Eric

Just published a quarterly update on the hyperscalers. As a (unnecessary) reminder, these are some of the best businesses to ever exist. Combined, the three biggest cloud providers (AWS, Azure, GCP) are at a ~$247B run rate, growing ~23.9% Y/Y. https://t.co/yyT4WPMhCw

1/ Semiconductor ecosystem 1. Nvidia: $115.3B in data center revenue (fiscal year ending Jan 28, 2025). 2. TSMC: ~$13-$14B in AI revenue (fiscal year ending Dec 31, 2024). Quote: “accounting for close to mid-teens percent of total revenue in 2024.” 3. Broadcom: $14.1B in AI revenue (Q1 ending Feb 2, 2025) 4. AMD: >$5B in annual data center AI revenue (fiscal year ending Dec 28, 2024). 5. HBM Market: SK Hynix ~ 50% market share, Samsung ~40% market share, Micron ~10% market share 6. Marvell: >$1.5B in AI revenue (fiscal year ending Feb 1, 2025). Combined, that gives us revenue estimates below:

I just published an updated Current State of AI Markets. I published my last one six months ago, when the topic du jour was “ROI on AI.” At an aggregate level, we’ve seen revenue accrue to the following categories: 1. Semiconductor Ecosystem: ~$160B 2. Data Center Infrastructure: Very roughly ~$115B (assuming half the costs of AI data centers are GPUs; according to Coreweave’s S1, 46% of their purchases went to Nvidia.) 3. Cloud Revenue: ~$25B 4. Foundation Models & AI Applications: <$10B I think the key moment in AI markets is what happens when AI Supply meets AI Demand. That'll tell us what the unhindered demand for AI applications is today. (The hyperscalers are hinting to that point being later in 2025). A breakdown of revenue from AI thus far:

I just published an updated Current State of AI Markets. I published my last one six months ago, when the topic du jour was “ROI on AI.” At an aggregate level, we’ve seen revenue accrue to the following categories: 1. Semiconductor Ecosystem: ~$160B 2. Data Center Infrastructure: Very roughly ~$115B (assuming half the costs of AI data centers are GPUs; according to Coreweave’s S1, 46% of their purchases went to Nvidia.) 3. Cloud Revenue: ~$25B 4. Foundation Models & AI Applications: <$10B I think the key moment in AI markets is what happens when AI Supply meets AI Demand. That'll tell us what the unhindered demand for AI applications is today. (The hyperscalers are hinting to that point being later in 2025). A breakdown of revenue from AI thus far:

If I'm adding up AI semiconductor revenue for the last four quarters (companies doing $1B in revenue at least), I get: 1. Nvidia: $115.3B in data center revenue (fiscal year ending Jan 28, 2025). 2. Broadcom: $14.1B in AI revenue (Q1 ending Feb 2, 2025) 3. TSMC: ~$13-$14B in AI revenue (fiscal year ending Dec 31, 2024). Quote: “accounting for close to mid-teens percent of total revenue in 2024.” 4. AMD: >$5B in annual data center AI revenue (fiscal year ending Dec 28, 2024). 5. Memory providers: SK Hynix, Samsung, Micron (revenue depends on HBM market size and share estimates) 6. Marvell: >$1.5B in AI revenue (fiscal year ending Feb 1, 2025). Anything else I'm missing?

If I'm adding up AI semiconductor revenue for the last four quarters (companies doing $1B in revenue at least), I get: 1. Nvidia: $115.3B in data center revenue (fiscal year ending Jan 28, 2025). 2. Broadcom: $14.1B in AI revenue (Q1 ending Feb 2, 2025) 3. TSMC: ~$13-$14B in AI revenue (fiscal year ending Dec 31, 2024). Quote: “accounting for close to mid-teens percent of total revenue in 2024.” 4. AMD: >$5B in annual data center AI revenue (fiscal year ending Dec 28, 2024). 5. Memory providers: SK Hynix, Samsung, Micron (revenue depends on HBM market size and share estimates) 6. Marvell: >$1.5B in AI revenue (fiscal year ending Feb 1, 2025). Anything else I'm missing?

4/ Option 2: Commoditize your complement There’s another path that companies like Meta are pursuing: commoditize your complement. The fundamental idea of commoditizing your complement is that a value chain acts like a water balloon. If you commoditize one layer of the chain, you squeeze all the value out of it (no profits), and that value flows to the rest of the value chain. Google owned search, expanded to Android and gave it away for free. Nvidia owned the GPU, expanded to CUDA, and gave it away for free. Companies with existing products (Meta, Google, xAI) can pursue this path.

With Anthropic’s new funding round, the foundation model companies are now three of the ten highest-valued unicorns in the world. But over the last few months, DeepSeek was released (showing both cost improvements from distillation and algorithm improvements). And Grok-3 was released, providing top-tier performance, and again making it unclear how to sustainably differentiate at the model level. So there’s a divide between expectations of future economics and the reality of economics today. A thread on the path to sustainability for foundation model companies:

Nvidia earnings call summarized: Numbers were great, Nvidia continues to crush it, but it’s expected now. Most importantly, there weren’t any signals that demand was slowing: “We have fairly good line of sight in amount of capital investments in data centers. We know going forward the last majority of software will be based on machine learning…we have forecasts and plans for our top partners…The startups are still quite vibrant, each one of them needs a fair amount of computing infrastructure.” Went on to call out that the "next wave is coming": agentic AI for enterprise, physical AI for robotics, and sovereign AI. Nvidia's quarterly revenue visualized:

6/ Alternate AI Chip Approaches There’s an ecosystem of chips around Nvidia aiming for a piece of their market: https://t.co/jBj9xVZEHW