Chamath Palihapitiya
· Chamath Palihapitiya
· May 31, 2026 at 15:29
· ⏱ 5 min read
| Read on Substack ↗
Summary
The newsletter recaps three major tech stories: Anthropic's $65B raise at $965B valuation overtaking OpenAI, Pluralis Research's decentralized AI training scaling at 20x/year, and Quantinuum's $1B IPO at $12.7B valuation. These developments highlight the accelerating capital intensity in frontier AI, the emergence of decentralized training as a disruptive force, and the first traditional IPO in quantum computing, signaling maturation of the sector.
•Anthropic raised $65B in Series H at $965B valuation, surpassing OpenAI's $852B close from March.
•Claude Opus 4.8 improved SWE-bench Pro to 69.2% from 64.3% and scored perfect on lazy-thinking benchmark.
•Pluralis-8B was trained across 50 cities using consumer GPUs (Nvidia 4090s) connected over ordinary internet via Protocol Learning.
•EpochAI calculated decentralized training scales at 20x/year vs frontier AI's 5x/year; at current rates, catch-up would take 5.5 years.
•Quantinuum filed IPO at $45-50/share on 21M shares, targeting $1.05B proceeds at $12.7B fully diluted valuation (ticker QNT, trading expected June 3).
•Quantinuum 2025 revenue $30.9M (up from $23M) with $192.6M net loss; Honeywell retains 49.1% voting power post-IPO.
Honeywell is the majority shareholder of Quantinuum and will retain 49.1% voting power post-IPO (article: 'Honeywell has remained the majority shareholder... After the IPO, Honeywell will retain about
Honeywell is the majority shareholder of Quantinuum and will retain 49.1% voting power post-IPO (article: 'Honeywell has remained the majority shareholder... After the IPO, Honeywell will retain about 49.1% of voting power'). The IPO at $12.7B valuation provides a public mark on Honeywell's quantum computing stake, potentially re-rating the parent.
Risk: Quantinuum's $192.6M net loss on $30.9M revenue and $2B+ cumulative R&D spend may weigh on investor perception; Honeywell's consolidated earnings could see dilution from minority stake.
Pluralis Research's decentralized training uses consumer-grade Nvidia 4090s and 6000-series cards (article: 'Most contributions came from consumer-grade Nvidia 4090s and 6000-series cards'). If decent
Pluralis Research's decentralized training uses consumer-grade Nvidia 4090s and 6000-series cards (article: 'Most contributions came from consumer-grade Nvidia 4090s and 6000-series cards'). If decentralized training scales at 20x/year (EpochAI data from article), demand for Nvidia consumer GPUs for AI compute could structurally increase, especially if 'protocol learning' gains adoption.
Risk: Consumer GPUs are lower margin than datacenter GPUs; growth in decentralized training may be slower than projected if coordination or bandwidth challenges persist.
This newsletter, published May 31, 2026,
features Chamath Palihapitiya
discussing HON, NVDA.
2 trade ideas extracted by AI with direction and confidence scoring.