Social Graph VC: Data Center Capital Spending to Exceed $1 Trillion by 2026

Huo Xing Cai Jing reports that venture capital firm Social Graph VC recently published an introductory market analysis stating that data center capital expenditures are projected to exceed $1 trillion in 2026, roughly double the 2025 spending of the four major hyperscale cloud providers. Jensen Huang anticipates annual spending could reach $3 to $4 trillion by the end of this decade, and potentially $10 trillion by 2031. AI investment currently accounts for approximately 0.9% of global GDP and is expected to rise to about 1.4% by 2028. The article notes that training each of the Fable 5 and GPT-5.6 models consumes less than 2 GW of power and costs approximately $120 billion. Each generation of models can recoup its own costs; laboratory losses stem from scaling the next-generation models by a factor of 10. Computing cannot be purchased at a single price point: the market rental rate for H100 is around $1.95 per hour, while hyperscale cloud providers charge $8–9 per hour, with hourly rates fluctuating by up to 137% annually. Of the approximately $3 trillion in global data center spending projected through 2028, there is a $1.5 trillion funding gap. The article also states that CME plans to list cash-settled monthly futures contracts for H100 and B200 rentals on NYMEX on October 5, 2026, pending regulatory approval, with settlement based on the Silicon Data Index; ICE has also announced GPU futures based on the Ornn Index.