Prince Mario-Max Schaumburg-Lippe: Meta Muse Passes 10 Million Users as AI Spending Rises

Ten million people is not a pilot program. It is a habit forming in real time. Meta says more than 10 million people in the United States and Canada have set up its Muse agent in about a month and a half, and roughly a quarter of them come back on any given day. For a personal agent that only launched in August, that pace tells a simple story: people did not just try it once and move on. They kept it.

The company paired that milestone with a bigger commitment behind the scenes. Meta raised its 2026 capital spending guidance to $135 billion to $145 billion, lifting the low end from $130 billion. It also said spending in 2027 will grow further, without giving a figure yet. That is data centers, chips and the compute that lets an agent answer quickly, remember context and actually finish a task instead of offering a polite paragraph about it.

Shopping Is Where Muse Found Its First Real Job

Ask what people are doing with Muse and the answer is refreshingly ordinary. Agentic shopping is its top use. Meta says a quarter of the users who shop through Muse spend over $200 on average per purchase, a detail that matters because it shows trust. Handing an agent a vague wish is easy. Handing it your money and letting it complete an order takes a different level of confidence.

Think of the small jobs that fill a week. Reordering a gift that worked last year. Comparing three nearly identical appliances and picking the one that fits the space you measured. Tracking down a sold-out size at a second store. None of that is glamorous, and that is exactly the point. An agent that handles the dull, multi-step errands well earns the right to help with bigger ones. Readers who followed how Nvidia is cutting AI data center build time in Berlin will recognize the other half of this story: the models get the headlines, but the buildings and the hardware decide how fast any of this can grow.

What $135 Billion to $145 Billion Actually Buys

Numbers that large can blur. So break it down. Capital spending at this scale buys capacity that has to be planned years ahead – land, power, cooling, servers and the engineers to run them. When Meta lifts the floor of its guidance, it is signaling that demand is not a short spike it expects to fade by spring. It is building as if agent use will keep climbing, and as if each user will ask the agent to do more, not less, over time.

For everyday users, the payoff is practical. More capacity means faster replies at busy hours, better memory across a longer conversation, and agents that can work through a task with several steps without stalling halfway. For small businesses, it means a storefront assistant that can answer sizing questions at midnight, check stock, and close a sale while the owner sleeps. That kind of help used to require staff a small shop simply did not have.

A Network of Agents, Not Just One

Meta has also teased a future Muse network, where a person’s agent could interact directly with the agents of friends, family and businesses. Picture it in plain terms. Your agent checks your calendar, a restaurant’s agent confirms a table for six on Friday, and your sister’s agent flags that she is running 20 minutes late – all before anyone opens a group chat with 43 unread messages. Scheduling a family dinner should not feel like air traffic control. If agents can negotiate the boring parts with each other, people get the evening back.

Meta says Muse has reached product-market fit and that its focus now shifts to scaling and monetizing it. That phrasing is worth pausing on. Plenty of AI tools launch to a burst of curiosity and then quiet down. A product that keeps a quarter of its early users active daily, six weeks in, is showing something sturdier. The next test is breadth: can the same agent that shops well also plan travel, sort a cluttered inbox, and help a student work through a problem step by step without rushing to the answer?

What Builders Should Take From This Week

If you build products, the Muse numbers offer a clear brief. Start with one job people already do every week. Do it completely, not halfway. Show your work so the user can check it in five seconds. And keep the cost of a mistake low, because trust grows when an agent asks before it spends and confirms after it acts. The builders who get those basics right will not need to chase attention. Ten million early users suggest people are actively looking for help they can hand real tasks to.

There is a wider shift here too, one this site has tracked as space data firms win concrete NASA work for their technology: useful AI is moving out of demos and into paid, practical jobs. Muse fits that pattern. It is not asking people to admire a model. It is asking to run an errand, book the thing, find the thing, finish the thing. That is a modest pitch, and a powerful one.

Meta still has plenty to prove. Ten million users in two countries is a strong start, not a finish line, and the company itself frames its ambition in billions. But direction matters. Spending guided up to $135 billion to $145 billion, an agent people return to daily, and shopping behavior that shows real trust – together, those point the same way. Personal agents are becoming part of the weekly routine, one completed task at a time.

Also published on: