This week, Meta Platforms (META) recorded its strongest weekly performance since February 2024, driven by a series of strategic advancements in artificial intelligence. The rally reflects significantly renewed market confidence in its AI capabilities and commercialization pathways. On Friday alone, Meta shares rose 6%, bringing the weekly cumulative gain to 14.8% and pushing its year-to-date stock performance back into positive territory at approximately 1.4%. The last comparable weekly surge occurred in February 2024, when the market responded favorably to its “Year of Efficiency” cost-cutting plan.
This breakout may signal that Meta is gradually shedding its market label as an “AI laggard,” opening up room for further acceleration of its AI strategy.
On the model front, Meta launched its flagship model, Muse Spark 1.1, on July 9. This is its first commercial model with near-frontier-level agentic programming capabilities, equipped with a paid API interface. Across multiple test benchmarks—including agent capability, coding, and multimodality—the model has already surpassed Google’s Gemini model. CEO Mark Zuckerberg emphasized on social media that the model’s pricing is “very low,” sparking widespread speculation that Meta is initiating an AI inference pricing war to pressure competitors. The API provides each account with a $20 free credit and operates on a pay-as-you-go basis, with input and output prices set at $1.25 and $4.25 per million tokens, respectively. Some analysts have pointed out that Muse Spark’s AI coding capabilities are approaching those of top-tier models, but at only one-quarter of the price, making it highly attractive to the mass market.
On the infrastructure side, Meta is actively pushing forward with the mass production of its self-developed chips and compute capacity expansion. The company plans to begin mass production of its in-house AI chip, codenamed “Iris,” this September. The chip is co-designed by Broadcom and manufactured by TSMC, and Meta has already signed long-term supply agreements with multiple partners. In terms of compute scale, Meta plans to deploy 7 gigawatts of capacity this year and double that to 14 gigawatts by 2027. To support this, the company is simultaneously constructing five gigawatt-scale “Titan” hyperscale data centers and deploying its own networking architecture to enable asynchronous scaling for complex training tasks.
Based on this expansion plan, Deutsche Bank analysts raised their projected incremental revenue from Meta’s third-party cloud services from $17 billion to $24 billion, noting that self-developed chips could open up a viable path for cost reduction and efficiency improvement.
Research firm SemiAnalysis released a report arguing that after a year of aggressive capital expenditure and structural reorganization, Meta Superintelligence could surpass Google in frontier AI capability rankings within the next six months, potentially shifting the current duopoly of Google and OpenAI toward a tripartite balance among Meta, OpenAI, and Anthropic. The core basis for this judgment lies in the pace of compute expansion—Meta’s growth trajectory in AI compute scale is expected to surpass the combined total of the other two by the end of the year. According to Reuters, Meta’s capital expenditure ceiling for AI infrastructure this year is as high as $145 billion. Additionally, the company has reassigned thousands of engineers to its internal reinforcement learning team and has invested heavily in recruiting top researchers from rival institutions. The report contends that evaluating MSL solely on current benchmark results is “missing the forest for the trees,” and that the true key lies in its momentum. If the current investment pace is maintained, Google could be permanently excluded from the top tier of global AI hyperscale players.
