U.S. technology stocks surged on Sept. 21 as enthusiasm over AI agents and the computing infrastructure needed to run them pushed the Nasdaq Composite to a record close, while AMD’s market capitalization rose above $1 trillion for the first time.
The Nasdaq gained 2.26% to finish at 27,122.09. The S&P 500 rose 1.49% to 7,764.70. Meta Platforms climbed about 11.4%, AMD added roughly 10%, Intel advanced around 12%, and Arm rose about 17%, concentrating much of the day’s gains in companies positioned to supply or deploy AI computing capacity.
The immediate consumer-facing catalyst was Muse, Meta’s personal AI agent, which reached the top of the free-app rankings in both the U.S. Apple App Store and Google Play less than two weeks after its release. The market reaction extended beyond the app’s early download numbers, focusing instead on the prospect that AI agents performing routine digital work could create a more persistent demand for data-center processing.
Meta’s Muse app climbs U.S. download rankings
Meta released Muse as a standalone app on Sept. 8 and also made it available through WhatsApp. The company has presented the software as an AI agent that can carry out multi-step tasks, including browsing the web, completing forms, managing email and booking travel.
The app is designed to seek user confirmation before taking sensitive actions such as sending emails or making payments. That requirement limits fully autonomous use, but it also frames Muse as a practical assistant for tasks that normally require repeated interaction with websites, calendars and inboxes.
Appfigures estimated that Muse had reached about 1.1 million cumulative installs by Sept. 18. Meta has not disclosed its own installation figure. Reaching the top free-app position on both major U.S. mobile storefronts gave the launch an early distribution signal, though download rankings do not show how frequently users return to the service or how many tasks the agent completes.
Muse’s release has sharpened a question that has become central to AI infrastructure spending: how much computing will be required when agents move from answering occasional prompts to continuously handling workflows for large numbers of users.
A conventional chatbot interaction may involve a short exchange. An agent that searches multiple sites, compares results, fills in information, checks a calendar and waits for user approval can require repeated model inference, tool use and scheduling. Inference is the process of running a trained AI model to generate an answer or perform an action. As usage rises, that workload can become a recurring demand rather than a one-time training project.
Chipmakers rally on demand for AI capacity
The gains in AMD, Intel and Arm reflected expectations that the next stage of AI competition will depend heavily on the servers, processors and power systems behind consumer applications.
AMD’s rise above a $1 trillion valuation placed it among the largest companies in the semiconductor sector. The company has been building its position in AI data centers with its Instinct graphics processing units, or GPUs, and EPYC server processors, seeking to compete for spending that has been dominated by a small group of large cloud and technology companies.
AMD reported $6.7 billion in data-center revenue for the second quarter, up 107% from a year earlier. The company attributed the increase to demand for EPYC processors and Instinct GPUs, showing that its AI business had already become a major source of growth before the latest stock-market advance.
Intel’s roughly 12% rise and Arm’s 17% jump broadened the move beyond one GPU supplier. Intel has been working to regain ground in data-center computing and AI hardware, while Arm’s processor designs are widely used across mobile devices and increasingly feature in servers built for energy-efficient workloads.
The rally suggests that traders are treating AI agents as a potential demand driver across several layers of the computing stack: chips, servers, networking equipment and the electricity systems required to operate large data centers. That view remains dependent on whether early consumer interest in products such as Muse translates into lasting engagement and revenue.
Meta and AMD have a large deployment plan
Meta and AMD already have an announced infrastructure relationship that gives the market a concrete link between the app developer’s AI ambitions and the hardware rally.
In February, the companies said they planned to expand their existing partnership, including deployment plans for up to 6 gigawatts of AMD Instinct GPUs and deeper collaboration around AMD’s EPYC server CPUs. A gigawatt is one billion watts of power capacity, illustrating the scale of equipment and electricity involved in large AI data-center programs.
AMD said products for the first 1 gigawatt of planned deployments are expected to begin shipping in the second half of 2026. The broader 6-gigawatt target should not be read as an immediate order schedule, but it provides a framework for how much capacity Meta could seek if its AI products continue to expand.
The focus on power has become as relevant as the focus on chips. Building AI data centers requires access to electricity generation, transmission connections, cooling equipment and physical space. Large clusters can take years to develop when utilities face grid constraints or when developers must secure new generation capacity.
That bottleneck could shape which companies benefit most from AI demand. A chipmaker may report strong orders, but revenue recognition and equipment deployment can depend on whether customers have power-ready facilities in which to install the systems.
Falling yields add support to technology shares
Broader market conditions also supported the technology-led advance. Oil prices and U.S. Treasury yields declined during the session as equities moved higher.
Lower Treasury yields can improve the appeal of high-growth technology companies because a larger share of their expected value is tied to earnings projected further into the future. They can also reduce financing costs for companies funding capital-intensive data-center construction, although the scale of current AI projects means that access to power and equipment remains at least as consequential as borrowing costs.
The Sept. 21 move does not establish that Muse’s early app-store performance will translate directly into long-term demand for semiconductors. Yet Meta’s deployment plans with AMD offer a clearer route from AI software adoption to infrastructure spending than many previous consumer AI launches.
The next tests will come through user-retention data for Muse, Meta’s disclosures on AI-related capital expenditure, and AMD’s updates on the timing and scale of its Instinct GPU shipments.
Explore how AI trading tools mirror Wall Street’s AI surge with Toobit’s AI copy trading guide today.
Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.
