// The Adjacent Brief

TL;DR: Microsoft withdrew from a £2bn data center project in Loughton, UK, after the local grid could not supply enough power, pushing the site's timeline into the 2030s. In China, more than 150 apps blocked Xiaomi's AI assistant from operating inside them. The Verge reorganized its audience team to prepare for a future with little or no Google search traffic.

Worth Reading

Brand & Growth

Owned channels can't carry the load alone

The Verge is restructuring its audience team around "Google Zero", the scenario in which search referrals fall to near nothing as AI answers absorb the click. A publisher with one of tech media's strongest brands is planning for the floor rather than hoping for a recovery. Any traffic chart is less telling than that decision. The practical move is to shift audience work from SEO toward direct relationships, such as newsletters, apps and community, where the publisher owns the connection to the reader.

The same pressure applies to brands, and The Landing Pad states it bluntly: brands that produce all their own content are going to fail. In-house teams can produce volume. On their own, they can't produce the credibility and reach that creators and outside partners carry. For a CMO, the budget question is changing from "how much content can we make?" to "whose audience are we borrowing, and on what terms?" The Verge is preparing for the moment it can no longer borrow Google's audience. Brands should look at their own dependencies the same way.

Embedding AI is table stakes

Stripe's survey of SaaS platform leaders finds them feeling competitive pressure from AI and planning to embed it across their products. That's the expected answer, and it is also the problem. If every platform adds AI features, those features stop differentiating anyone. What matters is whether AI changes how the product gets paid for, through usage-based pricing or outcome-based fees, or becomes a cost center that compresses margins. Stripe processes the payments, so its interest in pricing-model change is not incidental. Still, it's the right question for any product lead to ask before adding another copilot.

Connected World

Power, not compute, sets the build schedule

Klement on Investing makes the counterintuitive case that data center demand is speeding up the energy transition: utilities facing hyperscaler load are building generation faster than historical rates. The same demand runs into the constraint visible in the Loughton withdrawal above. Microsoft left over neither cost nor chip supply. The grid connection wouldn't arrive for years. Both things can be true at once. Capital is pulling new generation forward, but transmission and interconnection queues move on regulatory time rather than venture time. For anyone deciding where to site AI infrastructure, a firm grid connection now counts for more than cheap land or tax breaks.

The materials bottleneck has a domestic workaround

Canary Media reports that a cleaner fix for America's aluminum shortage is sitting in landfills. Recycled aluminum takes roughly 5% of the energy needed to produce new metal. With tariffs raising the cost of imported primary aluminum, recovering scrap looks more like a supply strategy than a sustainability project. Geopolitical friction has been turning circular-economy ideas into procurement decisions elsewhere too. The open question is whether collection and sorting infrastructure can scale fast enough to make a difference.

In a longer-horizon item, Nature reports that light and mirrors are bringing large quantum computers closer, with optical connections linking modules that can't scale individually. It's still a lab result. It follows the same logic as the data center story: the hard problem is increasingly the connections between units, rather than the units themselves.

The New Consumer

Apps are refusing to let agents in

More than 150 Chinese apps have blocked Xiaomi's AI assistant from operating on their platforms. The usual pitch for agentic commerce, "your assistant books, orders and pays for you," assumes that the apps where those transactions happen will cooperate. Chinese super-apps have answered that assumption: they won't hand their customer relationship and transaction data to a phone maker's agent. Western retailers and marketplaces will face the same choice when OS-level assistants start acting on users' behalf. Brands should decide now whether an agent arriving at their storefront counts as a customer or a competitor. The Chinese answer, so far, is a competitor.

Bodies and postcodes change faster than surveys can track

Two stories here are grounded in behavior rather than stated preference. In Australia, city dwellers are moving to tropical regional areas at rates closer to the pandemic peak than to pre-COVID levels. Predictions that remote-era moves would reverse aren't matching what people are doing. For retailers, health systems and property developers in regional Queensland, this demand is starting to look durable rather than temporary.

Vogue asks whether GLP-1s are changing how people dress and points to sheer and body-revealing silhouettes in SS27 collections. Designers are responding to changing customer bodies and buying patterns, not to polls about intentions. The commercial implications extend past the runway: sizing curves, fit returns and replacement-wardrobe purchases are all affected. Apparel brands that still plan inventory on pre-GLP-1 size distributions are taking on markdown risk they haven't priced in.

Commerce Rewired

AI's bill lands on other people's revenue lines

The Register argues that AI's huge appetite for capital is matched only by its ability to choke off existing revenue streams. The industry is raising money faster than it is earning it, while eroding the referral and advertising businesses that publishers and platforms depend on. The Verge's Google Zero planning is what that erosion looks like from the inside. The money pays for compute, and it also moves value from the open web into answer engines that don't yet earn back what they displace. Until AI revenue catches up with AI capital, someone else's business model absorbs the gap.

Culture & Signal

Regulatory middlemen get rich in the gaps

A STAT investigation finds that the nonprofit arbitrators created under the No Surprises Act are operating like "mini gods" and getting rich doing it. In one case, the arbitrator iMPROve Health went from financial distress to holding major power over healthcare payment disputes, all through a government contract. The law was meant to protect patients from surprise bills. It also created a dispute-resolution layer with high volume, little scrutiny and fees that reward throughput. The pattern extends well beyond healthcare. Proposed AI audit regimes, content-provenance certification and similar schemes all depend on new intermediaries. Before the rules are written, it's worth asking who will collect fees as the referee.

Machines & Minds

Jobs don't break down as neatly as the spreadsheet suggests

Josh Bersin examines "task intelligence", the practice of decomposing every job into discrete tasks that can be mapped to AI. HR and transformation teams are pursuing this as a planning tool, and task-level mapping does beat vague claims that "AI will change jobs." The weakness is that jobs include more than a list of tasks. Judgment, context switching and accountability sit between the tasks, and those parts resist decomposition. Leaders should use task maps to find automation candidates, and should not assume a role's remainder can be cut once its tasks are listed.

Buyers want proof on their own workflows

A LessWrong post on "holy shit" graphs and how to get more of them looks at what produces results so stark they settle an argument on sight. The same applies to enterprise AI. Benchmarks rarely move a budget. A before-and-after chart of one team's actual workflow usually does. Bersin's task maps and that kind of chart work best together: map the tasks, pick the one with the clearest measurable output, and produce a result that needs no explanation. Vendors that can generate that evidence inside a customer's own operations will win renewals that slide decks can't.

13 articles across 6 themes · 13 sources · Powered by Folo + Claude