Everything About AI
NextFin News - Meta's Muse agent has done in two weeks what most consumer apps never do: it reached the top of the U.S. app charts, forced Wall Street to rewrite its revenue model for Meta Platforms, and pulled the entire semiconductor complex higher with it. Advanced Micro Devices crossed a $1 trillion market value for the first time not because of a GPU order, but because investors decided a chatbot that can pack your bags, book a date night, and sell a car on your behalf is a CPU story. That second-order trade is bigger than Meta's own 20% rally — and it is the real story of the week.
The launch landed on September 8, when Meta released Muse as an action-oriented personal AI agent rather than a conventional chatbot. The basic version is free; heavier use costs $20 or $100 a month. Within days it was the No. 1 free app on both Apple's U.S. App Store and Google Play. Sensor Tower recorded more than 2.5 million cumulative U.S. downloads by September 19, with daily downloads topping 300,000; on September 19 alone the app logged 264,000 U.S. downloads, its third consecutive day above 200,000. Apptopia put global downloads at 2.8 million in the first 12 days. In its first six days, Muse logged 902,000 downloads, ahead of the 773,000 that Meta's predecessor AI app managed in the same window, according to data cited by Abe Yousef, a senior insights analyst at Sensor Tower.
The market reaction was immediate and broad. Since launch, Meta shares have risen more than 20%, adding over $200 billion in market capitalization to reach their highest level in more than seven months. On Monday alone, Meta gained more than 11%, with shares touching $757.27 intraday on September 22 before settling near $740. But the rally leaked far beyond one stock: Advanced Micro Devices jumped 9.95% to $615.52 to close at a $1.004 trillion valuation, its first trip past the 13-figure mark. Intel rose 12% and Arm 17% on the same session, and the Philadelphia Semiconductor Index advanced more than 4%. For a market that spent 2025 and early 2026 punishing AI spending with multiple compression, the reversal was sharp: the same investors who sold Meta on capex fears bought it on agent adoption.
Wall Street has moved quickly to monetize the narrative. Jefferies raised its Meta price target to $875 from $710, framing a scenario in which Muse reaches one billion users by the end of 2027 and converts 3% to paid plans — 30 million subscribers generating $10.8 billion in annualized revenue. The new target implies 25 times projected 2027 earnings, up from 20 times. Wells Fargo lifted its target to $796 from $640, and Bank of America kept a Buy rating with an $810 target while flagging privacy and trust as adoption risks. Of the 64 brokerages covering Meta, 57 rate it a buy or better, with a median price target of $760. JPMorgan analysts went further, writing that Muse has the potential to become the most widely used consumer AI application since ChatGPT, crediting product-market fit, Meta's distribution reach, and a free tier that lowers the barrier to trial. UBS Global Wealth Management said Muse provides early evidence that consumer AI agents could achieve broad adoption, offering new AI monetization opportunities beyond enterprise application.
Why a Chatbot Became a CPU Trade
The first-order read of Muse is simple — a popular app lifts its parent's stock. The second-order read is what moved AMD past $1 trillion. Agentic AI does not just answer questions; it executes multi-step workflows across other apps and websites, maintaining state, calling tools, and running long-horizon tasks with limited user intervention. That workload profile is heavier on CPUs, memory, and orchestration silicon than the training-centric GPU demand that has dominated the AI buildout. Michael O'Rourke, chief market strategist at JonesTrading, put it plainly: the market has interpreted Muse as a boon for the already significantly bottlenecked CPU market.
The market has interpreted Muse as a boon for the already significantly bottlenecked CPU market.
The distinction matters because the market has been pricing one kind of AI demand while a different kind arrives. Training is episodic — it clusters around model releases and then fades, which is why GPU orders arrive in lumps. Inference for agents is continuous and scales with user engagement: every delegated task, every background workflow, every autonomous transaction adds load, and unlike a chat query, an agent session can run for minutes or hours while it coordinates across services. Bank of America's research team noted that early reviews suggest functionality is resonating with users, particularly the agent's ability to execute multi-step tasks and workflows with limited user intervention — precisely the usage pattern that keeps silicon busy.
This is a structural shift in the AI value chain, not a cyclical rotation. If consumers adopt agents the way they adopted smartphones — and the download pace suggests they are willing to try — the compute requirement compounds rather than mean-reverts. AMD's own numbers back the timing: second-quarter revenue of $11.54 billion, up 50% year over year, with data-center sales of $6.7 billion, up 107%. The company has disclosed multi-gigawatt capacity deals, and analysts note that Meta is among AMD's largest hyperscaler customers, so Muse's success could be a major tailwind for a supplier whose stock had trailed Nvidia's for most of the AI cycle. AMD entered the week on a five-session advance of about 25%, lifting its gain since the start of 2026 above 180%.
Cyclical Excess Riding a Structural Move
The direction is structural; the velocity is not. AMD's roughly 41 times forward earnings — compared with about 16.3 times for Nvidia — show a market front-running a thesis that is still unproven in the revenue line. Monday's reporting did not show how much CPU capacity Muse actually uses, and the valuation gap between the two chip designers had never been wider in the AI era. The agentic-AI premium is being priced before the metered usage data exists. That is the cyclical leg riding on top of the structural one, and it is the part most vulnerable to a drawdown if adoption stalls.
History offers a caution. Download charts are a measure of curiosity, not retention. ChatGPT reached one million users in five days in late 2022 and never looked back, but most viral apps roll over within a month once the novelty fades. The comparison that matters is not Muse versus the Meta AI app in the first six days; it is Muse in month three versus month one. Sensor Tower's data shows Muse ahead of comparable launches for Claude and Gemini over similar periods, but those are still early windows. The mean-reversion risk is real: if daily downloads fall back below 100,000, the structural thesis survives but the multiple contracts fast.
The Monetization Bet Against the Spending Reality
Jefferies' $10.8 billion scenario is a clean story, but it sits against a much heavier balance-sheet reality. Meta has raised its 2026 capital expenditure forecast to a range of $130 billion to $145 billion, roughly double last year's investment and a large share of the more than $700 billion the four hyperscalers are projected to spend on AI infrastructure in 2026. The company plans to double overall computing power to 7 gigawatts this year and to double it again, to 14 gigawatts, next year, and currently has 32 data centers across the globe in operation or under construction.
The cash consequence is already visible. Free cash flow in the most recent quarter fell 91% to $784 million from $8.5 billion a year earlier, and sell-side models now expect negative free cash flow in 2026 and 2027. For a company that generated $43.6 billion of free cash flow in 2025, that is a regime change in its financial profile. Capital expenditures on servers, data centers, and network gear surged 83% to roughly $31.1 billion in the quarter alone. The market is being asked to underwrite a subscription-and-commerce future while the cash engine stalls — and to accept that the depreciation wave from this buildout will hit income statements for years after the spending peaks.
There is also a valuation asymmetry worth naming. At 25 times projected 2027 earnings, Jefferies' target prices in a future that requires flawless execution on adoption, conversion, and infrastructure returns. Bank of America's $810 target, held through the rally, implicitly asks for less from the agent and more from the core advertising business, which ended the second quarter with 3.6 billion daily active users across the family of apps and revenue growth of 28%. The two price targets are not just different numbers; they are different beliefs about which business will carry the next leg of Meta's growth.
The Walled-Garden Counterattack
There is also a friction that the rally has largely ignored. Amazon blocked Muse from shopping on Amazon.com, citing security, privacy, and transparency concerns, and asked Meta to remove the marketplace from the agent's experience. Users who send Muse to Amazon now see a warning that continued access by an unauthorized AI agent violates Amazon's Conditions of Use. Amazon said Muse appeared to capture and store customer credentials, scrape account data, and browse without identifying itself as an agent — a violation of its terms, which require agents to embed a text snippet in HTTP requests.
The standoff matters because commerce is central to the agent revenue thesis. Muse's most valuable actions are the ones that move money: booking travel, buying groceries, selling a car. If the largest online retailer, and potentially others, wall off their sites, the agent that does everything shrinks to the agent that does what platforms allow. Amazon told reporters it was not made aware in advance that Muse would access its store and did not authorize it. This follows a similar clash last year, when Amazon sued Perplexity AI over its shopping agent; the case was dismissed, but the court did not bar Amazon from blocking agents that breach its terms.
Meta is reportedly working to enhance Muse's data-protection guardrails, including an upcoming Muse Confidential VM, to prevent such access disputes. But the structural point stands: the agent economy will be negotiated territory, not open terrain. Every major platform — Amazon, Google, Apple, and the banks that hold payment rails — has a veto over how deeply an agent can reach into its systems. That negotiation power will shape who captures the economics of agentic commerce, and it is a risk that does not appear in Jefferies' conversion math.
What to Watch: The Signals That Settle the Thesis
The structural-agentic thesis holds if Muse retains users and converts them. Watch three metrics: daily downloads, which should stay above 200,000 in the U.S.; paid conversion, which needs to approach the low single digits within 90 days; and Meta's guidance at the Connect developer conference, where new partnerships with Shopify and Expedia are expected to be detailed. The single signal that would break the structural call is a sustained drop in daily downloads below 100,000 for two consecutive weeks, or a paid conversion rate under 1% after the first quarter — evidence that the agent is a novelty rather than a habit.
The base case is that Muse keeps its chart position, converts 2-3% of users to paid tiers, and Meta's agent partnerships expand across commerce and travel. Meta re-rates toward the Jefferies target, and the CPU and memory complex holds its gains. The upside case is that conversion exceeds 5%, commerce take rates add a second revenue layer, and agentic workloads become a durable, metered compute demand — AMD's 41 times earnings gets absorbed by growth rather than compressed. The downside case is that privacy concerns and platform blocks cap adoption, downloads roll over, and Meta's negative free cash flow forces a multiple compression back toward its historical range; the agentic premium in chip stocks unwinds fastest.
Across time horizons, the signals differ. Short term, the trade is momentum and positioning — AMD's 25% five-day move is a sentiment extension that can give back quickly on any soft download print. Medium term, the decisive data points are retention and conversion, not downloads; an app can be viral and still fail to monetize. Long term, the structural question is whether consumer AI agents become the next general-purpose interface — in which case today's chip rally is the beginning, not the top, and the platforms that control the agent's access to real-world actions, not the models with the highest benchmark scores, capture the durable economics.
Muse is not just Meta's answer to ChatGPT. It is the market's first real test of whether AI can be sold as a habit rather than an ad impression — and the CPU complex is betting, before the receipts exist, that it can.
Explore more exclusive insights at nextfin.ai.
