Meta, the parent company of Facebook, has begun exploring a potential entry into the cloud business, saying it will sell part of the artificial intelligence (AI) infrastructure it has invested in to outside parties. While the move is an effort to address criticism that its enormous AI investment in data centers and elsewhere has generated no revenue, market participants instead focused on a different signal: the fact that there are "spare resources in the AI infrastructure." Because analysts had repeatedly noted that supply could not keep up with AI demand, Meta's consideration of a cloud business is delivering a considerable shock to the market. Above all, chip-related stocks, which had risen two- to tenfold in the first half of this year, are taking a direct hit from oversupply concerns. On Wall Street, the mood is not to interpret Meta's business transformation as an immediate signal of slowing chip demand. Still, given that chip-related stock prices had risen too far ahead of fundamentals, investors are maintaining caution that the industry may be approaching its peak.
Meta Considers Selling Spare Computing Resources Externally... The Only One of Four Hyperscalers Without a Cloud Business
Bloomberg reported Monday that Meta has internally launched a "Meta Compute" plan and is envisioning a cloud business model utilizing its own data center infrastructure. According to sources, Meta is pursuing its cloud business in two broad ways. First, it plans to put its latest in-house AI model, "Muse Spark," on its own infrastructure and allow external developers to use it through an application programming interface (API). This is a platform-as-a-service (PaaS) offering similar to AWS's "Bedrock," Microsoft Azure's "AI Foundry," and Google Cloud's "Vertex AI."
The other approach is to lease out pure computing power itself — the raw material of a data center — to outside parties in bulk. This is similar to the infrastructure-as-a-service (IaaS) model used mainly by companies such as CoreWeave and Nebius. In effect, Meta has decided to sell off the spare resources among the computing infrastructure it built to develop artificial super intelligence (ASI). If Meta enters the cloud business, it is expected to shake up a market that had been divided three ways among Amazon Web Services (AWS), Microsoft (MS) Azure, and Google Cloud. Meta is the only one of the four hyperscalers — Amazon, Microsoft, Google, and Meta (companies that operate large-scale AI data centers) — that does not provide cloud services. Buoyed by such expectations, Meta surged 8.81% in the stock market Monday.
Meta's preparation of a cloud business is interpreted as stemming from the fact that its visible revenue model is very limited compared with the massive funds it is pouring into building AI data centers. Meta has recently laid out its ASI development plan and is aggressively building data center infrastructure, buying up AI chips including expensive Nvidia graphics processing units (GPUs).
Meta CEO Mark Zuckerberg answered a question about the possibility of entering the cloud market at the shareholders' meeting in May, saying it was "a perfectly viable option." "Almost every week, outside companies are asking us to provide API services, or asking whether there are computing resources they can buy even at a premium over the price we paid," he said. "When the point comes where we determine that we are in an overbuilt state, that option will be on the table."
Questions Over Existing Business Growth Such as Falling Social Media Users... Investors Anxious Over Massive Capital Expenditure
The fact that Meta has been ramping up AI data center investment while lacking cloud business capabilities has become a clear weakness for its stock. Indeed, Meta's stock fell as much as 8.55% on April 30, the day after it reported first-quarter earnings on April 29. Even though revenue ($56.31 billion) exceeded the market estimate of $55.45 billion and net profit surged 61% from the same period a year earlier, a question mark was attached to its growth potential. That was because first-quarter daily active people (DAP) for its flagship social network service (SNS) businesses — Facebook, Instagram, and WhatsApp — came to 3.56 billion, down more than 5% from the fourth quarter of last year. Revenue for the Reality Labs division, which includes the metaverse business and smart glasses, was $402 million, down 2.4% from $412 million a year earlier. The division's operating loss reached $4.03 billion. This is why investors still view Meta as a social media company rather than an AI outpost.
Zuckerberg went further at the earnings call, foreshadowing large-scale restructuring, saying, "People are using AI to perform their work more efficiently, and we are preparing the company's next phase around such talent." Meta, which laid off 8,000 employees — about 10% of its total workforce — in May, plans to carry out additional restructuring of a similar scale in the second half.
Moreover, investors are anxious about AI infrastructure capital expenditure (CAPEX) that is excessive relative to revenue. Without presenting a clear future revenue plan, Meta raised its capital expenditure forecast for this year from the previous $115 billion–$135 billion to $125 billion–$145 billion. As Google raised its capital expenditure plan for this year from $175 billion–$185 billion to $180 billion–$190 billion, and Microsoft raised its from $185 billion to $190 billion, Meta too competitively increased the scale of its investment. Amazon's annual capital expenditure plan for this year, at $200 billion, is the largest among the hyperscalers. Meta, after applying to issue its largest-ever corporate bonds worth $30 billion (about 45 trillion won) last October, also issued about $25 billion worth of bonds in April this year.
The news of Meta's entry into the cloud business unintentionally raised the question on Wall Street of whether "AI investment is in an excess state." The stray shots hit chip-related stocks, which had until then argued only about excess demand. In the New York stock market Monday, chip-related stocks fell across the board: Nvidia (-1.25%), Broadcom (-2.23%), Micron (-10.57%), AMD (-6.89%), Intel (-9.03%), Applied Materials (-9.97%), Lam Research (-9.71%), and SanDisk (-10.62%). CoreWeave and Nebius, "neocloud" firms that run cloud businesses similar to the service Meta envisions, also fell 13.92% and 17.01%, respectively.
The ruthless profit-taking in chip-related stocks continued Tuesday. Chip-related stocks plunged again for a second straight day: Nvidia (-1.41%), Broadcom (-2.36%), Micron (-5.49%), AMD (-4.26%), ASML (-4.00%), Intel (-5.25%), Applied Materials (-7.35%), Lam Research (-10.19%), ARM (-6.58%), SanDisk (-14.13%), Marvell Technology (-9.84%), Seagate (-10.38%), Western Digital (-9.92%), and Qualcomm (-3.12%). The Philadelphia Semiconductor Index also fell 6.27% Monday and 5.45% Tuesday. Because of the weakness in chip-related stocks, the tech-heavy Nasdaq Composite Index also dropped 0.66% Monday and 0.80% Tuesday. On Tuesday, Meta too sank 4.90%.
"AI Infrastructure to Spare?" Chip Stocks Plunge for Two Straight Days... No Immediate Slowdown, But Bad News for Peak-Price Fears
The AI over-investment controversy triggered by Meta also struck the Korean stock market. On Tuesday, Samsung Electronics (005930.KS) and SK hynix (000660.KS) plunged 9.06% and 14.57%, respectively. As leveraged (borrowed) exchange-traded fund (ETF) volumes tied to the single stocks of Samsung Electronics and SK hynix poured out late in the session, the KOSPI index also plunged 7.89% to close at 7,648.09. That was the lowest level in about a month since December 6 (7,484.41).
Chip-related stocks reacted particularly sensitively to Meta's move, in which it acknowledged that there are spare computing resources in AI infrastructure, because these stocks had until then
