Meta's $835 Case: Why Bank of America Thinks the AI Spend Scare Is Overdone

Bank of America's call stands out while Meta stock still bears AI spend baggage

Meta shares are down more than 7% year to date, and investors remained skittish after management again raised capital spending guidance. Bank of America, however, is treating that reaction as stale. Earlier this month, the bank put Meta on its US 1 List and kept a $835.00 price target, or about 25% upside from current levels. The number alone is not the whole case, but it does show a visible bull case still in force even as the stock trades as if the spending debate remains unresolved.

The core of BofA's argument is not that Meta should spend less. It is that the cost structure behind that spending may have improved faster than investors expected. If each gigawatt of AI capacity is cheaper to build than feared, the margin-squeeze thesis becomes less urgent and the investment looks more like infrastructure with a possible payback path.

Why the cost argument matters more than the headline spend

The market has focused on the sheer size of Meta's AI buildout. BofA is shifting attention to cost efficiency. A recent internal Meta memo says the company plans to deploy roughly 6.5 gigawatts of AI compute capacity during 2026, with 5.5 gigawatts coming online in H2 2026. BofA says that scale implies construction costs of about $22 billion per gigawatt, roughly half its prior $45 billion per gigawatt estimate.

That does not make the spending harmless. Meta is still expected to spend $125 billion to $145 billion on AI infrastructure in 2026, and free cash flow fell from $26 billion in Q1 last year to $1.2 billion in Q1 this year. But lower build costs can widen the window for Meta to demonstrate ROI before the market writes off the program.

How Iris, enterprise demand, and cloud optionality strengthen the case

Lower build costs change the ROI math

When investors anchor on total capex, it is easy to miss efficiency gains. If Meta can add more compute for far less than feared, the risk of a brutal margin hit is not gone, but it is less automatic. That is the practical shift behind BofA's optimism.

Iris and MTIA could reduce reliance on third-party chips

BofA also sees a longer-term edge in Meta's custom silicon effort. Meta is preparing to manufacture its custom AI chip, code-named Iris, as part of the MTIA architecture, in September. The chip was developed with Broadcom for design and TSMC for manufacturing, and BofA says it is meant to reduce reliance on expensive external suppliers over time.

That would matter most if it lowers per-workload costs after the current construction wave. It would not matter if the chips slip, underperform, or sit underused.

Enterprise and cloud demand provide a possible release valve

BofA is also pushing back on the idea that excess capacity would be pure waste. The bank says enterprise AI demand could create a more durable revenue stream, and Zuckerberg has said Meta may move into cloud computing if the buildout leaves spare capacity. He has also pointed to outside interest in Meta's APIs and computing resources.

BofA is not arguing Meta will displace the leading cloud platforms. Its point is narrower: even a modest share of a large enterprise AI market could be financially meaningful if the infrastructure is already in place.

What could validate the thesis - and what could break it

Meta has already shown signs of usage around its open-source model strategy. The company says Llama models are approaching 350 million downloads. Its own Llama 4 materials also claim best-in-class multimodal performance in certain comparisons. None of that settles the capex debate, but it does support the view that Meta is trying to build demand alongside capacity.

What investors should watch next

• September manufacturing of Iris as the first hard execution checkpoint for custom silicon.
• 5.5 gigawatts coming online in H2 2026: timing matters as much as cost.
• enterprise AI solutions and cloud capacity market expected to top $1 trillion by 2028: even early signs that Meta can use excess capacity here would matter.
• Evidence that better models are improving developer engagement, product usage, or advertising relevance.

What could invalidate the setup

• Iris manufacturing slips or fails to deliver the expected efficiency gains.
• The 2026 capacity rollout looks slower or less efficient than expected.
• Demand does not arrive fast enough to justify the scale of the buildout.
• building a cloud business is not easy turns out to be the more useful guide than the optionality case.

BofA's $835.00 target is really a call on economics, not just conviction. The bank is arguing that cheaper capacity and possible enterprise or cloud monetization make Meta's AI spend look less extreme than the market's current fear response. If that efficiency story and the demand story both hold, the stock has a clearer path to repricing.