Summary of Key Points
Meta has announced the sale of its idle AI computing power (renting out GPUs or hosting AI models), which has led to a divided market reaction: Meta's stock price soared by $127 billion in one day, while the stocks of its computing power suppliers (CoreWeave, Nebius) and storage chip manufacturers (Micron, Samsung, etc.) plummeted; traditional cloud giants (Amazon AWS, Microsoft Azure) saw increases in their stock prices. There was initial concern about an oversupply of computing power, but the ongoing rise in global GPU rental prices and extended delivery times indicate that demand for computing power remains high. Meta's move may be due to difficulties with its own large-scale AI models (Llama4 falling short of expectations, Muse Spark being delayed), which has resulted in idle capacity. However, this also opens up new revenue streams for the company.
1. Why Did AI Hardware Stocks Crash When Meta Started Selling Computing Power?
Meta is a major customer for CoreWeave and Nebius, with total contract values exceeding $60 billion. By selling its own computing power, Meta is essentially both stopping purchasing from these suppliers and competing with them for customers, which caused their stock prices to plummet. Additionally, storage chips (used in AI systems) are crucial for computing power, so there was concern that Meta's sale of computing power would reduce demand for these components, leading to declines in their stock prices as well. In short, Meta's actions directly impacted the suppliers' business models.
2. Is There Really an Oversupply of Computing Power? Don't Be Misled by Surface Symptoms
Despite Meta claiming to have idle capacity, the global demand for computing power still exceeds supply:
- NVIDIA GPU rental prices have skyrocketed: from $2.63 per hour to $5.1 per hour (a nearly doubling increase), with the H100 model increasing by 38%, and the high-end H200 model reaching $3.5 per hour.
- Orders for GPUs are taking 12–15 months to be fulfilled, with new orders scheduled for 2027.
- Amazon's AI computing power services have increased in price twice this year (15% on January 15th and 20% on July 20th).
Therefore, it's not a global oversupply; rather, Meta simply has more capacity than it needs.
3. Why Is Meta Selling Its Computing Power?
There may be issues with Meta's development of large-scale AI models:
- The Llama4 model, launched last year, did not meet expectations, leading Zuckerberg to restructure the AI department and invest heavily in hiring (e.g., bringing over leaders from Scale AI).
- The new Muse Spark model, intended to generate revenue through APIs like ChatGPT, has been delayed and is still not officially available.
Since these models are not generating profits, it makes sense for Meta to sell its idle computing power to turn a profit—a strategic and practical move.
4. Is Selling Computing Power a Sure-Better Deal for Meta? The New Revenue Stream
Meta previously lacked a cloud business, so selling its computing power can convert the substantial capital expenses incurred in building data centers and purchasing GPUs into actual revenue:
- Wells Fargo predicts that by 2028, Meta's AI computing power will grow to 13.2 gigawatts, generating annual revenues of $264 billion at an estimated revenue per gigawatt of $20 billion.
- This would also result in a net profit of $14.6 billion per gigawatt, boosting Meta's earnings per share by 16%.
- SpaceX is already earning $2.17 billion per month from selling computing power through partnerships with Anthropic and Google; Meta is following this approach to turn its idle resources into a profitable asset.
5. Why Is the Market Divided? Meta's Stock Rising, Hardware Stocks Falling, but Cloud Giants Growing?
The market's reactions seem contradictory, but each has its rationale:
- Meta's stock is rising because it has found a new source of revenue from selling idle computing power.
- Hardware stocks are falling due to concerns that Meta may reduce purchases or compete with them for business.
- Traditional cloud giants are growing because they are dominant in the global cloud market (AWS and Azure hold significant shares), and Meta's entry into this space does not pose a threat. Instead, the increased demand for AI computing power benefits these companies.
In essence, the market is responding in a targeted manner: what was expected to decline has indeed declined, while what was expected to rise has risen.
In Conclusion: Meta's sale of computing power is not a sign of an oversupply but rather a result of its own model development challenges. The global demand for AI computing power remains high, and this move allows Meta to address its idle capacity while opening up new revenue opportunities.