Why meta stock surged on a hidden AI cost breakthrough
Meta stock climbed about 3% early Friday after Bank of America flagged a hidden AI infrastructure cost breakthrough: an internal memo reported by Reuters suggests Meta can add compute far more cheaply than Wall Street assumed, with buildout costs nearer $22 billion per gigawatt instead of $45 billion. That shift is easing fears that Meta's massive AI spending will burn cash without returns.
BofA analyst Justin Post reiterated a Buy rating and an $835 price target on Meta Platforms (NASDAQ: META). The catalyst was not flashy software headlines—it was capacity economics buried in corporate planning documents that investors had been waiting to see.
Key Takeaways
- Meta stock rose after BofA said memo data reviewed by Reuters implies far lower per-gigawatt AI build costs than prior estimates.
- Meta targets about 6.5 gigawatts of AI compute in 2026, including 5.5 gigawatts in the second half of the year.
- Projected $145 billion in 2026 capital spending supports economics near $22 billion per gigawatt, versus BofA's earlier ~$45 billion assumption.
- CEO Mark Zuckerberg says exploring an AI cloud business makes sense, though Meta still uses all current computing capacity.
- Custom chip Iris, with Broadcom and TSMC, may strengthen long-term strategy but likely does not drive 2026 savings.
Why did Meta stock surge on Friday?
Shares moved after BofA Securities tied the rally to infrastructure math, not product launches alone. Post said the memo's 2026 capacity growth is well above prior BofA estimates, suggesting Meta engineered significant cost savings per megawatt.
For months, the bear case held that Meta's AI capex would incinerate cash. Friday's reaction flipped that script: if capacity can be built below $30 billion per gigawatt, the spending looks far more investable relative to cloud peers.
What did the internal memo reveal about AI costs?
According to BofA's analysis of Reuters reporting, Meta is working to add 14 gigawatts of total compute capacity across 2026 and 2027. The company deployed 1 gigawatt in 2026 so far and expects another 5.5 gigawatts in the second half.
Against Meta's expected $145 billion capex, actual costs track closer to $22 billion per gigawatt—roughly half what BofA previously modeled. Post noted building below $30 billion per gigawatt could compare favorably to Amazon and Google cloud revenue per gigawatt, and to recent large-scale compute deals elsewhere in tech.
Could Meta Compute turn spending into revenue?
Separately, Zuckerberg told Bloomberg that exploring an AI cloud business makes sense, putting his name behind the reported Meta Compute initiative. The plan could offer hosted models akin to AWS Bedrock or raw GPU capacity like neocloud providers.
SemiAnalysis notes Meta contracted more than 5 gigawatts across cloud and colocation in just the first half of 2026. Zuckerberg cautioned Meta is not sitting on surplus silicon—no one in the industry feels they have too much compute—but said high external demand could justify leasing capacity under long-term contracts.
What role does Meta's Iris chip play?
Reuters also reported Meta plans to begin manufacturing its custom AI chip, code-named Iris, with Broadcom and Taiwan Semiconductor Manufacturing later this year. BofA said Iris is unlikely to explain 2026 cost improvements, since manufacturing starts in the fall.
Still, a broader custom silicon roadmap extending into 2027 could support margins over time. For now, Wall Street's meta stock bid rests on proof that Meta is building an efficient AI factory—not just buying GPUs at any price. Follow more coverage in our Future Tech & AI Wonders section, and see the full Yahoo Finance report.