The Adjacent Brief

TL;DR: OpenAI is renegotiating its cloud dependency on Microsoft, routing workloads to Oracle and potentially Amazon as its infrastructure needs outgrow any single provider. Separately, more than half of Americans told Gallup their personal finances are deteriorating — a Fairphone breakout year and a Quartz look at manufactured workplace urgency sit in that same economic mood.

Worth Reading

• Social media cost Americans $2.1 billion in scam losses last year — FTC data puts a number on the liability platforms have long avoided; nearly 30% of all reported scam losses now route through social.
• Where the sweetest margins live in Jensen's five-layer AI stack — NVIDIA's CEO pitched a specific profit architecture; this reads which layer actually extracts the most value.
• How AI efficiency is quietly breaking the interactions that build strong teams — Automation removes the informal collisions that generate trust and creative problem-solving; worth reading alongside today's urgency-trap piece.
• AI hardware, Meta's display ambitions, and what VR/AR actually becomes — Ben Thompson on how Meta's hardware strategy is more coherent than critics credit — and what that means for the category.
• CEO dinner April 2026: the ghost in the shopping cart — Executives in a room debating agent adoption, echo chambers, and whether AI purchasing agents hollow out brand loyalty.
• Leading through AI hype — Practical framing for executives who need to validate AI investments without getting pulled into vendor narratives.
• The opportunity gap inside strength training — The most popular workout category still has no real data layer. A structural gap that fitness and wearables companies keep underexploiting.

Connected World

The physical cost of AI infrastructure is meeting organized resistance

Rural communities are drawing a line. According to a Financial Times analysis of Pew data, 67% of planned US data centers are sited in rural areas — land is cheap, zoning is loose, power is available — but 87% of those communities oppose adding AI infrastructure. The gap between where capital wants to build and where people want to live with the consequences is wide, and economic-development promises are no longer closing it.

Off-planet is not a metaphor

Rather than fight local resistance, Meta signed a contract with Overview Energy to pilot space-based solar power beamed to its data centers after dark — electricity generation that produces no local emissions, consumes no local land, and generates no local opposition. It is early-stage and expensive, but if terrestrial siting is becoming a constraint, orbit is the workaround. The deal is small; the direction of travel is worth watching.

A Numlock News dispatch today flagged the piracy and infrastructure beats running alongside these stories — a reminder that the physical layer of the internet remains contested even as the software layer draws most of the attention.

The New Consumer

Declining finances, declining upgrade cycles

Gallup's latest survey found more than half of Americans say their personal finances are getting worse, per Axios — a reading above the levels recorded during the 2022 inflation spike. That pessimism has a correlate in hardware: global smartphone sales are down, but Fairphone is posting a breakout year, per Android Central. The mainstream market contracts while a values-differentiated niche expands. Consumers who still buy in a down market are often buying on criteria other than price-performance — modularity, repairability, supply chain ethics — which gives that segment a durable edge commodity hardware can't undercut.

Manufactured urgency as a productivity tax

The Quartz piece on the urgency trap in modern workplaces lands squarely against that economic backdrop. Organizations structurally reward speed and constant responsiveness — fast replies, packed calendars, compressed timelines — in ways unconnected to actual business deadlines. That manufactured urgency consumes cognitive capacity that could go toward higher-leverage work. For leaders managing through an economic squeeze, the piece offers a concrete frame: audit the urgency, not just the headcount.

Machines & Minds

The cloud divorce is starting, bill still unclear

OpenAI is diversifying away from Microsoft's Azure, routing workloads to Oracle and exploring Amazon, per Axios. The relationship isn't ending — Microsoft remains a massive investor and distribution partner — but OpenAI's compute needs have grown large enough that single-vendor dependency carries real risk. This matters less as a drama between two companies and more as a precedent: when an AI lab reaches scale, it negotiates the cloud relationship rather than accepting it. Smaller enterprise buyers don't have that leverage, which is exactly the problem surfaced in the piece on vendor lock-in biting enterprise AI buyers in The Register — switching AI vendors takes longer than a week and costs more than procurement teams were told. Enterprises that signed three-year commitments to a single model provider are discovering what OpenAI figured out: architecture independence is a business asset.

AI deflation is the India services story no one wanted to tell

The Register's second piece is the sharper one: India's largest technology services firms — TCS, Infosys, Wipro, HCL — are seeing downward revenue pressure from AI as clients automate work those firms used to staff. The term the piece coins is "AI deflation" — clients paying less for the same output because AI handles more of the execution. Workforce reductions are still contained, but revenue compression is already showing up in quarterly guidance. For anyone selling labor-wrapped services into enterprises, this is the number to watch. AI's economic impact is hitting revenue lines before it hits headcount, which means the pain lands on suppliers before it becomes visible in employment data.

One-third of the web is already AI-generated

A 404 Media analysis of Internet Archive data finds that by mid-2025, roughly 35% of new websites published since ChatGPT's November 2022 launch were AI-generated or AI-assisted. Search indexes fill with undifferentiated machine output, and the incentive to produce genuinely useful original content weakens as AI-generated content gains equivalent surface area. Publishers, SEO practitioners, and anyone whose business model depends on organic discovery should treat this as a structural shift in the information environment, not a temporary anomaly.

Operating systems are rebuilding for AI workloads from the substrate up

The Pragmatic Engineer's look at how Ubuntu and Linux are architecting specifically for AI workload execution is a useful infrastructure signal. The OS layer is not just running AI tools — it's being redesigned around them. Memory management, security sandboxing, and developer environment tooling are all being reconsidered for inference and training contexts. For enterprise infrastructure teams, the practical question is planning horizon: the baseline OS assumptions under your AI stack will change, and "Part 1" signals this is a multi-installment examination worth following.

Commerce Rewired

Monopoly logic as the default operating environment

Matt Stoller's Monopoly Round-Up uses the Chinese finger trap uses the arm-wrestling metaphor as the structural frame for American commerce and politics: the harder incumbents pull against antitrust pressure, the tighter the constraint becomes. The piece is worth reading for the stock market analysis beneath the metaphor — Stoller argues that concentrated market structures are making capital allocation less responsive to actual economic conditions. For strategists, the practical implication is that regulatory risk is priced too low in sectors where market concentration has become the default assumption.

A piece in Semafor today ("Heir apparent") touches adjacent ground on succession dynamics in major institutions — a reminder that concentrated power at the top of organizations creates its own version of the same structural fragility.

13 articles across 4 themes · 14 sources · Powered by Folo + Claude