Meta’s Muse Could Require Massive Compute Spending at Facebook Scale
An estimate cited by Daytona CEO Ivan Burazin says Meta’s Muse would need 65,000 CPUs and 75PB of DRAM, about $2.8 billion in infrastructure, for every 100 million users. Scaling that to Facebook’s 3 billion users implies roughly $84 billion, exceeding Meta’s 2025 capex of $69.7 billion. The memory requirement alone could consume about 5% of annual global DRAM output, according to a widely shared estimate.
Daytona CEO Ivan Burazin posted on X that serving 100 million Muse users would take 65,000 CPUs and 75 petabytes of DRAM, or roughly $2.8 billion in infrastructure. Box CEO Aaron Levie and Elon Musk amplified the claim, with Musk saying people underestimate the scale involved.
Meta's capital spending rose from $18.7 billion in 2021 to $69.7 billion in 2025. A Facebook-sized Muse, at 30 times that 100 million benchmark, implies about $84 billion — more than that full year of capex. The estimate appears to assume fewer than one in ten allocated cloud computers runs at once.
If such projections hold, the memory and compute demands of consumer-scale AI assistants could ripple well beyond Meta. Chipmakers such as Micron may benefit from sustained pricing power, while cloud providers and device makers could face tighter component supplies. Ordinary users might see the tradeoff in subscription costs, feature limits, or slower rollouts of free tiers. Smaller developers, unable to match such infrastructure budgets, could find themselves at a widening disadvantage.