The Adjacent Brief

TL;DR: Uber is pushing into autonomous vehicle infrastructure and platform diversification as its core ride business faces pressure from AV competitors. Creators are leaving Substack over tax reporting requirements, European AI sovereignty claims are complicated by GPU rental dependency, and the IPO market is reopening to retail investors through private-market vehicles.

Worth Reading

• Nvidia's AI factory thesis is the real valuation story — The market hasn't priced the full capital cost of the AI factory buildout; Nvidia's pitch is that it's the infrastructure layer for all of it.
• Pinterest's production MCP rollout is a workflow story, not a tech story — Embedding brand data directly into creator tools via Model Context Protocol cuts the friction loop between brands and creators — worth watching for what it implies about B2B platform moats.
• Moore's Law broke a bargain, not just a curve — Azeem Azhar on why the end of predictable performance gains changes the economics of every hardware-dependent business.
• AI gig work is replacing waiting tables as the entry-level labor market — The displacement is hitting the bottom of the service economy first.
• Bambu Lab's legal threat killed an open-source project — and the community noticed — OrcaSlicer's shutdown is a case study in how hardware companies weaponize IP to neutralize forks that outperform their own software.
• Europe's battery law was written for chemistry that doesn't exist yet — Regulation optimized for lithium-ion is already misaligned with where battery technology is going.
• The IPO market is reopening — but not the way most people expect — Private companies like OpenAI and SpaceX are entering retail retirement accounts through new investment vehicles before they ever list publicly.

Brand & Growth

The platform pivot is the product

Uber's ride business is large, but the company is now building toward something structurally different — AV data provision, equity stakes in autonomous operators, and platform distribution for third-party fleets. The urgency is real: Waymo is already running paid rides in multiple cities without Uber's involvement, and the window to position as the distribution layer rather than the displaced incumbent is closing. The question is whether Uber ends up as the App Store of autonomous transport or just another operator competing on margin.

Cutting benefits in a tight labor market is a bet on desperation

Deloitte and Zoom both cut paid family leave this spring, which is notable less as a benefits story and more as a signal about employer confidence. These companies are calculating that talent supply has loosened enough to absorb the reduction without meaningful attrition. Whether that calculus holds depends on how long the current hiring environment persists.

Competitive hiring is a collective action trap

Seth Godin's framing of "Red Queen hiring" — organizations hiring because peers are hiring, not because the work demands it — names a dynamic that's been building since the 2021 overhiring wave. No single firm can stop without feeling exposed. The result is bloated headcount, diluted culture, and eventual painful correction. The companies that emerged cleanest from the 2022–2023 layoff cycles were the ones that never joined the race.

Connected World

Sovereignty you rent is not sovereignty

GPU-as-a-Service is becoming the dominant model for European AI infrastructure buildout — and it's hollowing out the policy argument underneath it. European governments point to domestic data centers as evidence of AI independence, but those data centers run on Nvidia H100s manufactured in Taiwan and sold through American chip channels. The compute is European; the supply chain isn't. When the next export control cycle tightens, the sovereignty claim will collide with the hardware reality. The gap between the political framing of AI independence and the actual architecture of AI infrastructure keeps widening.

Who pays for the data centers someone else built

Maryland ratepayers are looking at a roughly $2 billion share of a $22 billion PJM grid upgrade — most of which serves data centers located in Virginia and other states. Data center operators chose those locations for tax and land cost reasons; the grid upgrade required to power them is socialized across the region. Maryland is unlikely to be the only state having this fight.

Autonomous trucks and the quiet rewrite of freight economics

The interstate trade effects of autonomous trucking are getting more serious attention from economists — and the implications extend beyond labor displacement. If long-haul freight costs drop materially, the comparative advantage of proximity to major ports and distribution hubs weakens. That's a geographic reshuffling of manufacturing and warehousing economics, not just a trucking story.

Culture & Signal

Open source is only as open as the company that inspired it allows

Bambu Lab's legal threat forced OrcaSlicer — an open-source fork of Bambu's own slicing software — to shut down. OrcaSlicer was, by most accounts, more capable than Bambu's official software. The move follows a familiar pattern: a hardware company tolerates an open-source community when it's building distribution, then litigates when the fork becomes a competitive threat. The 3D printing community's reaction has been swift and hostile, which matters for Bambu's longer-term developer relations — but probably not more than the legal leverage it just established.

The regulation arrived before the science did

Brussels wrote battery regulations optimized for lithium-ion chemistry at precisely the moment when solid-state and sodium-ion alternatives are maturing. The result is a compliance framework that advantages incumbents running the old chemistry and creates unnecessary friction for next-generation producers. Technology-specific regulation codifies the present and taxes the future.

The research paper is forking

The question of whether AI kills the research paper is less interesting than what replaces it. The static PDF — published once, cited forever — was always a poor fit for fields where data updates continuously. AI-powered living documents that auto-update with new findings could serve science better than the current format, but they create new problems for attribution, priority, and peer review. The form is under pressure; the institutions built around it are under more.

The New Consumer

Substack created a tax problem and called it a platform

Creators are moving from Substack to Ghost, Beehiiv, and other platforms — and the trigger, per The Verge, is Substack's 1099 reporting requirement. Once a creator clears the income threshold, Substack issues tax forms that other platforms don't, creating a visible cost differential that's easy for a competitor to exploit. Substack's 10% revenue cut was always a bet that network and discovery benefits outweighed the fee; once tax friction enters the equation, that math gets harder to defend.

Ghost's ongoing effort to position itself as the professional publisher's alternative keeps gaining surface area as Substack accumulates friction. The migration isn't a mass event yet, but the direction is clear enough that platform retention teams should be paying attention.

Performative knowledge is the new status signal for Gen Z

Casey Lewis identifies a specific behavior worth tracking: Gen Z is packaging intellectual content — books read, ideas held, concepts understood — as social performance rather than actual knowledge acquisition. The "attention span panic" framing misses what's actually happening. Young people can focus; the incentive structure rewards demonstrating knowledge more than acquiring it. For brands and educators trying to reach this cohort, the distinction matters.

Commerce Rewired

The IPO window reopened in a different shape

The new IPO economy is private companies like OpenAI and SpaceX reaching retail investors through 401(k) vehicles and private market funds before they ever list publicly. The democratization framing is real but incomplete: retail investors are getting access, but at later stages and higher valuations than institutional investors paid. Whether this is good for returns remains to be seen. When the best price appreciation happens in pre-IPO rounds that retail can't touch, opening the door to Series D exposure is a different proposition than it sounds.

Machines & Minds

Local inference is becoming a real cost argument, not just a privacy one

The Register's case for local LLMs has shifted: the argument used to be about data sovereignty and privacy. Now it's about cost. Running inference on a laptop rather than in the cloud cuts per-query costs to near zero for workloads that don't require frontier-model capability. For enterprises running high-volume, lower-complexity tasks — document summarization, classification, internal Q&A — local deployment is increasingly defensible on economics alone, not just on compliance grounds.

The gap between agent demos and agent products is a judgment problem

The "judge layer" concept — a secondary AI evaluation step that reviews agent outputs before they execute — is emerging as the practical answer to why most agent demos don't survive contact with production environments. The failures are subtle reasoning errors that compound across multi-step workflows. Adding human judgment back into the loop defeats much of the efficiency argument for agents; building an automated judge layer that catches failures without adding latency is the actual engineering problem. That's where the serious agent development money is going — into the reliability infrastructure around capable base models.

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