AI money is exploding upward while a deepfake app apparently found a perfectly functional route through Meta's ad system. Quick catch-up before we get into it: Lanarkshire AI Growth Zone was already moving before this BE News follow-up. DataVita had secured £300 million to expand DV1 and build a second facility, backed by a £202 million National Wealth Fund guarantee. Dell Technologies had also chosen the zone's AI Innovation Park for its Scottish base, making Lanarkshire an early test case for AI infrastructure in Scotland. This is AI Daily Briefing. Today: an infrastructure forecast that doubled at warp speed, and Meta's safeguards running into its ad auction. Dell'Oro now puts global data-center capex at $3 trillion by 2030—nearly double its January forecast. That's well beyond a normal forecast update. Joseph Gabriel Lagonsin, writing in DataCenter News:
Dell'Oro Group has forecast that worldwide data centre capital expenditure will surpass $3 trillion by 2030, nearly double its January 2026 outlook. The revised forecast points to sustained spending on AI infrastructure, with high-end accelerators used in AI-optimised servers expected to take the largest share of data centre investment through the end of the decade.
Dell’Oro has pushed worldwide data-centre capex past $3 trillion by 2030—nearly double its own January forecast. In eight months, the whole buildout has been fundamentally repriced. The number is enormous. The speed of the revision is worse. If your 2030 model is off by roughly $1.5 trillion between January and August, nobody should treat that endpoint as a stable business-case input. Dell’Oro says high-end accelerators take the biggest slice, and the top four U.S. hyperscalers could account for about half the spending. So a very small set of balance sheets—and their power contracts—is drawing this decade's infrastructure map. Higher commodity costs also drove part of the revision, so more capacity doesn't automatically mean cheaper inference. More servers can still mean a nastier per-token bill. Simon Creasey, writing in BE News:
The investment will expand DataVita’s existing DV1 data centre and fund the construction of the new DV3 data centre. The capacity of both centres is contracted to AI cloud firm CoreWeave on a 15-year lease. The debt facility, which was unlocked by a £202m financial guarantee from the National Wealth Fund, has been raised with participation from a syndicate of lenders including ING
On the ground, DataVita has a £300 million facility for DV1 and DV3 in Lanarkshire. BE News says both are contracted to CoreWeave for 15 years. Good—an actual tenant and an actual lease. This is well beyond a minister at a podium talking about an AI Growth Zone. And the National Wealth Fund put up a £202 million guarantee against 80% of the main bank tranche. Public backing is helping turn private debt into committed capacity. It also concentrates the bet. CoreWeave underpins the whole borrowing, so £300 million is real financing—but the useful milestone is when the DV1 expansion and DV3 are live enough for customers to run jobs. Here's Ars Technica:
Meta platforms recently ran ads for an AI porn-generation tool that seemingly encouraged users to create deepfaked videos resembling female US politicians, despite the company’s policies against ads containing sexual material. It’s the latest in a series of failures by Meta to keep advertisements for tools that produce nonconsensual intimate imagery off its platforms. The tool, which is called Kromix, bills itself as an “AI image styler.”
Kromix called itself an “AI image styler,” then ran a Meta ad promising “no restrictions” and real people—while steering users toward sexual deepfakes of female politicians. That euphemism got through an ad system with actual money behind it. Meta's ad-review stack cleared a product pitching paid uploads for scenarios labeled “bedroom rape.” This wasn't some moderation miss buried in a billion uploads. What signal did that system see and decide was fine? We just spent two segments on trillions flowing into AI infrastructure. Great—some of that capacity is apparently being monetized through ads for nonconsensual porn. Meta's policy page is irrelevant if the auction keeps cashing the check. Here's one from Hacker News:
This is a potential lawsuit for Meta. Also, the app creator is in violation of US law and may be arrested if found to be a US citizen or in a country with an US extradition agreement.
A lawsuit may come, sure. But legal exposure after the ad runs is a miserable control loop. Meta needs to explain enforcement before an ad is bought, approved, and delivered—not point to a policy after somebody screenshots it. Computer Weekly, with Aaron Tan:
Alibaba Cloud has switched on its third datacentre in South Korea, a year after its second facility went live and four years after it first put infrastructure on the ground in the country. The Chinese hyperscaler announced the launch on 18 August 2026 without disclosing the capacity of the new site, which takes its global network to 104 availability zones across 30 regions.
Alibaba Cloud has opened a third South Korea site—one year after its second—but won't say how much capacity it added. After the numbers we just got from Dell’Oro and DataVita, “third datacentre” isn't a usable unit of measurement. Exactly. They can say 104 availability zones across 30 regions, but that tells a buyer nothing about available GPU capacity—or latency and inference costs under load. And the launch bundle covers agentic development, testing, operations, and security—the whole agent lifecycle. Fine. But the infrastructure announcement carries much of the sales pitch while the operating numbers stay offstage. A cloud region plus an agent platform is a tidy way to lock in both the workload and the serving stack. Gendive and Naver Z are real local use cases; now show enterprises what capacity they can actually buy. When a platform says it has deepfake safeguards, should we take that as proof it can protect people? What would show the rules work when harmful material is still getting views or ad money? Look beyond a policy page: clear labels, a way to report content, evidence the platform removes it, and an outside body that can challenge bad moderation calls. Meta's Oversight Board recently overturned Meta's decision to leave up an Instagram video reportedly generated by AI that impersonated a woman. It also called for stronger protections for non-public figures targeted by sexualized deepfakes. That's accountability in practice: a specific decision reviewed outside the company's usual moderation operation, with recommendations for change. The Board places the case amid new rules taking shape in India, the UK, and Spain. The European Union has also reached an agreement to prohibit AI systems that generate non-consensual sexually explicit or intimate content and child sexual abuse material. On the money side, Sarah Perez at TechCrunch reports that YouTube has clarified that three types of videos categorized as “inauthentic content” can't be monetized. And Jess Weatherbed of The Verge reported that the Oversight Board has called on Meta to expand AI-content labeling, including C2PA, a standard intended to carry information about a piece of content's origins. But labels and takedowns can sound reassuring after the fact. How do we tell whether a platform is catching this at scale instead of merely responding to the cases that become public? Watch the metrics. Are labels broadly applied? Do reports lead to removals? And do monetization rules actually cut off incentives for inauthentic material? Independent reviewers, regulators, and affected users need to be able to identify failures and force the platform to change its decisions. Have feedback, a story idea, or a correction? Email us at aidailybriefing at lantern podcasts dot com. Your notes help make the briefing more useful every day.
Links to every story are in the show notes, so take a look at the pieces that caught your attention and dig a little deeper. That’s AI Daily Briefing for today. This is a Lantern Podcast.