Muse not only boosted Meta but also gave the CPU sector a lift. This is because Muse handles a ton of high-frequency tasks, like cross-platform price comparisons (up to 146 searches for a single trip itinerary), processing massive backlogs of emails, and automated phone negotiations. Since Meta proactively provides persistent cloud execution environments for user agents, relying on headless browser automation and high-frequency API orchestration, the computational workload has significantly shifted back to CPUs. The market now expects that if Muse becomes widely adopted within Meta's social ecosystem, it will drive massive demand for server CPUs. Regarding Muse's significance to Meta, we talked about this back in early July: 'The real value and differentiation for Meta in the future lies in "proprietary models + proprietary user scenarios + proprietary ad/social data loops + proprietary inference infrastructure." The key point for Meta at this stage is the performance of its models. If they can make a breakthrough, the game-changer in the future will be the Agent. Of course, if the models don’t fully take off, Meta could just focus on building Agents, similar to Tencent's current approach. Whoever has better computing power and more cost-effective tokens will get the business—there’s still a chance. Meta pushing Agents also has its advantages.' Looking at it now, Meta's large model hasn’t fully caught up yet, but Meta’s Agent seems to be showing signs of overtaking the competition. https://(x.com)/qinbafrank/status/2073249685780000833?s=46&t=k6rimWsEbo2D2tXolYcM-A