{"id":6201,"date":"2026-07-23T21:54:37","date_gmt":"2026-07-24T01:54:37","guid":{"rendered":"https:\/\/redmonk.com\/sogrady\/?p=6201"},"modified":"2026-07-23T21:54:37","modified_gmt":"2026-07-24T01:54:37","slug":"meta-compute","status":"publish","type":"post","link":"https:\/\/redmonk.com\/sogrady\/2026\/07\/23\/meta-compute\/","title":{"rendered":"Meta Compute, Neoclouds and the Future of the AI Infrastructure Market"},"content":{"rendered":"

Earlier this month, Meta CEO Mark Zuckerberg told Bloomberg<\/a> that \u201cit may make sense, in some cases\u201d to consider renting out its existing infrastructure. This confirmed an earlier report, again from Bloomberg, about the new business unit, \u201cMeta Compute.\u201d That report produced an immediate impact in the market; Meta was up about 9%, and the neoclouds that both retailed the type of infrastructure Meta was purporting to sell and counted the company as a major customer, all dropped significantly – CoreWeave (-14%), IREN (-7%), Nebius (-17%).<\/p>\n

There is little debate that Meta making a credible entrance into either the retail or wholesale cloud markets would have significant market impacts on buyers and sellers alike. The important questions, then, are first, what would they need to roll out a cloud infrastructure offering, and second, what might those market impacts be?<\/p>\n

To evaluate those issues, it\u2019s necessary to consider them in order.<\/p>\n

Meta as a Cloud Infrastructure Provider<\/h1>\n

There can be no debate that, as a company, Meta has a deep understanding of infrastructure. Not only have they operated at a massive scale for decades, they have invested in infrastructure heavily, as we\u2019ll come back to, and they have attempted to advance the state of the art in the industry by sharing their internal hardware innovations and encouraging others to do so via the Open Compute Project (OCP) which began as an internal Meta initiative called \u201cProject Freedom.\u201d<\/p>\n

Much like AWS, Google and others before them, Meta was forced by both cost considerations as well as the novelty of their scale to rethink and reinvent its own hardware stack, as traditional off the shelf commercial solutions had not been designed for the load of Facebook\u2019s demand while nevertheless demanding premium prices. Meta and its partners in the OCP have pushed the industry forward, and enabled the creation of next generation hardware providers like the Oxide Computer Company.<\/p>\n

There is, however, a world of difference between understanding how to operate and run infrastructure versus selling it to and running it on behalf of third party customers.<\/p>\n

Meta needs several things to be successful as an infrastructure vendor, among them:<\/p>\n

    \n
  1. The ability to invest at scale<\/li>\n
  2. The ability to operate at a loss while the business ramps up<\/li>\n
  3. The ability to develop and run a business orthogonal to its core model<\/li>\n<\/ol>\n

    The good news for Meta is that the first two boxes should be easy to tick.<\/p>\n

    Investing at Scale<\/h2>\n

    Meta\u2019s deep understanding of infrastructure clearly informs its willingness to invest in same. As has been explored previously<\/a>, Meta spends an absolutely incredible amount of money on its infrastructure.<\/p>\n

    \"\"<\/a><\/p>\n

    Even amongst other cloud infrastructure standouts like Amazon, Google and Microsoft, Meta\u2019s percentage of revenue – not profit – directed to Plants, Property and Equipment (PP&E) spend is eye popping.<\/p>\n

    Investing at scale, in other words, should not be an issue. Meta can and would spend what the company believes necessary to be successful.<\/p>\n

    Operating at a Loss<\/h2>\n

    Whatever Meta\u2019s technical abilities, the reality is that profitability will lag, potentially significantly. Consider the following chart, which depicts the margin for AWS and Google\u2019s cloud business as measured by its income charted as a function of the cloud businesses\u2019 revenue.<\/p>\n

    \"\"<\/a><\/p>\n

    AWS, notably, started splitting out segment revenue in 2013, some seven years after foundational building blocks like EC2 and S3 were released in mid to late 2006. At that point, the business was profitable, but it seems to safe to assume there was a learning curve obscured by the delayed reporting.<\/p>\n

    Google\u2019s path to profitability, meanwhile, is clearer but suggests that the creation of a brand new, enterprise hyperscale class cloud offering is a multi-year process. It is also worth noting that Google\u2019s growth was an often painful exercise from an optics standpoint, with observers constantly questioning the company\u2019s commitment to the new market.<\/p>\n

    Meta Compute, in other words, should expect to lose money for a while and experience negative headlines while doing it.<\/p>\n

    Meta, however, has a demonstrated history of being willing to invest in businesses with no immediate profitability horizon. Meta\u2019s name, in fact, derives from its 2021 rebrand around its expected Metaverse product-line – one that could not render legs for its avatars, and has to date operated within a business unit that has lost almost $84B over the last five years.<\/p>\n

    \"\"<\/a><\/p>\n

    Being willing to lose money, then, should likewise not be an issue.<\/p>\n

    Developing an Orthogonal Business<\/h2>\n

    If the first two requirements seems straightforward for Meta, the third is not. To be fair, much depends on Meta\u2019s ambitions for the business. Selling excess capacity to a few large frontier model companies is a very different model than becoming a retail infrastructure provider – which is why there are those who believe<\/a> Meta will opt for the former. Per Bloomberg, Meta is still debating the various options, ranging from access to multiple hosted models a la AWS Bedrock to a SpaceX-style deal in which the company is selling excess capacity to two customers in Anthropic and Google.<\/p>\n

    If they go the former route at any customer scale, the list of new roles required would be extensive. Meta would need, among other functions, enterprise field sales aimed at CIOs and CTOs rather than CMOs, commercial quality technical support, compliance\/assurance\/trust, DevRel\/docs\/training\/certification, B2B marketing\/AR and enterprise IT legal.<\/p>\n

    There are multiple examples of organizations building just those sorts of capabilities – each of Amazon, Google and Microsoft have accomplished this to some degree – but it\u2019s not an overnight process, and it requires more than just spending. Genuine organizational commitment, sponsorship and oversight are a bare minimum. It can be done, in other words, if there is sufficient organizational will.<\/p>\n

    But this, much more than a willingness to spend or absorb losses, is the real wild card in any Meta compute business. It will be the tell as to how serious Meta is about infrastructure, and one early signal suggests it\u2019s serious indeed. Just last week it came to light<\/a> that Dave Brown, a longtime AWS executive who\u2019d decided to leave the company, was headed to Meta\u2019s Compute team.<\/p>\n

    It will be interesting to see if they poach other cloud talent, and from where.<\/p>\n

    Meta Compute and the Market<\/h1>\n

    If we assume that Meta is serious about becoming a credible infrastructure provider in some form, what does that mean for the landscape moving forward? The market, at least, has made its thoughts clear.<\/p>\n

    \"\"<\/a><\/p>\n

    Meta was up, then gave back some of those gains. The neoclouds got pushed down, rebounded slightly on quantification of an NVIDIA stake in Nebius among other events, but remain significantly depressed. In simple terms, the market believes Meta\u2019s entrance into the infrastructure supplier market would be good for Meta and bad for the neoclouds.<\/p>\n

    Which is an argument that makes sense on paper. When a major customer the size of Meta becomes a potential competitor, that\u2019s usually not a positive development. Meta can leverage economies of scale that smaller suppliers cannot, even as the latter enjoys preferential access to NVIDIA hardware relative to the wider market.<\/p>\n

    But there are three potentially mitigating factors for the neoclouds.<\/p>\n

      \n
    1. Timing<\/strong>: first, as discussed above, any Meta infrastructure business is likely to take time – potentially a substantial amount of time – to get off the ground and compete at scale. That offers a window to adapt, differentiate and compete.<\/li>\n
    2. Model<\/strong>: as mentioned, it\u2019s also unclear which model – retail or wholesale – Meta would pursue. Whichever path they go down might open another for their neocloud competitors. <\/li>\n
    3. Demand<\/strong>: lastly there\u2019s the question of demand. As has been documented previously<\/a>, the competitive window between closed frontier models and open weight alternatives has been narrowing. Most recently the introduction of Kimi K3<\/a> and the pending Qwen<\/a> release have caught the attention of the industry, as they have the capability to legitimately challenge the current frontier models. Add in the rising cost concerns for enterprises with said frontier models, and interest in open weight models is increasing in proportion to their capabilities. Some of these may be run on local hardware, but the most advanced – those most competitive with the capabilities of the frontier models – require just the sort of serious hardware that the neoclouds and potentially Meta can supply. Given how quickly the frontier models have scaled their user bases, then, it\u2019s entirely possible that any subsets that leave those tools in favor of open weight models could significantly expand the market for hosted AI infrastructure overall. Conversely, it’s possible that any easing of the AI hardware shortage could negatively impact demand. <\/li>\n<\/ol>\n

      Overall, it is simply too early to make any solid predictions in the impact of a potential Meta infrastructure business because there are too many unknown variables at play. But it also is worth thinking carefully through the wider market context, because there are potential paths ahead that are decidedly less than obvious – black and white though the industry narratives and market reactions have been to date.<\/p>\n

      In the meantime, watch Meta’s hiring. It will likely be the clearest unambiguous signal of their intent, because to attract senior talent, money typically isn’t enough – they’re already well compensated. What they want in most cases is to work on interesting problems, and be able to leave their mark on the business. To successfully sell the kind of people that can help them build out the infrastructure they need, Meta will need to articulate a clear vision of where it’s going. Even if that’s not shared publicly, then, the types of talent they are able to attract will make it possible to infer how compelling – or not – that vision might be.<\/p>\n

      Disclosure<\/strong>: AWS, Google, Microsoft and Oxide are RedMonk clients. CoreWeave, IREN, Meta, Moonshot (Kimi), Nebius and Qwen are not current clients.<\/p>\n","protected":false},"excerpt":{"rendered":"

      Earlier this month, Meta CEO Mark Zuckerberg told Bloomberg that \u201cit may make sense, in some cases\u201d to consider renting out its existing infrastructure. This confirmed an earlier report, again from Bloomberg, about the new business unit, \u201cMeta Compute.\u201d That report produced an immediate impact in the market; Meta was up about 9%, and the<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"spay_email":"","footnotes":"","jetpack_publicize_message":"","jetpack_is_tweetstorm":false},"categories":[599,19],"tags":[],"class_list":["post-6201","post","type-post","status-publish","format-standard","hentry","category-ai","category-cloud"],"jetpack_featured_media_url":"","jetpack_publicize_connections":[],"jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/posts\/6201","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/comments?post=6201"}],"version-history":[{"count":24,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/posts\/6201\/revisions"}],"predecessor-version":[{"id":6230,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/posts\/6201\/revisions\/6230"}],"wp:attachment":[{"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/media?parent=6201"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/categories?post=6201"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/redmonk.com\/sogrady\/wp-json\/wp\/v2\/tags?post=6201"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}