Overview
Morgan Stanley has evaluated Meta's (NASDAQ: META) contemplated move into cloud services and identified two distinct routes the company could pursue: a hosted API and model-access product comparable to existing hyperscaler offerings, and a neocloud-style approach focused on leasing raw silicon and compute capacity. The bank sets out five principal takeaways that map the monetization paths, the technical and execution risks, and the financial implications for Meta as it scales compute capacity over the next two years.
Take 1 - Execution risk: neocloud easier than a full hyperscaler stack
Morgan Stanley judges the neocloud option to be the more tractable near-term play for Meta. The analysts contrast that with the full hosted API and model-access business, which they view as carrying greater technology, hiring, and execution risk. A complete API offering - including models and application tooling - would require significant product maturity and integration and should be seen as a "show me" proposition relative to long-established hyperscalers.
The bank notes that performance on coding-focused benchmarks such as TerminalBench and SWE Bench Verified matters for third-party developer adoption, and that Meta's Muse model suite has not historically scored well on those tests. In Morgan Stanley's view, competing with frontier models like Gemini will demand substantial improvement from Muse models if Meta pursues a hosted API strategy.
Take 2 - Compute supply and leasing opportunity
Morgan Stanley models Meta increasing owned and operated IT capacity by roughly 2 GW in 2026 and about 3.5 GW in 2027, from an estimated year-end 2025 base of approximately 3 GW. Given that build profile, the analysts see an opportunity for Meta to lease compute capacity at least temporarily.
They observe that many customers remain compute constrained and make budgeting tradeoffs between training, inference, and other product uses while balancing near-term returns and long-term opportunities. As a point of comparison, Morgan Stanley expects hyperscalers such as Amazon (NASDAQ: AMZN) and Google (NASDAQ: GOOGL) to add about 5 GW and 9 GW of IT capacity, respectively, in 2027. That differential frames the theoretical opportunity for Meta to have surplus compute available to lease.
The bank clarifies that Meta likely will not be able to lease the roughly 2.5 GW the company currently rents from third parties such as CoreWeave (NASDAQ: CRWV), Nebius (NASDAQ: NBIS), GCP (NYSE: GCP), Oracle (NYSE: ORCL) and others. However, Morgan Stanley notes Meta would retain flexibility to monetize its own 1P capacity that it builds.
Take 3 - EPS implications from neocloud leasing
Morgan Stanley calculates that the neocloud approach could produce notable earnings-per-share accretion for Meta. The firm describes recent neocloud contracts as flexible, often smaller in scale and shorter in duration, with mutual opt-out clauses. Given that scarcity, the bank assigns material upside to monetizing Meta's freshly added capacity.
Concretely, Morgan Stanley estimates that every 250 MW leased for one year at a price point of $40 per Watt would add about $3, or roughly 8 percent, to its 2028 EPS estimate for Meta.
Take 4 - Strategic framing: neocloud as an EPS bridge while product bets scale
The firm's Overweight rating on Meta rests primarily on the company’s ability to develop new products that promote multi-year engagement and revenue growth across its platforms. Morgan Stanley positions the neocloud option largely as an "EPS bridge" that can provide near-term financial benefit while Meta’s core products scale.
The bank highlights a slate of product initiatives it views as central to achieving durable growth: MetaAI, business agents and messengers, diffusion offerings, and subscription revenue streams among others. Evidence of the development and scaling of these businesses will be the determining factor for the company to outperform and command a higher valuation multiple, in Morgan Stanley’s view.
Take 5 - Capex trajectory tied to capacity buildout
In Morgan Stanley’s model, Meta’s capital expenditures rise materially as the firm brings on about 3.5 GW of capacity in 2027. The bank projects capex reaching $175 billion in 2027 and $205 billion in 2028, up from $145 billion in 2026. These figures assume the new capacity is deployed primarily for Meta’s own first-party products rather than to operate a full hyperscale business offering an API and a comprehensive suite of models and tools.
While Morgan Stanley views a neocloud as a stopgap rather than a permanent scaled business, pursuing these leasing opportunities does require additional data center capacity. As a result, the analysts expect their 2027 and 2028 capex estimates to be biased upward if Meta follows this path.
Conclusion
Morgan Stanley’s assessment frames the neocloud option as a lower execution-risk way for Meta to monetize incremental compute while the company continues to develop product lines that could sustain long-term engagement and revenue growth. The hosted API and full hyperscaler product present higher execution challenges, especially given current benchmark performance of Meta’s Muse models. Financially, leasing even modest blocks of newly built capacity could meaningfully lift EPS in the near term, but it also implies higher capex as Meta expands its owned infrastructure.
Note: All projections and model figures above reflect Morgan Stanley’s estimates as described in their analysis.