> ## 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 — September 24, 2026
- URL: https://adjacent.media/briefs/2026-09-24/
- Published: 2026-09-24T14:15:03.000Z
- Updated: 2026-09-24T14:15:04.000Z
- Description: Meta’s Muse assistant took the #1 spot on the U.S. free iPhone chart, passing ChatGPT, the same week Google shelved its own consumer assistant ambitions.
- Author: Jonathan Greene
- Tags: #brief

**TL;DR:** Meta's Muse assistant took the #1 spot on the U.S. free iPhone chart, passing ChatGPT, the same week Google shelved its own consumer assistant ambitions. Financial Times reporting put two pieces of the AI supply chain under scrutiny: ByteDance's 75% share of Nscale's 2025 revenue routed through Norway, and Beijing's inventory count of Broadcom switches in state data centers.

## Worth Reading

- [Beijing is counting Broadcom switches the way you'd count inventory before a recall](https://www.ft.com/content/b97dde39-7813-4a51-8176-3699e2a468b1?ref=adjacent.media) (paywall) — SASAC's survey is the procedural step that precedes a substitution mandate.
- [Xbox bought its way to a billion users and shrank the thing people bought it for](https://www.theverge.com/games/998758/xbox-activision-blizzard?ref=adjacent.media) — Studio closures after the largest acquisition in gaming history is a scale strategy eating its own supply.
- [Open-weight models closed the gap, which resets what "frontier" is worth](https://thesignal.substack.com/p/the-open-models-have-caught-up) — If capability is commoditizing, the premium moves to distribution and data while model quality loses its pricing power.
- [Apple put ads in Settings, the one screen nobody visits voluntarily](https://mjtsai.com/blog/2026/09/22/persistent-ads-in-ios-settings/?ref=adjacent.media) — The privacy-premium brand is monetizing system chrome. Small move, meaningful precedent.
- [Your AI productivity gains are creating more work, not less](https://www.forrester.com/blogs/your-ai-productivity-gains-are-creating-more-work/?ref=adjacent.media) — Forrester's finding lands the same week Shopify's CEO told employees the same thing out loud.

## Brand & Growth

**Distribution beat capability, again**

John Gruber's link to Spyglass asks the question every Google product review should have asked two years ago: [Google had the models, the assistant history, the Android install base, and built none of it](https://spyglass.org/why-didnt-google-build-muse/?ref=adjacent.media). Meta shipped a consumer agent app and took the top of the iPhone chart. Google's assistant ambitions cannibalize the query volume that funds the company; Meta has no search revenue to protect. For anyone reading competitive positioning: the incumbent's biggest asset is often the thing that stops it from moving. Look at what a competitor *can't* afford to lose before assuming their scale is the advantage.

**Buying scale and shrinking the product are the same strategy running out of road**

Xbox closed studios and cut jobs while [publicly targeting "a billion people"](https://www.theverge.com/games/998758/xbox-activision-blizzard?ref=adjacent.media), retreating to Halo and a handful of owned franchises. The Activision Blizzard deal was supposed to be a content engine; what it produced was consolidation around proven IP and less capacity to make anything new. Microsoft is effectively running Xbox as a licensing and services business with a hardware legacy attached. That's a defensible model, but it's not the one the acquisition was sold on, and brands that buy distribution while cutting what's being distributed tend to discover the audience was there for the thing.

**Creator TV wants upfront money, which means it wants upfront measurement**

Spotter is arguing that [creator-made video deserves a seat at the upfront table](https://www.beet.tv/2026/09/why-spotter-says-creator-tv-deserves-a-seat-at-the-upfront-table.html?ref=adjacent.media) alongside broadcast and streaming inventory. The pitch is reasonable on audience math and harder on the part buyers actually care about: guaranteed reach, brand-safety controls, and comparable currency. The same ad industry now naming consumer trust, not targeting precision, as the binding constraint on effectiveness is being asked to buy inventory whose primary asset *is* trust — trust held by an individual who can post something unfortunate on a Tuesday.

## Commerce Rewired

**The unpaid layer under the AI boom is a payroll problem that companies misclassify as an ethics problem**

Papaya Global's pitch is that [the AI industry depends on millions of contract workers nobody has built payment infrastructure for](https://thenextweb.com/news/papaya-global-ai-workforce-infrastructure-payroll?ref=adjacent.media) — annotators, raters, red-teamers, spread across jurisdictions with inconsistent classification and payment rails. Read the business thesis rather than the moral one: someone is betting that AI labs will need auditable, compliant, multi-country workforce records, and will need them before regulators ask. That's a procurement bet, and it's the kind of unglamorous infrastructure that gets acquired.

The same classification mess shows up at the consumer end of the stack. A Roblox lawsuit over Tower Defense Simulator surfaced [developers paying staff in Robux with no formal contracts and contested ownership](https://aftermath.site/tower-defense-simulator-roblox-lawsuit/?ref=adjacent.media). Platform-native currency as compensation works until someone leaves, and then there's no employment agreement, no tax treatment, and no clean answer about who owns the game. Any platform running a creator economy with in-house currency is carrying this exposure whether or not litigation has found it yet.

**Perps found a product-market fit in things that don't trade well**

a16z crypto's read on [roughly $100B flowing through real-world-asset perpetual futures](https://a16zcrypto.substack.com/p/the-rise-of-the-100-billion-rwa-perp) is less about crypto than about liquidity structure. Perpetuals give continuous, leveraged exposure to assets — pre-IPO equity, commodities, private credit — that trade in thin or gated markets. The demand isn't ideological; the underlying markets are bad at price discovery and someone built a better tape. Expect the regulatory question to center on synthetic exposure to private assets; tokens will remain a secondary concern.

## Connected World

**Export controls regulate shipment but leave access unrestricted**

SEC filings and Financial Times reporting show [ByteDance made up close to 75% of Nscale's 2025 revenue, using the company's Norwegian facility to reach Nvidia hardware](https://www.ft.com/content/d1090476-7be6-4bba-ae2a-f417499e820a?ref=adjacent.media) (paywall). No chips crossed into China; compute did. The structure is legal, legible, and exactly the gap a shipment-based control regime leaves open. It also means a GPU cloud's "diversified customer base" can be one customer wearing a jurisdiction. Anyone underwriting neocloud capacity should be asking for revenue concentration disclosure before asking about power contracts.

Running the other direction, Beijing's state-asset regulator is [surveying how many Broadcom switches sit inside state-backed data centers](https://www.ft.com/content/b97dde39-7813-4a51-8176-3699e2a468b1?ref=adjacent.media) (paywall). Inventory counts precede substitution mandates. Networking has been the quieter half of the AI hardware stack, and it's the half where domestic Chinese alternatives are closest to viable — which makes Broadcom's China exposure a nearer-term line item than Nvidia's.

**AI glasses grew 263% by removing the part that made them glasses**

Counterpoint's H1 numbers put global AI glasses shipments up 263% year over year, driven mainly by Meta and EssilorLuxottica's display-*less* line. Smart glasses are winning by putting cameras, speakers, and a voice assistant into a frame people will actually wear in public—and doing so without AR. That's the Vision Pro lesson arriving from the opposite direction: the winning form factor cut the ambitious feature, hit a normal price, and looked like eyewear. Percentage growth off a small base is still a small base, but the composition of the growth is the useful signal for anyone planning a wearables roadmap.

## The New Consumer

**The install base doesn't hold a grudge**

Meta's Muse [took the top of the U.S. free iPhone chart, past ChatGPT](https://9to5mac.com/2026/09/18/metas-new-muse-ai-agent-app-overtakes-chatgpt-as-top-iphone-app/?ref=adjacent.media), which is what happens when you push an app to three billion existing accounts. Download rank measures distribution. It tells you nothing about retention. The number worth watching is 30-day usage; chart position is a distraction. The Verge's longer question about whether [a company can outrun its own reputation](https://www.theverge.com/cs/features/991659/meta-mark-zuckerberg-future-glasses-ai-instagram?ref=adjacent.media), reported against a deliberate reduction in Zuckerberg's public-facing role. Brand teams have a live case study here: Meta is testing whether separating founder perception from product perception restores advertiser and consumer confidence, and the early evidence is that most consumers never held the grudge the press assumed they did.

**Ad inventory is expanding into places that used to be neutral**

Apple is [placing persistent ads inside the iOS Settings app](https://mjtsai.com/blog/2026/09/22/persistent-ads-in-ios-settings/?ref=adjacent.media), per Michael Tsai's roundup of developer reaction. Settings is system chrome, the screen users visit to fix something, not to browse. Monetizing it is a small revenue line and a meaningful repositioning for a company whose privacy advertising is built on the claim that it isn't in that business. Services revenue growth targets eventually require new surfaces, and Apple has run out of easy ones. Anyone building on Apple's platform should assume the surfaces currently considered off-limits are inventory pending a quarterly miss.

## Machines & Minds

**Enterprise software's next rewrite is architectural rather than a matter of new features**

Forrester's argument is that [business applications are becoming agentic at the architecture level](https://www.forrester.com/blogs/how-business-applications-become-agentic/?ref=adjacent.media) — workflow orchestration moving from user-triggered to agent-initiated — rather than picking up AI as another feature release. The buying consequence is specific: procurement criteria shift from feature checklists to permission models, audit trails, and what happens when an agent acts wrong. If you're evaluating an enterprise suite this quarter, the question to ask vendors is not what the agent can do but what it's allowed to do and who signs off on the exception.

**Cheap and open is collapsing the premium on frontier**

The Signal's read is that [open-weight models have closed most of the gap that justified paying for frontier access](https://thesignal.substack.com/p/the-open-models-have-caught-up), arriving the same week DeepSeek's V4.1 Flash reportedly trained for roughly $10M. If capability converges, the durable margin sits in distribution, proprietary data, and integration depth — which is precisely why Meta can win a chart position with a middling model and why Google's hesitation cost it more than a product.

Convergence has a less comfortable corollary. NewsGuard's testing found U.S. and Chinese AI tools fail in strikingly similar ways, repeating false claims at comparable rates regardless of origin or governance regime. The policy framing that treats model provenance as a proxy for reliability doesn't survive contact with the evaluation data. For enterprises, "we use a Western model" has become a procurement talking point rather than a meaningful measure of quality control.

---

*15 articles across 5 themes · 12 sources · Powered by Folo + Claude*