Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028

Dylan Patel – Anthropic & OpenAI will have most of the world’s compute by 2028 – Dwarkesh Podcast

Market views

Bullish

ANTHROPIC— Anthropic has dramatically improved unit economics, going from negative margins to $50M revenue per megawatt, with revenue per megawatt potentially reaching $50-80M by end of 2027. The company is rapidly scaling compute capacity and turning profitable while capturing increasing share of world's compute.
OPENAI— OpenAI has transitioned from venture-funded losses to profitability with dramatically improved margins. The company is rapidly scaling compute capacity (3-4x this year) and generating high revenue per megawatt, with potential to reach $50-80M per megawatt by 2027.
META— Meta is strategically hoarding compute infrastructure using its balance sheet, positioning to either monetize internally or sell to labs at premium prices ($25-50M per megawatt). The company is undervalued given its compute assets and ability to generate optionality from infrastructure investments.
TSLA— SpaceX (Elon Musk entity) is strategically building compute infrastructure and selling it to labs at premium prices ($25-40M per megawatt vs $13M cost basis), demonstrating strong unit economics and optionality in the AI infrastructure market.
NVDA— NVIDIA is positioned to benefit from massive AI infrastructure buildout with pricing power as demand from labs willing to pay premium prices increases. The company has exclusive relationships with key players like SpaceX and is central to the AI compute supply chain.
DATACENTER— Data center infrastructure is experiencing explosive demand with pricing power increasing from $10-15M per megawatt to potentially $25-50M as AI labs compete for capacity. The sector benefits from structural shortage and high returns on investment.
POWER— Power generation infrastructure is a critical bottleneck for AI data centers with massive capital requirements ahead. The sector will see significant investment as power plants must be built years in advance of data center deployment.
HBM— Memory companies (HBM suppliers) are rapidly raising prices as AI infrastructure demand surges. The supply chain is recognizing pricing power with memory companies raising prices very quickly in response to structural demand.

Bearish

TRADFI— Traditional financial institutions and non-AI sectors will face severe pressure from rising interest rates driven by AI infrastructure demand. Banks will suffer as credit spreads blow up and their assets reprice slower than their debt.
EQUITIES— Non-AI equities will face severe valuation compression as rising discount rates from AI-driven interest rate increases crater present values. Traditional stable stocks will be particularly vulnerable as discount rates rise from 3-5% to 8-10%.
EMERGINGMARKETS— Emerging market sovereign debt faces default risk as AI-driven interest rate increases create a second Volcker shock. Countries with high debt, low tax revenue, and frequent debt refinancing (Pakistan, Nigeria) are particularly vulnerable.
RETAIL— Consumer-facing businesses and retail sectors will face severe pressure from rising interest rates as AI infrastructure crowds out traditional lending. Consumer packaged goods and retail companies that rely heavily on debt financing will be particularly vulnerable.
CHINA— China faces severe compute disadvantage due to export controls, with only sub-10% of incremental global AI compute deployment and inferior chip quality. By 2028, China may have only 30 gigawatts versus much larger US deployments, with chips significantly worse than NVIDIA/Google equivalents.

Not investment advice.

Mentioned

Market views