagged by Techmeme. Operators have 30 days to report available generating capacity and 60 days to defend or revise regional electricity rates. Data centers will bear interconnection costs directly. As TechCrunch reports, FERC also told grid operators to consider "alternative transmission technologies"—language that opens a regulatory door for solid-state transformers, superconducting lines, and similar grid-tech startups—and to be more accommodating to behind-the-meter power arrangements.
Why it matters. The orders address the queue, not the underlying supply problem. At the end of 2023, grid connection requests for power plants exceeded the total capacity of the existing fleet, meaning generation is itself stuck in line. Wholesale electricity rates are up as much as 267% over five years, per Bloomberg figures cited by TechCrunch, and data center demand is projected to nearly triple through 2035. By prioritizing large loads without resolving generation, FERC effectively concentrates scarce capacity toward hyperscale buyers.
Who is affected. Hyperscalers and colocation developers gain a clearer timeline and a sanctioned path for behind-the-meter generation, which had become a costly workaround. Grid operators—PJM in particular, which TechCrunch describes as "descending into something resembling chaos"—face a 90-day procedural clock. Residential and industrial ratepayers in affected regions face continued upward pressure on wholesale prices. Grid-tech vendors get a regulatory tailwind. The order lands as the Trump administration paid $765 million to Invenergy to cancel offshore wind leases, with that capital being redirected toward Midwest gas and Western geothermal—a reshuffling that further tilts the generation mix.
What to watch next. The 30-day capacity reports will be the first concrete read on how much headroom actually exists in each ISO footprint. Watch PJM's filing closely given utility threats to withdraw, and watch whether FERC's "alternative transmission technologies" language translates into procurement activity at MISO and ERCOT.
Sources:
2. The Mythos/Fable standoff hardens into an ad-hoc US AI licensing regime
What happened. Nearly a week after the Trump administration issued an export control directive forcing Anthropic to pull Claude Mythos and the safeguarded Fable 5 model offline, the two sides remain at an impasse, according to Wired's Maxwell Zeff. The trigger, as The Verge's Hayden Field reconstructed on Decoder, was a 90-minute ultimatum delivered Friday after Amazon CEO Andy Jassy raised concerns to Treasury Secretary Scott Bessent about a potential jailbreak surfaced by Amazon researchers. A parallel concern involved Anthropic's sharing of Mythos with SK Telecom, which US officials allege has ties to China. The order bars all foreign nationals—including Anthropic's own non-citizen staff and customers such as Apple, Meta, and most of the Fortune 500—from access. Bloomberg, via Techmeme, reports that roughly 200 companies in Anthropic's Project Glasswing preview program have nonetheless retained access despite the shutdown order.
Why it matters. As Wired notes, the episode functionally converts what the White House had described as a "voluntary" early-testing regime into a licensing one. A former White House technology official quoted by Wired put it bluntly: "It seems very clear that what they are now doing is a licensing regime." The criteria are opaque—David Sacks's X post is the closest thing to a public explanation—and the bar Anthropic is reportedly being asked to clear, a guarantee against jailbreaks, is not technically achievable for probabilistic systems.
Who is affected. Anthropic's enterprise customer base is locked out of frontier capability. Competitors are recalibrating: Cohere CEO Aidan Gomez told Wired that "advance notice, advance access" has become the operating expectation. As Pragmatic Engineer's Gergely Orosz argues, the move likely accelerates international adoption of Chinese open-weight models such as GLM, as non-US firms now have to treat US political risk as a business-continuity variable. The Verge reports that backup contracts with non-US vendors and open-weight deployments on alternative hardware are already being signed.
What to watch next. Whether the foreign-nationals restriction is rolled back in the next round of DC meetings; whether OpenAI or Google receive similar treatment for their Mythos-class equivalents (industry sources cited by The Verge argue GPT-5.5 exhibits comparable capabilities); and the 60-day clock on the June 2 executive order, which will determine which models fall under formal review.
Sources:
- The White House Is Making Up Its Rules for AI in Real Time — Wired
- Sources: about 200 companies accessing Mythos Preview through Project Glasswing have preserved their access despite a recent US government shutdown order (Bloomberg) — Techmeme
- Who decides when AI is too dangerous? — The Verge AI
- The Pulse: Big implications of US banning Anthropic's new model, Fable — Pragmatic Engineer
- Meta's AI Workers Are Revolting, Peter Thiel's Secret Society, and SBF's Plea to Trump — Wired
3. Google's $3.2B New York data center bet—and Amazon's Trainium pitch—open a two-front campaign against Nvidia
What happened. Google is providing $3.2 billion to fund a New York data center that will rent TPUs to Anthropic, according to a Wall Street Journal report summarized by Techmeme. The Journal frames the move as Google explicitly adopting Nvidia's customer-financing playbook to seed external demand for its in-house silicon. The same day, TechCrunch reports that AWS AI chief Peter DeSantis told Bloomberg the cloud provider is in early talks to sell its Trainium chips directly to other companies for data center deployment—a reversal of AWS's longstanding policy of keeping the silicon captive.
Why it matters. Both moves attack Nvidia from different angles. Google is using its balance sheet to manufacture a reference customer and rental market for TPUs—the same financing-and-allocation tactic Nvidia has used to anchor demand at Oracle, CoreWeave, and others. Amazon, meanwhile, is testing whether it can convert internal scarcity into an external product line. CEO Andy Jassy disclosed in his April shareholder letter that, if treated as a standalone, Amazon's chip business would run at roughly $50 billion annually—comparable to Intel's revenue, though still small against Nvidia's $326 billion run rate.
Who is affected. Anthropic deepens its hardware diversification, having now committed to compute relationships across Google, AWS, and Nvidia-based infrastructure—a hedge that looks increasingly prudent given the regulatory turbulence detailed above. Nvidia faces its first credible multi-vendor challenge at the silicon layer, though TSMC capacity is a binding constraint: AWS would have to displace Nvidia in TSMC allocation to scale external Trainium sales meaningfully. Hyperscaler customers gain optionality and, potentially, pricing leverage.
What to watch next. Construction milestones at the New York TPU site and whether Google extends similar financing structures to other anchor tenants; identification of Trainium's first external buyers, which DeSantis declined to name; and Nvidia's response, particularly whether Jensen Huang's recently announced $200 billion CPU push is timed as a counter-offensive into Intel and AMD territory.
Sources:
- Sources detail how Google is using Nvidia's playbook to build an AI chip business, including providing $3.2B to fund a NY data center renting TPUs to Anthropic (Wall Street Journal) — Techmeme
- Amazon hopes to challenge Nvidia more directly by selling its AI chips — TechCrunch AI
The three stories share a common thread: the constraints on AI build-out are no longer primarily about model capability. Power interconnection queues, executive-branch discretion over model access, and the supplier concentration in accelerators are now the variables that determine which firms can ship frontier products and on what terms. The hyperscalers that emerge best positioned over the next 18 months will likely be those that can simultaneously secure megawatts, maintain working relationships with an unpredictable White House, and reduce single-vendor exposure on silicon—a triangulation that, as of today, no single company has fully solved.