Leaked Meta Memo Reveals Plans to Double AI Computing Capacity to 14 Gigawatts

In a recently leaked memo, Meta has disclosed ambitious plans to significantly expand its computing capacity, aiming to double it to 14 gigawatts (GW) by 2027. This expansion includes a substantial deployment of 7 GW scheduled for this year alone. The memo outlines not only the scale of this initiative but also the strategic long-term supply contracts that Meta has secured for essential components such as memory, flash storage, and fiber-optic equipment. These contracts come at a time when a global memory shortage has begun to affect consumer hardware prices, raising concerns about the overall sustainability of such rapid growth.

Meta's Silicon Strategy and Supply Agreements

The memo reveals that Meta's in-house AI accelerator, Iris, is set to enter production at the Taiwan Semiconductor Manufacturing Company (TSMC) this September. This follows a successful six-week bug validation process, indicating that the project is on track. Broadcom continues to serve as the design partner for these chips, with an agreement that extends through 2029. Meta plans to release a new chip approximately every six months until 2027. Importantly, these chips are intended to complement, rather than replace, the externally sourced graphics processing units (GPUs) that Meta has been using.

Additionally, Meta has secured a multiyear agreement with AMD for up to six gigawatts of Instinct accelerators. The memo notes that the integration of the latest external GPUs has proven to be challenging and time-consuming at Meta's scale, which underscores the complexity of managing such a massive computing infrastructure.

The memo also highlights that Meta has established long-term contracts for critical components, including memory from Samsung, flash storage from Sandisk, and fiber-optic equipment from Sumitomo Electric. These agreements are particularly noteworthy given the ongoing memory shortage in the market, which has led to increased costs for consumer hardware.

Market Reaction and Financial Implications

Following the leak of the memo, Meta's stock experienced a partial recovery in the morning but remained in negative territory as investors grappled with the implications of such a large-scale expansion. Analysts have pointed out that the memo's content runs counter to the market's recent preference for capital discipline, especially after a period where investors had rewarded Meta for its more conservative financial strategies. Earlier this month, reports revealed that Meta was launching an internal cloud business called Meta Compute to sell surplus capacity, which had previously led to a nearly 9% increase in its stock price.

However, the stock price took a hit after CEO Mark Zuckerberg acknowledged in leaked remarks that the development of AI agents had not progressed as quickly as anticipated. This context makes the memo's announcement of a doubling of capacity and a six-month cadence for new chips appear more ambitious, potentially raising concerns about Meta's financial commitments.

The memo estimates that the cost of establishing a one-gigawatt AI data center could range from $50 billion to $100 billion. Therefore, Meta's planned incremental capacity of 7 GW could translate to a staggering investment between $350 billion and $700 billion. Furthermore, Morgan Stanley's analysis suggests that Meta's reported capital expenditures may not fully capture the extent of its financial commitments due to various purchase obligations and construction projects that are still in progress.

Implications for AI Capex Boom and Future Financing

Looking ahead, Morgan Stanley forecasts that Meta's free cash flow could remain flat or even negative by 2026, amidst off-balance-sheet commitments that are approaching $1.8 trillion across the hyperscaler sector. Goldman Sachs has projected that capital expenditures for hyperscalers could reach up to $1.1 trillion by 2027, with a potential upside of $1.4 trillion. This raises significant questions about how Meta will finance its ambitious expansion plans, especially given its current capex guidance of $125 billion to $145 billion.

Additionally, the rapid expansion of AI data centers poses challenges for the U.S. power grid, which was not originally designed to handle such high levels of demand. Analysts have noted that a single query to AI models like ChatGPT consumes about ten times more energy than a standard Google search. Local officials have raised alarms about the environmental and infrastructure implications of increased data center construction, warning that it could lead to blackouts and water shortages.

In response to these concerns, the Trump administration is reportedly developing policies that would require major tech companies to cover the costs associated with electricity, water, and grid infrastructure for their expanding AI data centers. As Meta moves forward with its plans, the intersection of technology, finance, and environmental sustainability will be crucial to monitor, as these factors will significantly influence the future of AI development and deployment.