About half of Americans say they use AI. Only 4.5% pay for it. That's the number I can't stop thinking about this week, and it's today's lead. Behind it: Google gave its AI agents their own email inboxes, AI's appetite for memory chips is pushing up iPhone prices, and three of the biggest book publishers in the US got caught using AI on covers and copy without saying so.
Half of Americans use AI. Fewer than 5% pay for it.
The venture firm a16z published the seventh edition of its Top 100 Consumer AI Apps report, and for the first time it ranks apps by what people actually spend on their cards, not by traffic. The data comes from a US card panel. In August 2026, 4.5% of US consumers had a paid personal subscription to ChatGPT, Gemini or Claude. A year earlier it was 2.1%. So it doubled, but from a tiny base.
The spending is wildly uneven. The median paying user spends about $25 a month. The top 1% spend an average of $903 a month, and that small group accounts for 19.5% of all observed consumer AI spending. That's more than the bottom half of payers combined. And here's the detail I like most: 29 of the top 50 companies by spend don't show up on a16z's web or mobile traffic lists at all. Small, focused tools are quietly making real money while the famous chatbots collect the visits.
Two caveats. It's US-only, and it's panel data, not company revenue. a16z says so itself.
So what does this mean for you? First, your customers meet AI mostly for free, and mostly inside apps they already use. Don't build a sales pitch that assumes they're AI fans who pay for tools. Second, the money in AI goes to tools that do one job well. If you're deciding what to adopt, pick by the job, not by the brand. Our breakdown of AI tools for Instagram sellers is sorted that way on purpose. Third, that "top 1% pays a fortune" pattern? Your DM inbox probably has the same shape. A few repeat buyers carry the month. Know who they are.
Google's AI agents now get their own inbox
On October 8 Google launched a Gemini agent for work, and the strangest part is the "coworker agents." Each one gets its own Workspace account: an email address, a calendar, Drive storage and a seat in the company directory. Colleagues hand it work like they would a new hire. Its actions are logged under its own name, not yours.
It's in private preview for a small group of business customers. Google says it will come at no extra cost where Gemini Enterprise is available. There's no date for wider release, and questions about who is responsible when the agent gets something wrong are still open.
You won't get one of these next week. But watch the pattern. The big companies are no longer selling AI as a chat window. They're selling it as a team member with a name, a job and a paper trail. That's the right way to think about AI in your shop too. Give it a clear role, know exactly which messages it sent, and decide in advance what it is not allowed to touch. If you've ever turned an AI assistant off because you stopped trusting it, the reason is usually missing rules, not a bad model. We wrote about why sellers turn their AI assistant off, and the fix is boring and effective.
AI data centers are making your phone more expensive
Nikkei Asia reported on October 9 that Apple cut October component orders for the iPhone 18 Pro and Pro Max by 15% to 20% below its original requests. The reason given: memory chip prices have shot up because AI data centers are buying memory in huge volumes, Apple raised prices, and demand softened. The Pro models start at $1,199 and $1,299, which is $100 more than last year.
Keep this one in proportion. The report rests on anonymous sources, Reuters couldn't confirm it right away, and Macworld points out that "iPhone production cut" stories show up nearly every autumn. Apple also pushed the regular iPhone 18 to early 2027, which can make component orders look odd.
The memory squeeze itself is real, though. TrendForce estimated in August that the parts in a 256GB iPhone 18 Pro cost about 38% more than last year's model, mostly because of memory. If you sell phones, laptops, tablets or accessories, expect higher landed costs and pickier buyers. When the price goes up, the questions in your DMs go up too: which storage size, is it worth it, can I pay in installments. That's exactly the moment a fast, accurate answer wins the sale.
Book publishers used AI quietly. Their staff aren't happy.
Wired spoke to more than two dozen current and former staff at HarperCollins, Simon & Schuster and Hachette. They describe AI chatbots being used for back-cover copy, publicity text, cover art and emails to literary agents, often at executives' request, and without telling authors or readers. Some staff are pushing back. Hachette, which has publicly positioned itself against AI, reportedly draws a line between "operational" uses it allows and "creative" uses it opposes.
These are anonymous accounts, and the publishers gave only limited responses. But the lesson doesn't depend on the details. Using AI isn't what upsets people. Hiding it is.
For a seller, this is a cheap lesson. AI-written captions and AI-touched product photos are fine. What gets you in trouble is when the AI version promises something the real product doesn't deliver: a color that isn't quite that color, a fabric that looks thicker than it is. Customers forgive a tool. They don't forgive feeling fooled. If a photo is a render, say so, and keep a real photo one tap away.
Quick takes
- Microsoft ships a tiny "decision" model. Microsoft-Decision-1 launched October 9 in Microsoft Foundry. It doesn't write text. It picks from fixed options (yes/no, categories, scores) and returns a probability for each, at $0.042 per million input tokens with free output. The speed claims are Microsoft's own. This is the kind of quiet engine that sorts messages behind the scenes: "is this a complaint, an order or a price question?"
- Perplexity open-sources its search embeddings. Two models, a small 0.6B one and a 9B one, released under the MIT license. They share one embedding space, so a big model can build the index and a small one can serve it cheaply. Expect better product search on smaller sites in the coming months.
- Cloudflare buys Deno. The JavaScript runtime team is joining Cloudflare. Deno Deploy, its hosting service, is set to shut in about six months. If your developer hosts your shop's code there, ask them about it this week.
- Nvidia and Reflection AI. The Financial Times reports Nvidia is in early talks to buy or invest more in Reflection AI. Nothing is confirmed, so treat it as a rumor for now.
The thread running through this week: AI is moving out of the chat box and into work with names, roles and price tags, and the winners are the tools that do one clear job. That's the bet Vardast makes in your inbox. It learns your products, answers customers on Instagram, WhatsApp, Telegram and your website, and hands the conversation to you when a human should take over. If you're weighing the cost, our guide to AI assistant pricing for Instagram walks through the math.