Meta shares jumped as much as 10% on Wednesday after Bloomberg reported the company is developing a cloud infrastructure unit, internally called Meta Compute, that would sell access to spare AI computing capacity and to hosted AI models such as its Muse Spark family. The plan, still unconfirmed by Meta, would put the company in direct competition with AWS, Microsoft Azure and Google Cloud — the very firms it currently pays to cover compute it can't build fast enough itself.
Meta expects to spend up to $145 billion on AI infrastructure this year, on top of $70 billion in 2025, and has already committed $21 billion to CoreWeave alone. A pivot toward selling capacity rather than only buying it would test whether that spending is a hedge or a genuine new revenue line. The reaction elsewhere in the "neocloud" market was immediate: CoreWeave fell as much as 14% and Nebius dropped 17% on fears a hyperscaler-scale seller would undercut their pricing. The effort is reportedly led by infrastructure chief Santosh Janardhan alongside Meta Superintelligence Labs' Daniel Gross and Meta president Dina Powell McCormick.
MGX, the Abu Dhabi-based investment firm, said this week it had closed one of the largest funds ever dedicated to artificial intelligence deals, drawing institutional and private investors from the Middle East, North America, Asia and Europe. The raise exceeded its original $45 billion target, and Bloomberg reported the firm has already deployed capital from the fund in recent weeks.
MGX has emerged over the past two years as one of the most active sovereign-linked vehicles in AI, backing ventures spanning frontier labs, data centers and chip supply — part of a broader Gulf push to convert oil wealth into a stake in the AI buildout. The size of the raise underscores how much non-U.S. capital is now chasing the same scarce assets, compute, power and frontier-model equity, that American hyperscalers are racing to lock up.
SoftBank Group has reopened talks with a lending consortium expected to include Goldman Sachs, JPMorgan Chase and Mizuho Financial Group for a $10 billion margin loan secured by its OpenAI stake, Reuters reported, after an earlier attempt stalled over the difficulty of valuing a private company. To win the banks over, SoftBank is now offering a repayment guarantee that gives lenders recourse to the parent company itself if the pledged shares lose value — a concession it did not make the first time, when the ask reportedly shrank from $10 billion to about $6 billion for lack of interest.
SoftBank has committed more than $60 billion to OpenAI and related infrastructure including the Stargate data center venture, and separately faces a March 2027 deadline to repay a $40 billion bridge loan tied to the same bet. OpenAI's confidential June IPO filing could eventually make the stake easier to value and sell, but for now Son is financing one of history's biggest AI wagers the same way SoftBank financed its Arm stake — by borrowing against paper gains before they're realized.
President Lee Jae-myung unveiled a national investment drive this week worth more than 800 trillion won ($518 billion) from Samsung Electronics and SK Hynix, the world's two largest memory chipmakers, alongside suppliers, to build two new fabrication sites apiece in the country's southwest. A further 81 trillion won ($52.5 billion) is earmarked for a chip-packaging cluster near Seoul. "We must secure the core elements of AI faster than any other country," Lee said, framing the buildout as existential rather than merely commercial.
The announcement lands as the global AI chip supply chain strains under a well-documented crunch: TSMC's advanced-node and packaging capacity is sold out through 2027, HBM memory is being rationed toward AI accelerators at the expense of conventional RAM, and server CPU lead times have stretched to six months in some markets. South Korea's bet is that owning more of the memory layer it already leads globally is the surest way to stay indispensable as the AI buildout accelerates.
Physical-security company Verkada announced this week that NVIDIA has taken an equity stake and is collaborating with it to bring more capable AI to the built environment, following an earlier strategic investment from Alphabet's CapitalG. The partnership applies NVIDIA's Cosmos world foundation models and its Physical AI Data Factory synthetic-data toolkit to the models behind Verkada's video search — finding a specific person, object or moment across recordings from more than 2.4 million connected devices in 170 countries.
Since the collaboration began, Verkada says mean average precision for spatial-temporal search queries has climbed 68%, and the company is now building a multi-model search agent aimed at flagging safety incidents on factory floors and shrinkage in retail stores. "Verkada has been building and deploying Physical AI before the term existed," said co-founder and CEO Filip Kaliszan.
European Commissioner Henna Virkkunen held talks with Apple CEO Tim Cook this week after Apple delayed the European rollout of its upgraded, more AI-capable Siri, citing privacy and security complications tied to the Digital Markets Act's interoperability requirements. The Commission counters that the DMA's rules, designed to force dominant gatekeepers to open their platforms to competitors, are not optional, and that Apple must find a way to comply rather than withhold features from EU users.
The standoff is becoming a template case for a wider tension: as AI assistants gain deeper access to a device's data and other apps to be useful, they run directly into the same interoperability and data-sharing obligations regulators imposed to curb Big Tech gatekeeping power. Neither side has signaled a near-term resolution, and how it's settled will likely shape how fast, and on what terms, the next generation of AI assistants reaches European users.