Wall Street’s enthusiasm around Meta’s recent stock rally extends beyond the headlines praising its Muse Spark 1.1 model and a move to a paid developer approach. The more consequential development is a technical revelation tucked into an internal memo that surfaced publicly, showing a capacity roadmap and implied cost structure that materially alters the economics of Meta’s AI buildout.
The leaked document, reviewed by analysts at BofA Securities, outlines a multi-year compute expansion that, if accurate, suggests Meta is achieving capacity at a much lower cost per unit than the Street had assumed. This conclusion helped BofA analyst Justin Post to maintain a Buy rating on Meta Platforms (NASDAQ: META) and keep an $835.00 price target in place.
What the memo says
- Scale - The company is reportedly planning to add about 14 Gigawatts (GW) of total compute capacity across 2026 and 2027.
- Near-term deployments - Meta has deployed 1GW so far in 2026 and expects to add another 5.5GW in the second half of the year, implying 6.5GW of capacity growth for 2026.
BofA’s read of these figures drives a striking recalibration of Meta’s per-GW build cost. The bank had previously modeled roughly $45 billion per GW to achieve comparable capacity. Combining the memo’s capacity outlook with Meta’s expected $145 billion overall capex, the effective cost implied by the memo is closer to $22 billion per GW.
Why that matters
For months, sceptics argued that Meta’s AI ambitions would require enormous capital outlays that could sap cash without delivering clear returns. The new capacity math flips that narrative: if Meta can expand compute at roughly half the cost anticipated by Wall Street, the return profile of its AI investments looks substantially more attractive.
BofA frames the potential upside by comparing estimated returns across cloud customers. The bank’s analysis contrasts Meta’s potential per-GW economics with estimated annual cloud revenues per GW for major providers, which it pegs at $10-16 billion for Amazon and Google, and with recent SpaceX capacity deals that were characterized as ranging $40-50 billion per year per GW. In this context, achieving build costs below $30 billion per GW could yield favorable economics for Meta relative to peers and recent market deals.
Custom chips and timing
The memo and subsequent reporting also describe Meta’s chip roadmap. After successful testing, the company plans to begin manufacturing a custom chip code-named Iris in the fall, working with Broadcom and TSMC. Meta reportedly intends to introduce new custom chips roughly every six months through 2027 and has pursued multi-year supply agreements with partners including Broadcom and TSMC.
Crucially, BofA stresses that Iris does not appear to be the driver of the 2026 cost improvements. Since Iris manufacturing begins in September, the bank reasons that the cost efficiencies reflected in the 2026 capacity numbers are happening independently of the custom-chip roll-out, making the chip program an incremental long-term positive rather than the proximate source of near-term savings.
Analyst perspective
Justin Post highlighted the gulf between his team’s prior estimates and the memo’s figures, noting that the memo’s 6.5GW 2026 capacity build is markedly above BofA’s internal projection of 2.6GW. If the memo’s numbers are even close to correct, Post wrote, Meta may have engineered material cost savings that push capacity cost per GW well below both BofA and broader Street expectations.
From an investor standpoint, that combination of faster-than-expected capacity growth and lower implied capital intensity helps address the principal concern that has weighed on Meta’s valuation: whether massive capex will generate convincing returns. The memo suggests Meta is not merely chasing scale at any price but pursuing a more efficient, vertically integrated approach to building AI infrastructure.
Market reaction and framing
The market reaction reflects a reappraisal of Meta’s infrastructure economics rather than celebration of product-level announcements. While media attention has centered on Muse Spark 1.1 and the change to a paid developer model, the under-the-hood details in the memo appear to be the dominant force behind renewed investor confidence.
Equally important, the chip program remains a strategic positive. BofA views progress on custom silicon as supportive of long-term margin contribution from specialized processors, even if the initial manufacturing start date places Iris outside the primary driver of 2026 savings.
Bottom line
The leaked internal memo reframes the debate around Meta’s AI spending by suggesting materially lower per-GW build costs than previously modeled. That shift in the underlying economics explains much of the recent upward movement in the stock and has convinced at least some of Wall Street’s largest bulls to reiterate bullish ratings and elevated price targets. For investors and sector observers, the critical takeaway is that Meta’s competitive story may now rest as much on infrastructure efficiency and vertical integration as on software milestones and model releases.