Meta Platforms shares are gaining momentum after investors focused on evidence suggesting the company is building its artificial intelligence (AI) infrastructure far more efficiently than previously expected, easing concerns over its record capital expenditure programme.
While recent attention has centred on Meta’s new Muse Spark 1.1 AI model and its move towards a paid developer model, analysts say the stronger driver of investor optimism is an internal company memo first reported by Reuters, which points to significantly lower costs for expanding the company’s AI computing capacity.
BofA Securities analyst Justin Post reiterated a Buy rating on Meta and maintained a price target of $835 following the Reuters report, arguing that the company’s infrastructure economics could materially improve long-term returns on investment.
According to BofA’s analysis, Meta plans to add approximately 14 gigawatts (GW) of computing capacity during 2026 and 2027. The internal memo indicates that the company has already deployed 1GW of capacity in 2026 and expects to add a further 5.5GW in the second half of the year.
The figures imply that Meta is building new AI infrastructure at a considerably lower cost than previously estimated.
BofA had earlier projected that each gigawatt of AI computing capacity would cost around $45 billion to build. However, based on the capacity outlined in the memo and Meta’s expected capital expenditure of about $145 billion, the investment bank estimates the actual cost is closer to $22 billion per gigawatt.
Post said the reported expansion significantly exceeded BofA’s previous estimates, suggesting Meta may have achieved substantial cost efficiencies that are well ahead of broader market expectations.
The development addresses one of the principal concerns surrounding Meta’s AI strategy. Investors have questioned whether the company’s massive spending on data centres and computing infrastructure would generate sufficient financial returns.
BofA believes the improved economics could fundamentally alter that narrative.
The bank noted that if Meta can continue building AI capacity for less than $30 billion per gigawatt, the economics compare favourably with estimated annual cloud revenues generated per gigawatt by rivals such as Amazon and Google, while also appearing attractive relative to recent AI infrastructure agreements involving SpaceX.
The Reuters report also revealed that Meta plans to begin manufacturing a custom AI chip, codenamed Iris, in partnership with Broadcom and TSMC later this year to complement its purchases of graphics processing units (GPUs).
However, BofA believes the custom chip is unlikely to be responsible for the cost savings already evident in 2026, as production is only expected to begin in September.
Instead, the bank argues that Meta’s existing infrastructure improvements are delivering efficiencies independently of its custom silicon programme, making the company’s long-term chip strategy an additional positive rather than the primary driver of current gains.
According to Reuters, Meta intends to introduce new generations of custom AI chips roughly every six months through 2027 while securing multi-year supply agreements with key manufacturing partners, including Broadcom and TSMC.
BofA said continued progress in custom chip development should strengthen Meta’s AI competitiveness over the longer term, improve infrastructure margins and reinforce management’s confidence that the company’s substantial AI investments will generate attractive returns.
The latest developments suggest investors are increasingly rewarding Meta not simply for expanding its AI software capabilities, but for demonstrating that it can build the underlying computing infrastructure more efficiently than previously expected, potentially strengthening both profitability and its competitive position in the global AI race.
