Chamath Palihapitiya
· Chamath Palihapitiya
· March 22, 2026 at 16:34
· ⏱ 5 min read
| Read on Substack ↗
Summary
The newsletter summarizes NVIDIA's GTC 2026 keynote, arguing that the company is in a massive revenue supercycle driven by the shift from training to inference, with Jensen Huang forecasting $1 trillion in cumulative Blackwell+Rubin revenue. However, the data center buildout faces a severe power bottleneck, with only a third of the 241 GW pipeline under active development, potentially slowing hyperscaler CapEx growth. Separately, the SEC's crypto taxonomy and Nasdaq tokenization pilot create a regulatory framework that legitimizes digital assets and could accelerate equity tokenization.
•Jensen Huang claimed computing demand has increased 1 million times in two years and now sees at least $1 trillion in revenue from Blackwell and Rubin (2025-2027), double the prior $500B forecast.
•NVIDIA's Vera Rubin platform delivers 10x inference efficiency and 4x training performance vs. Blackwell, with a unified rack integrating seven specialized chips and a purpose-built Vera CPU.
•The US data center pipeline reached 241 GW at end of 2025 (up 159% YoY), but only one-third is under active development due to grid capacity and generating capacity constraints.
•Hyperscalers are expected to spend ~$650 billion in 2026 (up $200B from 2025), but the key constraint has shifted from capital to energy availability, interconnection rules, and project delivery timelines.
•The SEC (with CFTC) issued a formal taxonomy dividing crypto into five categories — digital commodities, collectibles, tools, stablecoins (all not securities) and digital securities — clarifying the regulatory status of airdrops, staking, and mining.
•The SEC approved Nasdaq's pilot to tokenize traditional securities (Russell 1000 stocks and major ETFs) via the DTC, with tokenized shares fungible with traditional counterparts and carrying identical rights.
The article directly quantifies NVIDIA's product performance leaps (10x inference efficiency, 4x training) and revenue forecast ($1T cumulative), validating the company's dominant position in the AI i
The article directly quantifies NVIDIA's product performance leaps (10x inference efficiency, 4x training) and revenue forecast ($1T cumulative), validating the company's dominant position in the AI infrastructure buildout as it shifts from training to low-latency inference for agentic AI.
Risk: Power bottlenecks and interconnection delays could slow hyperscaler CapEx ramp, potentially dampening near-term demand even if the long-term thesis remains intact.
This newsletter, published March 22, 2026,
features Chamath Palihapitiya
discussing NVDA.
1 trade idea extracted by AI with direction and confidence scoring.