Until now, the dominant narrative was based on the idea that compute capacity is scarce, GPUs are in short supply, and any company that is capable of locking down megawatts, chips and data centers can monetize them at high prices, to the delight of semiconductor companies and "neo-clouds" such as CoreWave and Nebius.

However, Meta's project suggests that part of this shortage may not reflect only explosive end demand, but also a hoarding phase typical of gold rushes. Fearing they will run out of compute, the big tech groups have secured massive capacity beyond their needs. When capital spending becomes too heavy for balance sheets, those capacities ultimately have to be used, rented out, or monetized.

The project reported by Bloomberg would involve marketing part of Meta's surplus capacity via a cloud business dedicated to artificial intelligence. The initiative would not necessarily be limited to simply renting raw power, but could also include access to its AI models via its infrastructure, operating in a way similar to Bedrock at AWS.

Even if Meta obviously will not turn all of its infrastructure into an external cloud, because the group needs a considerable share of its compute for its own operations, the order of magnitude remains impressive. Based on its guidance, Meta would have about 5.5 GW of planned or reserved capacity by mid-2026, versus 1.5 GW for Oracle Cloud, 1.2 GW for CoreWeave, 0.5 GW for SpaceX/Starcloud and 0.4 GW for Nebius. As a result, even 10% of Meta's capacity made available to the market would already represent a compute volume comparable to that of some specialized players.

The shortage may mostly be masking a hoarding phase

Meta's decision matters because if groups this large have accumulated capacity beyond their immediate needs, the apparent shortage could in fact reflect a phase of defensive stockpiling rather than a truly durable supply deficit. The market read this buildout as proof of structurally insatiable demand, when it could instead be the result of classic "gold rush" cycle behavior.

In that scenario, everyone buys shovels, land, and concessions ahead of others, prices rise, capacity is reserved far in advance, then financial pressure gradually forces some players to put part of those assets back on the market. The Meta news is significant because it shows that the compute amassed during the euphoric phase now has to generate a visible return, putting ROI back at center stage.

Meta's decision also echoes SpaceX's, whose move into the neo-cloud space is not only an opportunistic strategic choice, but also an effort to monetize surplus compute capacity, notably tied to the xAI ecosystem. Meta is simply taking that logic to a different scale.

Capital discipline catches up with the AI rush

For Meta, the market may see it as a welcome monetization option. The group remains heavily dependent on advertising, and its massive AI investments now need to result in more visible revenue. The stakes are even higher because Llama, despite its adoption in the open-source ecosystem, remains less directly monetized than the AI offerings integrated into the clouds of Alphabet, Microsoft, or Amazon, while newer closed models still have to prove themselves. By providing access to its models and infrastructure, Meta would therefore be looking to turn a strategic cost center into a real revenue line.

For pure-play compute companies, the signal is far more worrying. The arrival of a player with that kind of capacity could weigh on rents, margins, and valuation multiples, especially if it encourages other hyperscalers to monetize their own excess capacity in turn. In that scenario, neo-clouds would no longer be facing competition only from other compute specialists, but from the very largest hyperscalers.

Overall, the announcement therefore looks more negative for markets, because Meta's entry into the compute market points to excess capacity at some major players and balance-sheet pressure too large to ignore, raising the odds of a pullback in AI investment, and therefore in the revenues of upstream companies such as semiconductors.