Industrial Securities: Meta's Leasing and Selling of Computing Power Should Not Be Interpreted as Overcapacity
Claire Weston
After Meta announced plans to lease spare compute, global tech stocks sold off again. Industrial Securities argues this is not an oversupply story — the real drivers of the pullback are fragile positioning and an earnings-visibility gap, not a deterioration in compute fundamentals. The dip, the firm says, opens a positioning window.
Is Meta selling compute because it has too much?
Industrial Securities says no, citing four reasons: the news is not new — reports surfaced as early as May; this week alone, Google was reported restricting Meta's access and Meta signed a compute deal with Crusoe. This means → Meta's own capacity is still short of what it needs.
Meta monetises AI through advertising, not cloud services. In plain terms = its business model differs from AWS or Azure — exploring cloud revenue is about adding a cash-flow stream, not pivoting the core business.
Hardware demand is driven by the inference side — models serving live user requests — not by redundant training purchases. This reflects a structural shift in where compute demand sits, not a contraction in total volume.
What does Wall Street think? How wide is the disagreement?
Deutsche Bank argues Meta is leasing older, non-core compute to recoup cash; high-end Rubin-series resources remain reserved for its long-term superintelligence strategy.
Jefferies draws a parallel to early-stage Amazon AWS: Meta still has 35% idle compute, and reselling it gives the company more firepower for future capex.
Morgan Stanley's estimate is more aggressive — if Meta materially pursues cloud services, capex could surprise to the upside. Its model forecasts $175 billion in 2027 and $205 billion in 2028.
Why is the 2027 capex range so wide?
JPMorgan notes the market's 2027 capex estimate for Meta spans $175 billion to $275 billion. This means → the gap between the highest and lowest forecasts is $100 billion — roughly the entire market cap of a mid-size tech company.
The split is clean: hardware analysts lean toward no capex cuts, while internet-sector professionals lean toward contraction. Put simply = the supply-side view and the demand-side view reach opposite conclusions.
How does this pullback compare to history?
Industrial Securities benchmarks two eras. In the 1995–2000 dot-com run, aside from a nearly 30%, ~80-day correction triggered by the 1998 Asian financial crisis, each pullback averaged roughly 10% over ~30 days.
In the post-ChatGPT AI run since 2022, aside from the 20%+ corrections triggered by DeepSeek and trade-war headlines, each pullback also averaged roughly 10% over ~30 days. This reflects a remarkably stable "normal volatility" band across major tech cycles.
The firm's bottom-fishing signals: watch for a maximum drawdown of 10% and implied volatility above 30% — when both thresholds are hit, the dip is more likely a positioning window than a trend reversal.
What comes next? Where is the key verification point?
Industrial Securities attributes this correction to fragile positioning, an earnings-visibility gap — high-frequency data on cloud providers' capex plans and ROI is scarce compared to hardware — and a liquidity headwind, rather than any real deterioration in compute fundamentals.
Mid-to-late July is the critical checkpoint: U.S. earnings season will re-confirm whether the sector's earnings advantage holds, and inflation data will show whether the peak is in. In plain terms = if both earnings and inflation flash green, tech likely begins a new leg higher; if not, the correction could extend.
Content is for reference only, not financial advice.