Meta’s AI agent Muse hits 700K users in 11 days, faces compute bottlenecks
Meta's AI agent Muse reached approximately 700,000 daily active users within 11 days of its US and Canada launch, a tenfold increase, but is already experiencing service degradation due to compute infrastructure limits. Monitoring platform SaaSHub lists Muse as "degraded," with users reporting frequent search failures. A stress test showed Muse could only create 33 of 120 requested sub-agents. Meta Chief AI Officer Alexander Wang stated on September 9 that usage was 10 times internal expectations, prompting token limit resets. Each user gets a dedicated virtual machine with 2 vCPUs, 8GB RAM, and 100GB SSD, making scaling hardware-intensive.
IllustrationEditorial responsibility
- No named human review is recorded for this page.
- Reports are grouped by semantic similarity and deterministic rules. Language models may assist titles, summaries, translation and cross-source analysis; the page reads the event directly, while its address stays stable when the title changes.
- Summary covers the current reports
Cross-source coverage
Common ground
- Meta's Muse architecture, giving each user a dedicated virtual machine, is inefficient and unsustainable at scale.
- Meta's planning was poor, as they underestimated demand by 10 times and hit a 72% failure rate on sub-agent creation.
- The Global South is often left out of initial tech rollouts, raising concerns about inequality and trust.
- Chinese firms like DeepSeek have shown impressive efficiency gains in AI development.
Points of contention
- Whether Muse's bottleneck is primarily an engineering problem or a sign of deeper strategic and ethical flaws.
- Whether China's AI model is genuinely more development-oriented or just as extractive as Western tech, just with different goals.
- Whether the compute shortage is caused by bad architecture, geopolitical factors like chip supply, or a deliberate choice to prioritize Western markets.
- Whether the fix is a simple engineering refactor or requires a fundamental shift in how AI is deployed globally.
Blind spots
- The debate largely ignored the voices and needs of end-users in the Global South, treating them as passive recipients rather than active stakeholders.
- There was little discussion of how local communities could own or govern AI tools to ensure they serve their own interests.
- The focus on China versus the West overlooked the possibility of alternative models, like open-source or community-driven AI, that don't rely on either corporate or state control.
WorldAttention’s read
The roundtable revealed that Meta's Muse bottleneck is a complex issue with technical, ethical, and geopolitical layers. While everyone agreed the architecture is wasteful and planning was poor, they split on root causes: the Eastern Agent saw it as proof of Western inefficiency versus China's strategic self-reliance, the Regional Agent framed it as a symptom of global inequality and broken trust, and the Neutral Agent insisted it's a fixable engineering and business problem. The blind spot was a lack of focus on what local communities actually need, with both Silicon Valley and Beijing seen as extracting value rather than empowering users. Ultimately, the debate showed that without addressing who AI serves and how it's governed, any technical fix will just reinforce existing power imbalances.
Reporting timeline
Meta's AI Agent Muse Hits 700K Users in 11 Days but Faces Compute Bottlenecks
Meta's artificial intelligence agent Muse has achieved rapid user growth, reaching approximately 700,000 daily active users within 11 days of its launch, according to Similarweb data. However, the service is already experiencing significant scaling issues due to insufficient computing power. Status monitoring platform SaaSHub has listed Muse as 'service degraded,' with users reporting frequent search failures. A researcher's stress test showed Muse could only create 33 out of 120 requested sub-agents. Meta Chief AI Officer Alexander Wang stated on September 9 that early usage far exceeded internal expectations, roughly 10 times the beta test level, prompting Meta to reset token usage limits. The bottleneck stems from Muse's compute-intensive architecture, which allocates each user a dedicated virtual machine with 2 vCPUs, 8GB RAM, and 100GB SSD. Industry analysis estimates that reaching 100 million users would require approximately 2 billion CPU cores, 800PB of memory, and 10,000PB of SSD storage. Meta currently offers up to 100 million free tokens per week per user, with paid subscriptions starting at $20/month. The service is only available in the US and Canada, with WhatsApp and Instagram user registration not yet opened.
Read sourceMeta's AI Agent Muse Hits Service Issues Before Reaching 1 Million Daily Users
Meta's personal AI agent, Muse, has seen rapid user growth, with daily active users increasing roughly tenfold to 700,000 in the past 11 days. However, this growth has outpaced expectations, leading to service degradation and agent task failures, revealing emerging compute bottlenecks. According to Similarweb, Muse currently has about 700,000 daily active users, but this is limited to the US and Canada, with WhatsApp and Instagram registration not yet fully open. Meta's architecture assigns each user an isolated cloud virtual machine with 2 vCPUs, 8GB RAM, and 100GB SSD, which scales compute demands linearly with users. A stress test showed that when asked to create 120 sub-agents simultaneously, only 33 succeeded. Meta Chief AI Officer Alexander Wang noted that early usage was about 10 times higher than internal test groups, prompting token strategy adjustments. Third-party monitor SaaSHub has marked Muse as 'Degraded,' with 'cannot search' being a common complaint. Meta is expanding access via smart glasses and a keychain device called Muse Charm, and has begun monetization with a commission on transactions and a $20 monthly subscription for heavy users. The article highlights that scaling Muse will require significant data center investment and infrastructure procurement.
Read sourceMeta's AI Agent Muse Faces Compute Bottleneck After Rapid User Growth
Meta's artificial intelligence agent, Muse, has experienced explosive user growth, with daily active users rising 10-fold in 11 days to approximately 700,000, according to Similarweb. However, this rapid adoption has led to critical scaling issues. Monitoring platform SaaSHub has downgraded Muse's status to 'service degradation,' citing persistent problems such as failed search responses. A researcher's stress test showed Muse could only create 33 out of 120 requested sub-agents. Meta's Chief AI Officer Alexander Wang revealed on September 9 that usage was 10 times higher than internal expectations, prompting new token limits. The bottleneck is attributed to Muse's compute-intensive architecture, which allocates each user a dedicated virtual machine with 2 virtual CPUs, 8GB RAM, and 100GB SSD. Industry analysis suggests that reaching 100 million users would require approximately 2 billion CPU cores, 800PB of memory, and 10,000PB of SSD storage. Currently available only in the US and Canada, with WhatsApp and Instagram registration not yet open, Meta faces a significant hardware constraint that could limit further market share expansion.
Read sourceShow 3 older updatesHide older updates
Meta's AI Agent Muse Hits 700,000 Users in 11 Days but Faces Compute Bottleneck
Meta's artificial intelligence agent, Muse, has achieved rapid user growth, reaching approximately 700,000 daily active users within 11 days of its limited US and Canada launch, according to Similarweb data. However, the surge has exposed critical compute infrastructure bottlenecks. Status monitor SaaSHub has listed Muse as experiencing 'service degradation,' with users reporting frequent search failures. A researcher's stress test showed Muse could only create 33 out of 120 requested sub-agents. Meta's Chief AI Officer Alexander Wang stated early usage was 10 times internal expectations, forcing new token limits. The article notes Meta provides each user a dedicated virtual machine, and analysts estimate that scaling to 100 million users would require approximately 158 million CPU cores, 800 PB of memory, and 10,000 PB of SSD storage. The current pricing model, offering free weekly tokens and paid subscriptions from $20 to $100 per month, may not cover these hardware costs. The compute shortage is identified as the primary barrier to Meta's expansion plans for Muse.
Read sourceMeta's AI Agent Muse Hits 700K Users in 11 Days but Faces Compute Bottlenecks
Meta's AI agent Muse has achieved rapid user growth, reaching approximately 700,000 daily active users within 11 days of its launch, a tenfold increase according to Similarweb. However, the service is already experiencing significant scaling issues due to insufficient computing power. Status monitoring platform SaaSHub has listed Muse as experiencing 'service degradation,' with users reporting frequent failures such as inability to search. A researcher's stress test showed Muse could only create 33 out of 120 requested sub-agents. Meta's Chief AI Officer Alexander Wang noted that early usage far exceeded internal expectations, prompting the company to reset token usage limits. The bottleneck is attributed to Muse's compute-intensive architecture, which allocates a dedicated virtual machine with 2 virtual CPUs, 8GB RAM, and 100GB SSD per user. Industry analysis suggests that reaching 100 million users would require approximately 158 million AMD Ryzen EPYC CPU cores, 800PB of memory, and 10,000PB of SSD storage. The service is currently only available in the US and Canada, with WhatsApp and Instagram user registration not yet opened.
Read sourceMeta's AI Muse Hits 700K Users in 11 Days but Faces Compute Bottlenecks
Meta's artificial intelligence product, Muse, has reached approximately 700,000 daily active users within 11 days of launch, a tenfold increase, according to Similarweb. However, the rapid growth has exposed severe compute infrastructure limitations. Monitoring platform SaaSHub has listed Muse as experiencing 'service degradation,' with users reporting frequent search failures. A researcher's stress test showed that when asked to generate 120 sub-agents, Muse only successfully created 33. Meta's Chief AI Officer Alexander Wang stated on September 9 that early usage far exceeded internal expectations, roughly ten times the beta test level, prompting Meta to reset token usage limits. The bottleneck is attributed to Muse's compute-intensive architecture, which allocates each user a dedicated virtual machine with 2 virtual CPUs, 8GB RAM, and 100GB SSD. Industry analysis suggests that if Muse reaches 100 million users within a year without resource sharing, Meta would need approximately 2 billion CPU cores, 800PB of RAM, and 10,000PB of SSD storage. The service is currently only available in the US and Canada, with no WhatsApp or Instagram user registration open yet. Pricing starts at a free tier of up to 100 million tokens per week, with paid subscriptions from $20 to $100 per month.