An Assistant that Does the Work

While most AI chatbots just answer questions, Muse is designed to handle tasks. Users can ask it to manage emails and calendars, book restaurants and travel, fill out forms, research products, track spending, and shop online. Muse keeps working even after the app is closed and can break big jobs, like planning a trip, into smaller steps.

Each user’s Muse runs on its own virtual computer in the cloud, with its own browser. This setup lets it browse regular websites like a person. Meta says users have to approve sensitive actions like sending emails or making purchases, and that passwords and payment details are kept separate from the AI model.

The app is free, but paid options start at $20 and go up to $100 a month for heavier users. Early ratings are nearly five stars on both app stores, though these may change as more people join.

The Spending Finally Has a Face

This is important because Meta is spending huge amounts on AI. The company expects to spend $130 billion to $145 billion this year, with $31 billion in the second quarter alone. This heavy spending has pressured its cash flow. Quarterly free cash flow dropped to $784 million, while operating cash flow was $31.9 billion.

Before this, investors had to trust that Meta’s AI spending would pay off. Now, Muse gives them a product they can track week by week, with several possible ways to make money, like subscriptions, commerce, and business use. Meta’s biggest strength is its reach—few AI companies start with billions of users on Facebook, Instagram, Messenger, and WhatsApp.

Analysts noticed the change. Wells Fargo raised its price target for Meta, saying Muse makes Meta look more like a leader in consumer AI, not just a major chip buyer.

Why Chip Stocks Care, and Why It’s CPUs this Time

The biggest reaction was in semiconductor stocks. Arm jumped about 17%, Intel rose 12%, and AMD nearly 10%, with AMD closing above a $1 trillion market value for the first time. The broader Philadelphia Semiconductor Index gained 4.3%, showing that CPU makers led the way.

The reason is that AI agents need a different mix of computing power. Large models still mostly use GPUs, but an agent like Muse also needs regular server CPUs to run its virtual machine, browser, databases, and networking, and to handle the many small tasks that come with each call to the AI model.

A single server CPU can run many virtual machines, so one Muse user does not equal one chip. Still, if AI agents reach tens or hundreds of millions of users, investors see this as a new source of demand for data-center CPUs as well as GPUs. In short, the market is using Muse’s download numbers as an early sign of chip demand.

The Hurdles Ahead

Getting downloads is the easy part. Amazon has already blocked Muse from shopping on its site, saying the agent is not authorized and does not properly identify itself when browsing.

This mirrors Amazon’s dispute with Perplexity’s Comet agent. If major websites decide AI agents need permission to operate, tools like Muse could become much less useful. This issue could shape the entire category.

Privacy is another big challenge. An assistant that reads your messages, calendar, and accounts must convince people that the convenience is worth giving up some privacy. With Meta’s track record on personal data, trust could be just as important as features.

Meta’s valuation is not as high as the recent rally might suggest. The company trades at about 28 times its trailing earnings, which is above the S&P 500 average but much lower than many other AI companies. The stock is near its 52-week high of $770.60, so expectations are now much higher than a month ago.

What to Watch

• Meta Connect this week. More detail on Muse, Meta’s AI models, and the product roadmap.
• Paid conversion. Early signs of how many users move to the $20 and $100 tiers.
• Website access. Whether other large retailers follow Amazon or open their sites to AI agents.
• CPU demand. Commentary from Arm, AMD, and Intel on whether agent workloads are showing up in orders.