RSI Compute Demand Could Fill $1.9 Trillion AI CapEx Gap

AI infrastructure pricing has shifted from penalizing cloud CapEx to rewarding revenue, as the sector's June-July correction erased 39 percent from Korea's KOSPI and opened a new compute demand narrative around recursive superintelligence.

"CapEx alone no longer moves the needle — investors now demand proof that spending converts into orders, revenue, and profit," Liu Chenming, chief strategy analyst at GF Securities, said in a research note published Aug. 9.

The shift was visible in the July earnings cycle. Amazon and Microsoft rebounded 21.1 percent and 28.0 percent from their lows after raising CapEx guidance while beating revenue and operating profit estimates. Meta fell 7.95 percent the day after its report, with free cash flow down 91 percent year over year. The four major cloud providers now project 2026 CapEx of roughly $175 billion at Microsoft, $220 billion at Amazon, $195-205 billion at Alphabet, and $130-145 billion at Meta — a combined increase of $25 billion from earlier guidance.

The critical question is whether growth can be sustained. Cloud CapEx growth is projected to slow from 105 percent in 2026 to 45 percent in 2027 and just 9 percent in 2028. Maintaining the 2026 pace would require an additional $1.94 trillion in 2028 spending — a gap that GF Securities argues could be filled by model companies and new cloud providers pursuing recursive superintelligence, or RSI, where AI models train other models at unprecedented scale.

Cloud CapEx Growth Slows From 105% to 9% by 2028

The correction began June 22 when South Korean regulators signaled they would revisit leveraged ETF approvals, triggering forced liquidations across the region. Apple's June 25 price increases on Mac and iPad models — attributed to memory costs — pushed the Philadelphia Semiconductor Index down 5.29 percent and the Nikkei 225 down 4.15 percent the next day. The selloff spread to A-shares, with the ChiNext index only modestly recovering since.

The four major cloud providers' full-year CapEx guidance now ranges from about $70 billion at Oracle to roughly $220 billion at Amazon. Microsoft's guidance was revised from $190 billion to $175 billion for calendar 2026, but the reduction reflects an accounting reclassification of data center leases from finance to operating — not a cut in equipment purchases. Alphabet's quarterly free cash flow turned negative for the first time in over a decade, and Amazon reported $8.82 billion in negative quarterly free cash flow.

RSI Opens a Second Wave of Application Demand

GF Securities projects that model companies and new cloud providers — including OpenAI's Stargate program, Anthropic, xAI, and Coreweave — will add roughly $461 billion in CapEx in 2027 and $525 billion in 2028. OpenAI and Anthropic have annual recurring revenue exceeding $47 billion and $30 billion respectively, while xAI's 2027 AI CapEx is projected at $73.9 billion. Nvidia's approximately $5 billion strategic investment in SSI, a company pursuing recursive superintelligence, has strengthened the model-company compute demand thesis.

The RSI narrative extends beyond training demand. Current AI agents operate at L1 — executing coding and debugging subtasks per human instructions. At L2-L3, models would autonomously modify their own constraints and algorithms, with current-generation models participating in training the next generation. This could open a second wave of application demand, expanding from coding agents to enterprise-level work agents, similar to how 4G networks unlocked new application categories after 2G and 3G.

The market has already begun pricing this transition. Microsoft gained $720 billion in market cap over three sessions from July 29 to Aug. 3, while the four cloud giants added a combined $1.86 trillion. Microsoft trades at 24.9 times forward earnings versus 21.2 for the S&P 500, a premium investors appear willing to pay for the company's 45 percent operating margin and $19.6 billion in quarterly free cash flow.

But GF Securities cautions that stabilization does not mean a return to one-way rallies. The sector is likely entering a high-volatility, high-rotation phase, with the first cyclical top already formed. The rotation path runs from hardware to cloud providers to model companies to applications. The key metrics to track: model company ARR growth, whether cloud revenue growth can cover expanding CapEx, and whether token price declines translate usage growth into revenue. A true phenomenon-level application with user scale, retention, and monetization has not yet emerged — but it remains the endpoint of the entire narrative chain.

This article is for informational purposes only and does not constitute investment advice.