Digital Colliers Daily Briefing — June 8, 2026
Three storylines define today's agenda, and each points to the same underlying pressure: the AI buildout is colliding with the constraints of physics, geography, and unit economics. Google is reaching beyond traditional hyperscaler supply chains to source compute from SpaceX as power scarcity reshapes data center strategy. Apple opens WWDC 2026 this afternoon with a Siri overhaul that will be measured against rivals that have spent a year extending their lead in conversational AI. And Microsoft's pivot to per-token pricing for GitHub Copilot — already nicknamed the "Tokenpocalypse" inside customer organizations — is the clearest signal yet that the subsidy era of generative AI is ending.
1. Google–SpaceX compute deal puts power, not silicon, at the center of the AI race
What happened. Google has agreed to source compute capacity from SpaceX, a sourcing arrangement Ben Thompson at Stratechery frames alongside Broadcom's latest earnings as broadly bullish for Nvidia. The deal lands on the same day the Wall Street Journal's Yusuf Khan details Ireland's "Bring Your Own Power" (BYOP) policy, which now requires new data centers to either build on-site generation or contract for nearby supply. Separately, Nvidia and LG announced an expanded partnership covering an AI factory in Korea and co-designed future data center architectures, with Jensen Huang appearing at LG's Seoul headquarters.
Why it matters. A hyperscaler buying compute from a launch and satellite company would have read as a category error two years ago. It reads today as a logical response to a market in which the binding constraint is megawatts, not GPUs. SpaceX has both interconnected power assets and a willingness to operate at the edges of where conventional utilities cannot move quickly. Ireland's BYOP regime, meanwhile, is the policy version of the same admission — that grids cannot absorb hyperscale load on hyperscale timelines without externalizing costs onto retail ratepayers.
Who is affected. Nvidia benefits regardless of who builds the building: more sites mean more accelerators. LG gains a co-design seat at the table that Korean conglomerates have historically been denied in U.S.-anchored cloud architectures. Irish operators — Microsoft, AWS, Meta, and Google all have significant Dublin footprints — face higher capex and longer permitting paths. Citizens of the host countries are the implicit second-order beneficiaries if BYOP-style policies spread.
What to watch next. Whether other power-constrained jurisdictions (the Netherlands, Singapore, Virginia) adopt BYOP analogs; whether SpaceX formalizes a compute-leasing business beyond the Google arrangement; and whether Broadcom's outlook, which Stratechery reads as confirmatory, translates into upward revisions for custom silicon programs at Google and Meta.
Sources:
- Google Buys Compute From SpaceX, Broadcom's Outlook, Apple's AI Politics — Stratechery
- A look at Ireland's Bring Your Own Power policy, which requires new data centers to have their own power plants on site or contracts for energy produced nearby (Yusuf Khan/Wall Street Journal) — Techmeme
- Nvidia and LG expand their partnership in AI, robotics, and mobility tech, including building an AI factory and co-designing future data center architectures (Choi Eun-Kyung/The Chosun Daily) — Techmeme
2. Apple's WWDC 2026 keynote stakes its AI credibility on a rebuilt Siri
What happened. Apple's WWDC26 keynote begins today at 10 a.m. Pacific, streaming from Apple's developer site and YouTube. The Verge and Wired both lead their previews with the same expectation: the long-promised Siri overhaul is the marquee item, alongside the annual refresh of iOS, macOS, watchOS, and Apple's other operating systems. Wired notes that, unlike WWDC 2023 — which surfaced the Vision Pro — this year's event is expected to remain software-only.
Why it matters. Apple Intelligence shipped in 2024 with a Siri roadmap the company subsequently had to delay, and the gap between Apple's assistant and the conversational systems offered by OpenAI, Google, and Anthropic has widened through 2025. Stratechery framed today's preview around what Ben Thompson calls "Apple's AI politics" — the increasingly awkward question of how much of the intelligence layer Apple builds itself versus licenses, and how it presents that dependency to customers and regulators. A credible Siri demo would reset the narrative; another deferred promise would not.
Who is affected. The installed base is the largest single audience of any software event — over a billion active iPhones plus Macs, Watches, and iPads. Roughly 34 million registered Apple developers are watching for new APIs, particularly around on-device model access and the App Intents surface that any rebuilt Siri will need to drive. Partners — OpenAI most visibly, potentially Anthropic or Google — have leverage in any framework where Apple ships a router rather than a model.
What to watch next. Whether Siri ships at GA or in a staged developer beta with a fall consumer release; the degree of on-device versus server-side processing Apple commits to; new entitlements for third-party assistant integration; and any concessions to EU Digital Markets Act obligations that have been pending since last year.
Sources:
3. GitHub Copilot's per-token pricing forces the AI industry to confront unit economics
What happened. Microsoft has restructured GitHub Copilot pricing toward per-token billing, replacing flat-rate access for higher tiers of usage. The shift was severe enough that, according to TechCrunch, at least one enterprise customer is internally calling it the "Tokenpocalypse" — a term that surfaced on Reddit and migrated into the latest episode of TechCrunch's Equity podcast with Anthony Ha, Kirsten Korosec, and Sean O'Kane.
Why it matters. The Equity panel's framing is sharper than the pricing change itself. As Ha noted, the ecosystem has been "heavily, heavily subsidized by investor money," and Microsoft — the company best positioned to absorb GPU costs — moving to meter usage is a statement that subsidy economics are running out even at hyperscaler scale. Korosec pointed to Uber's experience inside its own organization: in roughly six weeks the company moved from overshooting its AI budget to capping internal usage. The pattern is becoming common, and it complicates IPO storytelling. O'Kane asked, only half-rhetorically, how many token-related risk factors will appear in Anthropic's eventual S-1.
Who is affected. Enterprise Copilot customers face line-item volatility they could previously plan around. Independent developers and small teams will likely feel the change first as caps tighten. Every AI-native startup that priced against the assumption of declining inference costs now has to model a scenario where vendor pricing moves the other direction. And as Korosec observed, regulators are arriving as the model shifts — President Trump this week signed a narrow executive order giving the government a review window over powerful AI models.
What to watch next. Whether Cursor, Replit, and other Copilot competitors hold flat-rate pricing as a differentiator or follow Microsoft's lead; how Anthropic's eventual S-1 frames token-cost risk; and whether Uber-style internal usage caps become a standard procurement pattern across the Fortune 500.
Sources:
Today's three stories rhyme more than they overlap. Google buying compute from a rocket company, Ireland forcing operators to bring their own power, and Microsoft passing token costs through to developers are all expressions of the same recognition: the inputs to AI — power, silicon, inference cycles — are scarcer and more expensive than the 2023–2024 pricing models implied. Apple's keynote this afternoon will be judged in that context, since whatever Siri becomes has to run on infrastructure with newly visible costs and against competitors whose own economics are about to be tested in public markets.