> ## Content Index
> Fetch the complete content index at: https://adjacent.media/llms.txt
> Use this file to discover other available public pages before exploring further.

# The Adjacent Brief — October 16, 2026
- URL: https://adjacent.media/briefs/2026-10-16/
- Published: 2026-10-16T14:15:02.000Z
- Updated: 2026-10-16T14:15:03.000Z
- Description: Fortune published capex-versus-revenue figures arguing the AI investment cycle is nearing its peak, and Noahpinion compared the buildout to the 19th-century railroad boom.
- Author: Jonathan Greene
- Tags: #brief

**TL;DR:** Fortune published capex-versus-revenue figures arguing the AI investment cycle is nearing its peak, and Noahpinion compared the buildout to the 19th-century railroad boom. WIRED reported that safety guardrails on AI models can be stripped in minutes to produce extremist content, ASML said it cannot meet chip-tool demand in 2027 or 2028, and a single IoT flaw let attackers compromise hundreds of French supermarket refrigeration systems.

## Worth Reading

- [ASML's sold-out order book caps the AI buildout through 2028](https://thenextweb.com/news/asml-q3-capacity-2030-orders?ref=adjacent.media) — With orders reaching into 2029–2030, the binding constraint on new fabs is a single Dutch supplier. Capital is no limit.
- [Data center backup batteries could become grid assets instead of dead weight](https://www.canarymedia.com/articles/data-centers/on-energy-ups-battery-grid-interruption?ref=adjacent.media) — Batteries that stabilize the grid give operators a stronger argument in interconnection queues and local permitting fights than any sustainability report.
- [China's market for selling your face to AI pays almost nothing](https://www.sixthtone.com/news/1019126/Chinese%20Users%20Find%20Little%20Money%20in%20'Selling'%20Their%20Faces%20to%20AI?ref=adjacent.media) — Likeness licensing was pitched as creator income. Platforms collected the faces and kept the margin.
- [MyChart lets you opt out of sharing your health data, but the friction does the work](https://www.statnews.com/2026/10/15/health-data-privacy-medical-record-opt-out-barriers-investigated/?utm%5Fcampaign=rss&ref=adjacent.media) — A first-person account of an opt-out that exists mainly to be cited in compliance reviews.
- [Fairbuds 2 makes the battery, not the earbud, the replaceable part](https://www.yankodesign.com/2026/10/15/fairbuds-2-earbuds-let-you-swap-worn-batteries-instead-of-buying-a-new-pair/?utm%5Fsource=rss&utm%5Fmedium=rss&utm%5Fcampaign=fairbuds-2-earbuds-let-you-swap-worn-batteries-instead-of-buying-a-new-pair) — Planned obsolescence in audio has always lived in the battery. A swappable cell is a direct shot at the upgrade cycle.

## Brand & Growth

**Human signoff is the bottleneck**

McKinsey's framing of AI-speed analysis colliding with human-speed judgment names the operational problem most strategy teams are running into. Generating options costs almost nothing, [while deciding among them costs the same executive hours it always did](https://www.mckinsey.com/capabilities/strategy-and-corporate-finance/our-insights/the-judgment-bottleneck-balancing-ai-and-human-speeds-in-strategy?ref=adjacent.media#/). Companies that bolt AI onto existing planning cycles end up with more decks and no faster decisions. McKinsey's proposed redesign centers on governance: who is allowed to decide what, at what threshold, without escalation.

Media buying is where this gets tested first, because the feedback loop is measured in hours. Industry voices at Beet argue that [agents will execute buys while humans keep control of the agenda](https://www.beet.tv/2026/10/ai-agents-will-reshape-media-buying-but-humans-still-set-the-agenda.html?ref=adjacent.media). Read it with some skepticism, since "humans still set the agenda" is also what agency holding companies need clients to believe to justify their fees. The useful question for a CMO is narrower: which line items in the media budget are execution, and which are judgment? Agent pricing will compress agency margins on execution.

PSFK's look at [how work with AI will be structured and paid in 2027](https://open.substack.com/pub/retailinnovationweek/p/how-youll-work-with-ai-in-2027-and) raises the compensation question the other pieces skip. If output scales with agents but judgment does not, pay will eventually follow judgment. Org charts built on headcount-as-status have no slot for that yet.

## Commerce Rewired

**Chinese automakers are competing on product, not just price**

Chinese brands [moved from the margins to center stage at the Paris Motor Show](https://www.nytimes.com/2026/10/15/business/chinese-evs-europe-paris-motor-show.html?ref=adjacent.media) (paywall), and their European expansion is continuing despite EU countervailing duties. Tariffs raised the price floor but did not close the feature gap. Trade policy bought European incumbents time, and it is not obvious they are using it. Brand leaders at legacy automakers should plan for a market where Chinese EVs are a permanent tier.

**Intermediaries tax the transaction they claim to optimize**

Matt Stoller's BIG newsletter makes the case that [Expedia uses its marketplace position to push hotel prices up](https://www.thebignewsletter.com/p/how-expedia-rigs-travel-markets-and?ref=adjacent.media). The mechanism is familiar from Amazon's seller fees: hotels absorb commissions, then build them back into rates across every channel. The hotel's direct-booking price never ends up undercutting the platform. For hospitality brands, direct-booking programs only work if rate parity rules let them be cheaper.

The opposite dynamic is playing out in AI compute. Tomasz Tunguz documents [falling intelligence-per-dollar as chip and model pricing compete downward](https://tomtunguz.com/intelligence-per-dollar-2026/?ref=adjacent.media). When frontier capability converges, pricing becomes the competition, and buyers gain leverage they did not have a year ago. Procurement teams locked into single-vendor AI contracts signed in 2025 should be renegotiating.

## Connected World

**The railroad analogy cuts both ways**

Noah Smith compares the AI buildout to the railroad era and its [boom, overbuild, and lasting infrastructure](https://www.noahpinion.blog/p/ai-vs-the-railroads?ref=adjacent.media). The comparison is usually offered as reassurance: investors lost money, but society kept the track. The less comfortable half is that railroad financing collapsed when debt came due before traffic arrived. AI data centers carry a similar maturity-mismatch risk. GPUs depreciate in years, not decades, so the "we keep the infrastructure" defense is weaker than it was for steel rail.

**One vulnerability, hundreds of stores**

Ars Technica traced how [a refrigeration hack that cost millions and sparked fires](https://arstechnica.com/security/2026/10/fridge-hack-costing-millions-and-sparking-fires-stemmed-from-poor-security/?ref=adjacent.media) came down to basic security failures. More than 700 supermarket systems in France were compromised through a single flaw. Grocery cold chains are now networked infrastructure but are procured like appliances, often from vendors with no security budget. Any retailer running connected HVAC, refrigeration, or lighting should ask its facilities team, not just IT, who owns the patch cycle.

## Culture & Signal

**Retailers are making the political call for readers**

NME reports that ['Normal People' is being pulled from UK book shops](https://www.nme.com/news/tv/normal-people-removed-uk-book-sally-rooney-pro-palestine-support-action-3973182?utm%5Fsource=rss&utm%5Fmedium=rss&utm%5Fcampaign=normal-people-removed-uk-book-sally-rooney-pro-palestine-support-action) over Sally Rooney's pro-Palestine activism. Booksellers hold the decision, and publishers and readers do not. Retail shelf space becomes a political instrument in a way it rarely is for mainstream literary fiction. Brands that sell culture should expect to be asked which side their inventory is on.

**Adding capacity is not the same as adding outcomes**

Researchers writing in The Conversation found that in one California region, [more rehab centers coincided with more overdoses, not fewer](https://theconversation.com/more-rehab-centers-should-mean-fewer-overdoses-in-one-california-region-we-found-the-opposite-285569?ref=adjacent.media). Supply metrics like bed counts and facility openings are easy to fund and report. Treatment quality is harder to measure. Any program measured by inputs will eventually optimize for inputs.

The New York Review of Books on [the dream of American AI primacy](https://www.nybooks.com/articles/2026/11/05/the-dream-of-ai-primacy/?ref=adjacent.media) shows a similar gap between appearance and substance. The piece examines a policy posture built on assumed dominance over China, an assumption the competitive evidence does not clearly support. Policy built on a narrative tends to fund the narrative.

## The New Consumer

**Opt-out exists on paper**

The Markup's [guide to opting out of health data sharing](https://themarkup.org/privacy/2026/10/15/tips-how-to-opt-out-of-data-sharing?ref=adjacent.media) is useful mainly as evidence of how much effort it takes. Health apps comply with privacy rules by technically offering the choice, then make exercising it a multi-step chore. Regulators are starting to treat that kind of friction as a design choice rather than an accident. Health and wellness brands that rely on obscured opt-outs carry legal exposure they may not have priced.

**Labels set the floor and leave the fix undone**

YouTube Music is [adding AI labels as synthetic tracks flood streaming](https://www.androidcentral.com/apps-software/youtube/youtube-music-adds-ai-labels-as-synthetic-songs-flood-streaming?ref=adjacent.media). Labeling protects the platform's credibility more than it protects listeners, who mostly play what the algorithm queues. What matters is whether labeled tracks get demoted in recommendations and royalty pools. Until YouTube says so, treat this as a disclosure.

**Timing a career against a bubble**

Klement on Investing looks at [how asset bubbles affect lifetime earnings](https://klementoninvesting.substack.com/p/the-impact-of-bubbles-on-lifetime). The question sharpens next to today's AI-capex warnings. Workers who start careers or pivot into sectors at peak valuation tend to carry the downside long after the market recovers. For talent leaders: be cautious about hiring sprees justified by valuation rather than revenue.

## Machines & Minds

**The capex math is meeting the revenue line**

Fortune lays out [the numbers suggesting the AI boom is near its top](https://fortune.com/2026/10/14/ai-capex-revenue-top-ai-boom/?ref=adjacent.media): spending on chips and data centers running well ahead of the revenue those assets generate. Fortune is not calling a collapse. The question has moved from "can it be built" to "who pays for it." Enterprise buyers should expect vendors to push harder on multi-year commitments to backfill that gap, and should negotiate accordingly.

**Guardrails are a setting that can be changed**

WIRED reports that [AI models can be easily modified to assist terrorist and extremist activity](https://www.wired.com/story/ai-models-can-be-easily-modified-to-assist-terrorist-and-extremist-activity/?ref=adjacent.media), with guardrails bypassed in minutes. Zvi Mowshowitz argues that [the safety culture at frontier labs is broken](https://thezvi.wordpress.com/2026/10/14/the-safety-culture-at-the-labs-is-broken/?ref=adjacent.media), following criticism this week that outside safety reviews function as cover rather than governance. Together, they describe safety commitments that hold only until someone tries to break them. Companies deploying these models should not substitute vendor safety claims for their own red-teaming.

**Your archive is part of the prompt**

Models that can draw on a person's public corpus can personalize, and identify, far beyond what users assume. Executives with long public writing trails are the easiest people to profile and impersonate.

**The experts feel it first**

Derek Thompson writes on [why mathematicians are mourning as AI moves into their field](https://www.derekthompson.org/p/i-know-why-the-mathematicians-weep?ref=adjacent.media). Mathematics is a useful early case because results can be checked: either the proof holds or it doesn't. The grief concerns what it means to be the person who knows, more than jobs. That reaction will reach other expert professions, including law, medicine, and strategy, as machines become able to verify their work.

---

*20 articles across 6 themes · 19 sources · Powered by Folo + Claude*