Back to the AI frontline? Meta surges 15% in a single week

Meta recorded its strongest weekly performance since February 2024 this week, driven by a series of AI strategy advancements. Market confidence in its AI capabilities and commercialization trajectory is clearly rebounding.

On Friday, Meta’s stock surged 6% in a single day, accumulating a weekly gain of 14.8%—the best single-week performance since at least February 2024—turning its year-to-date increase positive at around 1.4%.

The last time Meta recorded a similarly strong weekly performance was in February 2024, when investors responded positively to early results from the company’s “Year of Efficiency” cost-cutting plan, which aimed to restore financial discipline and rehabilitate the negative image caused by heavy metaverse investments.

This latest stock price breakout suggests that Meta may be gradually shedding its market label of “AI laggard,” potentially clearing the way for further acceleration of its AI strategy.

Wallstreetcn reported that on July 9, Meta launched its flagship model Muse Spark 1.1, which has now surpassed Google’s Gemini model in several test areas, including agent capability, programming, and multimodal tasks.

At the same time, according to Reuters, Meta is pushing forward with plans for mass production of its in-house chips and is significantly expanding its compute infrastructure. Deutsche Bank analyst Benjamin Black accordingly raised his projection for Meta’s potential incremental third-party cloud service revenue from $17 billion to $24 billion.

Research firm SemiAnalysis released a report predicting that Meta’s Meta Superintelligence (MSL) could, within the next six months, surpass Google in frontier AI capabilities, shifting the AI competitive landscape from the current Google/OpenAI duopoly to a triumvirate of Meta, OpenAI, and Anthropic.

The Muse Spark 1.1 model launched by Meta this week is its first commercialized model approaching frontier-level agent programming capabilities and is equipped with a paid API.

On Thursday, CEO Mark Zuckerberg posted on X, emphasizing that the model is priced “very cheaply,” fueling broad market speculation that Meta intends to initiate an AI inference price war to put pressure on its competitors.

The Meta Model API provides each account with a $20 free credit and charges on a pay-as-you-go basis: input price is $1.25 per million tokens, and output price is $4.25 per million tokens.

Richard Windsor, founder of Radio Free Mobile, pointed out in a research note on Friday that the launch of Muse Spark 1.1 corroborates recent reports about Meta’s plan to start a new compute power sales business. Windsor writes:

There is increasing evidence that, given the currently attractive returns, Meta will launch a new business selling compute power to third parties.

He further noted that Muse Spark’s AI programming capabilities are now close to top-tier models, “but priced at just 25% of their cost,” making it extremely attractive for the mass market.

According to Wallstreetcn, Meta plans to start mass production of its in-house AI chip, code-named “Iris,” in September this year. The chip was co-designed with Broadcom and manufactured by TSMC, passed testing in just six weeks, and Meta has signed multi-year supply agreements with Samsung, SanDisk, and Sumitomo Electric.

In terms of compute scale, Meta plans to deploy 7 gigawatts this year, doubling to 14 gigawatts by 2027.

Backing these targets are five gigawatt-scale “Titan” hyperscale data center clusters under simultaneous construction, along with Meta’s proprietary “AI-Backbone” network architecture, which enables asynchronous expansion of complex training tasks across thousands of kilometers.

Deutsche Bank analyst Benjamin Black stated in a Thursday report that these compute expansion plans mean Meta’s potential incremental revenue from third-party cloud services is around $24 billion, significantly higher than the previous $17 billion estimate.

He also noted that Meta’s in-house chip development could provide the company with a viable path to cost reduction and efficiency improvement.

Wallstreetcn reported that research firm SemiAnalysis believes Meta, after a year of aggressive capital investment and organizational restructuring, has the potential for MSL to overtake Google in frontier AI capability rankings within the next six months.

The report points out that the existing Google/OpenAI duopoly will be rewritten as a three-way competition among Meta, OpenAI, and Anthropic.

SemiAnalysis’ core thesis centers on the speed of compute scale expansion: Meta’s growth trajectory in AI compute will enable it to surpass the combined compute capacity of OpenAI and Anthropic by the end of this year.

Reuters, citing an internal memo, reported that Meta’s capital expenditure cap for AI infrastructure this year is as high as $145 billion.

On the talent side, Meta has reassigned 3,000 engineers to its internal reinforcement learning environment factory, building proprietary data that commercial data vendors cannot easily replicate, and invested $14.3 billion in Scale AI to help recruit top researchers from organizations like OpenAI and Anthropic.

SemiAnalysis believes that evaluating MSL solely on current benchmark test results is to “miss the forest for the trees”; what truly matters is future development potential, not just the starting point.

The report notes that if Zuckerberg maintains his current pace of capital investment, Google could potentially be permanently excluded from the global first tier of hyperscale AI players.