The Adjacent Brief

TL;DR: Opponents blocked or delayed 75 US data center projects worth roughly $130 billion in Q1 2026, as community resistance to AI infrastructure buildout spreads to 49 states. Separately, KPMG pulled a published report on agentic AI after named organizations — including UBS and the NHS — said the case studies attributed to them were fabricated.

Worth Reading

• India's AI ambitions hit a wall when Anthropic's models go dark — Export controls on Anthropic reveal how much of India's AI strategy is built on infrastructure it doesn't control.
• Europe calls Anthropic's model suspension a "wake-up call" on US dependency — When a US company can suspend access to sovereign digital infrastructure overnight, "AI strategy" means something different on each side of the Atlantic.
• How to run a news company when AI slop is everywhere (paywall) — NBC News chief Cesar Conde's bet: institutional brand trust outlasts the content flood. Worth stress-testing that assumption.
• Hyperscalers cut buybacks to fund AI capex — and what that does to markets (paywall) — Alphabet's ~$85B in planned equity offerings is a structural shift in where AI investment actually shows up on the balance sheet.
• Apple's AI photo tools: mostly working, occasionally not — A hands-on from The Verge worth reading for the texture: the tools are competent, not magical, which is exactly where mass-market AI features need to land.
• The passive cooling vest that actually works without batteries — A graduate design project using phase-change materials to manage heat stress for outdoor workers — a reminder that not every hard problem needs a chip.
• Solid-state battery aircraft completes first crewed flight — Helios Horizon's milestone matters less as an aviation story than as a data point on solid-state energy density finally crossing a threshold.

Connected World

$130 billion worth of data centers didn't get built last quarter

The scale of the resistance is hard to minimize: opponents blocked or delayed at least 75 US data center projects in Q1 2026, representing roughly $130 billion in planned investment. Organized opposition groups have doubled to 833 across 49 states. Opposition has spread well beyond data-center-dense Virginia — it's a distributed, replicating pattern of community resistance the industry has no playbook for.

The timing is pointed. Hyperscalers are simultaneously pulling back on share buybacks to fund AI infrastructure — Alphabet alone has signaled roughly $85 billion in equity offerings tied to AI capex. The capital commitment has never been larger, but the physical buildout is running into permit denials, zoning challenges, and organized local opposition at a pace the financial models didn't price in. When the constraint on AI infrastructure turns out to be community relations rather than capital availability, that's a different kind of problem than the industry has managed before.

The physics of not-quite-magic

Two hardware stories this week sit at very different points on the readiness curve. A graduate designer's passive cooling vest for outdoor workers uses phase-change materials — no battery, no electronics, no power source required — to manage heat stress in jobs that kill people during summer months. The solution works because the constraint it's solving (no access to power, harsh environments, cost sensitivity) is real and legible. Compare that to Helios Horizon's first crewed solid-state battery flight, a genuine technical milestone — but the distance between "first crewed flight" and "commercially viable aircraft" is measured in decades, not product cycles. The vest ships. The plane is a proof of concept.

Culture & Signal

Platforms moderate when senators call, not before

Spotify's removal of 57,000 fake podcast episodes promoting illegal drugs — along with 3,500 associated accounts — happened after Senate pressure, not before. The content had been live long enough to build a meaningful distribution network. This is the same pattern across every major platform: detection systems exist, enforcement happens when the political cost of inaction exceeds the operational cost of acting. The question worth asking isn't whether Spotify can find this content — clearly it can — but why the default position is to wait for a senator to make a phone call.

KPMG's AI report contained things that weren't true

KPMG pulled its published report on agentic AI after UBS, the NHS, and other named organizations said the case studies attributed to them were fabricated. A Big Four firm publishing work built on hallucinated citations is a process failure, not a technical one. Someone signed off on this report. The technology produced plausible-sounding text; the humans in the loop didn't verify it. That's a workflow problem, and the reputational damage lands on KPMG regardless of which piece of software drafted the copy.

Both stories share the same structure. Whether it's drug-spam podcasts or fabricated consulting case studies, the infrastructure for producing and distributing bad content at scale outruns the infrastructure for catching it. The KPMG case is notable because it involves a firm whose entire value proposition is institutional credibility.

The AI job cut number is getting harder to contextualize

Challenger, Gray & Christmas data shows employers cited AI as the reason for roughly 88,000 US job cuts through May 2026 — up from about 54,000 in all of 2025. That's a real acceleration in a single attribution category, and it's the third consecutive month AI topped the list of stated reasons. The methodological caveat matters: "employer cited AI" is a disclosed rationale, not a causal audit. Companies have incentives to frame cuts as modernization rather than demand weakness. But even discounting for strategic framing, 88,000 in five months versus 54,000 in a full prior year is a number that deserves to be taken seriously.

The New Consumer

The second screen has no home

The 2026 FIFA World Cup is the first major global sports event to unfold without a clear real-time social venue. Twitter — now X — no longer functions as the communal second screen it was during every World Cup from 2010 to 2022, and nothing has replaced it. The Verge's Andrew Webster puts the problem plainly: algorithmic timeline disruption, platform fragmentation, and the departure of sports media personalities who made the experience work have scattered fans across Bluesky, Threads, WhatsApp group chats, and Discord servers — none of which have the critical mass to replicate what Twitter did at its peak.

This is a genuine gap in the attention economy, not nostalgia for a platform. Real-time shared sports experience was one of the few remaining reasons to be on a social platform at a specific moment. Whoever solves the live sports second-screen problem — whether that's a platform feature, a standalone product, or a broadcaster-owned experience — is looking at a use case with demonstrated demand and no current owner. It's a product gap, not a cultural one.

Commerce Rewired

Buybacks out, capex in — and the market is absorbing it

The structural story in how the AI boom is reshaping US equity markets (paywall) centers on what the hyperscalers are giving up to fund the headline capex numbers. Share buyback programs, the primary mechanism for returning capital to shareholders and supporting equity valuations for the past decade, are being scaled back as Alphabet, Microsoft, and peers redirect cash toward infrastructure. Alphabet's planned ~$85 billion in equity offerings tied to AI is the clearest example: the company is effectively asking public markets to fund the buildout it believes will generate future returns.

The bet is legible — own the infrastructure layer before the market prices it in. The risk is also legible: if the AI revenue models that justify the capex take longer to materialize than capital markets expect, the companies that cut buybacks to fund buildout will face a different kind of shareholder pressure. This is a 2028 story priced on 2026 commitments. But the capital decisions being made now set the parameters.

Machines & Minds

Cheap intelligence isn't the same as usable intelligence

Nate's Substack puts it plainly: your company is about to get cheap intelligence, and that is not the same as being able to use it. The bottleneck isn't access to the model — it's organizational capacity to deploy it against real work. Most companies are about to find themselves with abundant raw capability and scarce ability to apply it. That gap, between access and operational use, is where the next wave of enterprise software competition will play out.

This connects directly to why the KPMG report failure matters beyond the embarrassment. The firm had access to capable AI tools; what it lacked was a verification workflow that matched the output volume those tools generate. Cheap intelligence deployed without the scaffolding to validate its outputs produces confident-sounding errors at scale.

Mozilla is trying to build what the KPMG failure demonstrates is missing

The Mozilla Data Collective's attempt to build AI's data economy around consent and trust is, in concept, an answer to a real structural problem: the training data market has no provenance layer, which means the outputs it produces have no auditable lineage. Whether Mozilla can execute that vision as a commercial proposition — rather than an advocacy project — is the open question. The organization has a credibility advantage few others can claim in this space. It does not have a revenue model that obviously sustains the infrastructure the project would require. Watch the funding, not the mission statement.

11 articles across 5 themes · 9 sources · Powered by Folo + Claude