Meta Platforms closed at $669.21 on Friday, up 5.97% on the day. The stock touched $677.85 intraday before easing into the close. That capped a stretch that saw the stock climb about 15% over the week, its best five-day run in months.
The proximate trigger was a report card, not an earnings release. Independent chip-and-AI research firm SemiAnalysis published an assessment of Meta's push toward advanced AI and gave CEO Mark Zuckerberg's year-long overhaul high marks. The firm wrote that Meta had "finally finished paying down" the cost of rebuilding its AI organization, and called Meta "the only hyperscaler" on track to be world-class at the three things a frontier model needs: data, talent, and compute.
From Llama 4 to a rebuilt team
The turnaround matters because the market had written Meta's AI effort off. Last year's Llama 4 model was widely viewed as a disappointment. Zuckerberg responded by tearing the AI group down and rebuilding it, a move that cost time and drew steady skepticism through the spring.
That skepticism is now reversing. This week Meta released a model called Muse Spark, which Zuckerberg described as a strong agentic and coding model priced well below rivals. Early reviews put it in the same conversation as top systems from OpenAI and Anthropic. For a company that spent a year absorbing setbacks, a competitive model plus an outside endorsement was enough to move a stock this size 6% in a session.
The SemiAnalysis note added one caveat worth keeping in view. It said Meta's revamped team is "still basically at step 1." The direction has changed. The results have not yet shown up in products at scale.
The bigger prize is Meta Compute
The more durable part of this story is not a single model. It is what Meta plans to do with the enormous computing footprint it has been building. Reports this week point to a new unit, Meta Compute, that would sell computing power and models to outside customers directly.
That would drop Meta into the cloud business alongside Amazon's AWS, Microsoft Azure, and Alphabet's Google Cloud, and put it in direct competition with AI-cloud specialist CoreWeave, a roughly $48 billion company whose entire model is renting out AI compute. It would also reframe Meta's capital spending. For two years investors treated Meta's data-center buildout as a cost to be tolerated. A cloud unit turns that same capacity into a potential revenue line.
The math is why traders care. Wolfe Research estimated that every gigawatt of compute Meta monetizes at roughly a $25 billion annual rate could lift earnings per share by about 20%. Meta operates data centers measured in many gigawatts, so even partial success would be material to a company already earning at an 82% gross margin.
What the valuation says
Even after the run, Meta trades near 24 times earnings, cheaper than most of the megacap AI group. The stock carries a market value around $1.7 trillion and pays a token dividend yielding about 0.33%. Its 52-week range runs from $520.26 to $796.25, so Friday's close still sits well below the high set earlier in the cycle.
The read-through hit the rest of the group unevenly. Nvidia, Meta's main chip supplier, rose about 4% to $210.96 as the AI-spending story firmed. Amazon slipped less than 1%, and Oracle fell about 2.5%, a reminder that a new cloud entrant is a threat to incumbents as much as a validation of the theme.
The next real test is Meta's second-quarter report later this month. Investors will want to see the AI narrative show up in two places: guidance that reflects rising engagement and ad performance, and any concrete detail on how, and when, Meta Compute starts billing customers. A report card buys a week. Numbers buy a re-rating.
