Quick Facts
- DRAM prices jumped 90-95% in Q1 2026, the largest single-quarter increase on record, with memory now accounting for up to 40% of total rack costs
- Hyperscalers plan nearly $700 billion in data center spending for 2026, with AI infrastructure alone consuming $1.37 trillion globally
- NVIDIA has visibility to $500 billion in Blackwell and Rubin revenue through end of 2026, with new Rubin platform delivering 10x reduction in inference costs
The global AI infrastructure race has reached trillion-dollar scale as companies scramble to build what NVIDIA calls “intelligence factories” ahead of the GTC 2026 conference starting Monday.
Memory has emerged as the primary bottleneck. DRAM contract prices surged 90-95% in Q1 2026 alone, marking the largest single-quarter increase on record. By 2026, memory could consume 30% of hyperscaler capital expenditures, with the four major cloud providers forecast to spend a combined $650 billion this year.
“AI infrastructure economics are now defined at the rack and factory level, not at the chip level,” said industry analyst Dave Vellante. Memory now accounts for up to 40% of total rack costs, fundamentally reshaping data center economics.
NVIDIA CEO Jensen Huang will deliver the keynote Monday at SAP Center, with more than 30,000 participants from 190 countries expected to attend. The company has visibility to $500 billion in revenue from its Blackwell and Rubin platforms through end of 2026.
The new Rubin platform, in full production for second-half 2026 availability, delivers up to 10x reduction in inference token costs and 4x reduction in GPUs needed to train mixture-of-experts models compared to Blackwell.
“Intelligence scales with compute,” said OpenAI CEO Sam Altman. “The NVIDIA Rubin platform helps us keep scaling this progress so advanced intelligence benefits everyone.”
McKinsey projects $5.2 trillion to $7.9 trillion in infrastructure investment will be required by 2030 to meet AI compute demands. Supply chain constraints extend well beyond 2026, with TSMC reporting CoWoS packaging oversubscribed through at least 2026 and HBM memory fully allocated through the same period.
The infrastructure war has created what analysts call a “GPU appreciation paradox” where AI hardware increases in value over time as models become more powerful, reversing traditional technology depreciation patterns.
Big Tech companies have lost more than $1.3 trillion in combined market value in early 2026 as Wall Street scrutinizes massive AI spending commitments. Microsoft shares fell nearly 16% year-to-date amid investor concerns about return on AI investments.
“The revolution will be built by those who control the unglamorous physics of power generation, semiconductor fabrication, and cooling systems,” the analysis notes, highlighting how infrastructure constraints now define AI progress more than algorithmic advances.
Read more: GTC preview: Inside the AI factory — The $1T infrastructure war under the hood of the AI economy