As a PC gamer, you’ll be all too familiar with the sky-high prices of memory and storage components. To put it most simply, these frankly eye-watering price tags are largely due to big tech’s major players buying up as much DRAM and flash chips as possible in order to fuel their AI ambitions. Well, that strategy may have backfired as some are now trying to figure out what to do with an excess of compute capacity.
Meta is moving into the cloud business in order to sell off its glut of AI compute capacity, according to Bloomberg this week. The plan is in its earliest phase, with those close to the matter claiming the company is still considering its approach.
One idea reportedly being considered is to sell access to AI models already hosted on Meta’s existing infrastructure, à la Amazon Web Services’ Bedrock. In other words, developers would pay to access AI models like Muse Spark running from Meta’s own hardware. Apparently, another plan is to skip the model middleman and simply sell Meta’s ‘raw’ compute power.
Zuckerberg said during Meta’s annual shareholder meeting back in May that moving into the cloud business was “definitely on the table.” In response to an investor question, he claimed that there’s already demand from other companies to “stand up an API service” or asking to buy compute from Meta “at some premium.”
However, Zuckerberg then went on to clarify, “We haven’t done that yet, because we think that we have a use for the compute. But obviously, if we get to a point where we feel that we have overbuilt, then that is an option that we have, and that is partially what gives us confidence in investing in building this out.”

Investors are getting increasingly anxious about seeing a return on AI investments. With the company’s full-year capital expenditure projections blowing past $140 billion, it’s not hard to see why. Meta’s shares saw a plunge back in April, so selling off compute would be one way to reassure investors.
It’s worth noting Meta hasn’t committed to any one plan yet, and its strategy may change significantly, but the company’s share value did jump up 9.3% to $615.55 on Wednesday, which Bloomberg describes as “the biggest intraday gain since April”.
Damir Tokic of Seeking Alpha argues that this may not actually be good news for Meta. First, he highlights that Meta’s revenue is highly dependent on advertising across all of its platforms, and that it started building out AI infrastructure to better support this revenue stream.
He elaborates, “Meta is unlikely to continue investing in AI infrastructure with the specific aim to rent it—it likely means that Meta will need to reduce or possibly stop AI investment in datacenters, as it already has excess capacity. Further, Meta would have to increase debt to continue AI capex, and it already borrowed around 20B in 2025 and 2026, with operating cash flows dangerously decreasing.”

It’s also not just Meta potentially going down the route of selling compute capacity, either. SpaceX acquired Elon Musk’s xAI back in February and has since started renting out its own excess compute capacity to Anthropic.
Renting out your extra capacity is all well and good, but it’s a strategy that may become less viable over time if competitors start to also realise they’ve overinvested and the wider industry starts to even more closely resemble a snake eating its own tail. In other words, if the AI bubble hasn’t yet burst, it’s certainly looking strained.