Gas turbines are visible at an xAI data center on Riverport Rd in Memphis, TN on April 25, 2025.
Brandon Dill | The Washington Post | Getty Images
The NAACP filed a lawsuit against Elon Musk’s xAI on Tuesday, accusing the artificial intelligence company of violating the Clean Air Act with its use of natural gas-burning turbines to power data centers in and around Memphis, Tennessee.
The suit, filed in the U.S. District Court for the Northern District of Mississippi, alleges that between August and December 2025, xAI and its subsidiary MZX Tech, LLC, installed and operated 27 gas turbines in Southaven, Mississippi, “without an air permit or regard for the health and safety of people living nearby.”
The turbines emit smog-forming pollutants and particulate matter that can lead to increased health risks and an unpleasant odor, among other things.
The NAACP is seeking declaratory and injunctive relief for the companies to “cease operating the Colossus Gas Plant unless and until they obtain the required permits; to apply the necessary pollution controls; and to pay appropriate civil penalties for each day of violation.”
“Our right to clean air is not up for negotiation, especially when companies prove expediency not people is their priority,” Abre’ Conner, NAACP Director of Environmental and Climate Justice, said in an e-mailed statement.
xAI did not immediately respond to a request for comment.
Now owned by SpaceX after its merger this year, xAI has been using the natural gas-burning turbines for months at its facility in Southaven, Mississippi, claiming that no federal permit was required because they were only for temporary use.
The company operates the Colossus 1 and Colossus 2 data centers in Memphis, just across the Mississippi state line. The company is planning to build another named Macrohardrr in Southaven, where it is also building a more permanent power plant that will use 41 natural gas-burning turbines to power the data centers.
Tens of thousands of people live, work and study near xAI’s power plant, according to the suit, and hundreds of thousands more live in greater Memphis. The filing notes that “a much larger share of this population is Black than that of the country’s population as a whole.”
Separately, the NAACP wants Mississippi state regulators to revoke a permit granted to xAI in March that allowed Musk’s company to build out its massive power plant in Southaven with 41 permanent turbines.
Musk, who is the world’s richest person, has been counting on the greater Memphis area to serve as the backbone for xAI’s buildout, as it tries to compete with OpenAI, Anthropic and Google in the booming AI market.
The company sparked regulatory probes by the EU Commission and U.K. online safety regulators, among others, after its Grok image generator and chatbot enabled the widespread creation and sharing of deepfake, AI porn based on images of real people, including children, who did not consent.
Amid that controversy, SpaceX acquired xAI in February in a transaction that values the combined entity at $1.25 trillion, ahead of what’s expected to be a record IPO in the coming months.
The NAACP is being represented in the case by Earthjustice and the Southern Environmental Law Center.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
OpenAI has announced it’s opening its first permanent London office with a capacity of over 500 team members, after the company halted a major AI infrastructure project in the U.K.
The U.S.-based AI firm said it had signed the lease for an 88,500 square foot space on Monday. In February OpenAI announced it would make London its largest research hub outside of the U.S. The company currently employs around 200 people in the U.K. capital.
“The UK has an incredible depth of talent and a strong track record in AI,” Phoebe Thacker, London site lead at OpenAI, said in a statement. “London is already a key hub for our research and teams, and this new office gives us the space to keep building here.”
OpenAI’s new office space will be located in the King’s Cross area of London, home to a slew of AI companies including Google DeepMind, Meta, Synthesia and Wayve.
The news comes just days after OpenAI confirmed it had paused plans for its U.K. Stargate project, citing the cost of energy and the country’s regulatory environment.
Discussions between the company and its partner on the project, Nscale, are still ongoing, according to a source with direct knowledge of the matter. However, the decision to halt plans was considered a blow to the U.K.’s AI buildout.
Industrial energy prices in the U.K. are among the highest in the world. Critics of the U.K.’s AI infrastructure buildout previously told CNBC the high cost of energy and delays in accessing the national grid were key stumbling blocks.
The UK’s AI offensive
The U.K. has been looking to position itself as a leading AI hub globally, but it still lags far behind top ecosystems like the U.S. and China in both technological innovation and funding.
It launched its AI Opportunities Action Plan at the start of 2025 and AI startups have increasingly been raising large sums from VCs.
Funding has hit $6.7 billion so far this year, according to Dealroom, already just shy of the $8.2 billion picked up across the whole of 2025.
AI cloud company Nscale raised $2 billion in March, in Feburary autonomous driving startup Wayve picked up $1.2 billion, and AI voice startup ElevenLabs has raised $500 million.
Following the Pentagon’s spat with OpenAI rival Anthropic, U.K. officials stepped up attempts to court the company, including proposing office expansion in London and a dual listing, the Financial Times reported earlier this month.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
If one thing became clear at the HumanX conference in San Francisco this week, where 6,500 executives, founders and investors gathered to talk about artificial intelligence, it’s that OpenAI no longer dominates the conversation in their industry. For now, at least, that distinction belongs to Anthropic.
Anthropic’s viral coding agent, Claude Code, was the tool on everyone’s lips, even as many attendees acknowledged that OpenAI, Cursor and Google are offering strong alternatives.
Despite its spat with the Pentagon that went public last month and quickly made its way to the courtroom, Anthropic has only gained momentum. The Department of Defense blacklisted Claude, but after opposing rulings in two courts, Anthropic can keep working with other federal agencies while the cases play out.
Anthropic’s early strength in the enterprise has positioned it to benefit from the soaring popularity of AI coding agents, which are used to generate, edit and review code. So while OpenAI kicked off the generative AI boom with the launch of ChatGPT in 2022, Anthropic may be best set up to win contracts from the biggest spenders.
CNBC spoke with 19 executives and investors at HumanX, some of whom asked not to be named in order to speak freely. Here are the top three takeaways.
Claude Code launched to the general public in May 2025, and as of February was generating more than $2.5 billion in annualized revenue. Arvind Jain, CEO of enterprise AI company Glean, said Claude Code has inspired “Claude Mania,” which is putting pressure on business leaders to deploy it.
“It has become a religion, that’s the level of that mania,” Jain said in an interview. “Everybody, if you go and ask them today, ‘Hey, if I gave you one AI tool, what tool would you want?’ The answer would be Claude.”
On Tuesday, Anthropic announced a new AI model, Claude Mythos Preview, with advanced cybersecurity capabilities thanks to its strong coding and reasoning skills. The model sparked a lot of buzz at HumanX, even though its rollout is limited to a select group of roughly 50 companies.
Victor Riparbelli, CEO of AI video company Synthesia, said Anthropic has managed to demonstrate focus and restraint with its models and product, which can be difficult for a young hyper-growth company.
“The guys at Anthropic were just like, ‘We’re not going to do anything about video, we’re not going to care about voice models, we’re just going to solve code gen,’ and now we’re here,” Riparbelli said in an interview. “OpenAI has had the problem of having to market six different products, which just takes up mind space for the consumer.”
One investor cautioned that while Anthropic has been consistent and managed to identify a sticky AI use case, the industry is still young, and momentum could easily swing in another direction.
AI change management
As tech companies work to usher their customers into the AI era, they’re also grappling with how to leverage and deploy agents internally. Even for Silicon Valley startups, keeping up with the pace of change is no easy feat.
Ashwin Sreenivas, president of AI startup Decagon, said the advent of coding agents has led to a number of shifts within his company. Decagon has changed its interview process to allow candidates to use the tools, and the company is able to rely on smaller teams of engineers.
A project that may have required four or five engineers “becomes two engineers because everyone can move a lot faster and go a lot farther,” Sreenivas said in an interview.
For Navrina Singh, CEO of AI governance startup Credo AI, the proliferation of new AI tools has been simultaneously exciting and anxiety inducing for her. Overcommunicating, particularly with her customers, has become essential, she said.
“The things that I could not do last year and I needed to hire 10 people, I can actually build over a weekend and deploy for myself and for the company,” Singh said. “The anxiety is I can’t control my roadmap, and I can’t control my commitments to the enterprise customers who love more clarity and who like a little bit more stability.”
Big tech incumbents are navigating similar changes.
Cisco President Jeetu Patel said roughly 85% of his company’s engineering workforce, or about 18,000 employees, are using AI, but the path to getting there was unlike what he’d anticipated. Patel said Cisco initially learned it i needed to prioritize adoption over outcomes, and to trust that model capabilities will continue to improve.
“You can’t think of these as tools, you have to think of these as digital coworkers that are joining your team, because your composition of your scrum team changes,” Patel said at the conference. “You might not have a scrum team of eight people. You might have a scrum team of two people and six agents, or two people and infinite agents.”
The race against China
Qwen3 is Alibaba’s latest large language model, which it says combines traditional LLM capabilities with “advanced, dynamic reasoning.”
Rather, they’re focused on another looming geopolitical problem: China’s open-weight models.
In AI, a model is considered open weight if its parameters, or the elements that improve its outputs and predictions during training, are publicly available. As of April, Chinese open-weight models, including GLM-5.1, Kimi K2.5 and Qwen3.5, dominate industry benchmarks.
American companies are swarming to China’s models. Cursor built its Composer 2 model using Kimi 2.5. Airbnb CEO Brian Chesky told CNBC in October that his company’s chatbot was largely dependent on Alibaba’s Qwen.
Given the importance the U.S. AI industry is placing on beating China when it comes to innovation, there’s a big emphasis domestically on closing the gap in open weight. Two investors told CNBC they’re dedicating a lot of their time and resources to that effort, and a third said it’s one of the key problems for the industry to solve right now.
Glean’s Jain said having multiple options is critical.
“The trend that we see is that enterprises today, they’re very wary of depending on one or two providers for all of their AI,” Jain said. “They don’t want to work with just one model company, because they know that innovation is happening across many and also in open source. You want to have a choice.”
WATCH: OpenAI slams Anthropic in memo to shareholders
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
The Trump administration is increasingly on a collision course with the European Union over Big Tech fines.
Google, Apple and Meta are contesting fines from the EU over violations of the bloc’s antitrust and competition laws, which total over 6 billion euros, or $7 billion, since the start of 2024.
They’re an increasing bone of contention, as both companies and the White House say the fines reflect the bloc’s hostility to innovation, while the EU tells CNBC that its tough line is getting companies to make decisions that benefit consumers.
Six fines have been imposed since 2024:
March 2024: Apple fined €1.84 billion under antitrust rules for abusing its dominant position in the market for the distribution of music streaming apps.
November 2024: Meta fined €797 million under antitrust rules over practices benefiting Facebook Marketplace.
September 2025: Google fined €2.9 billion under antitrust rules for anti-competitive practices in its advertising technology business.
April 2025: Apple fined €500 million for failing to comply with “anti-steering” obligations. Meta fined €200 million under the Digital Market Actfor requiring users to consent to sharing their data with the company or pay for an ad-free service.
December 2025: X fined €120 million under the Digital Services Act for breaching transparency obligations.
“All companies doing business in the EU are accountable to the European people and should respect the rules meant to protect them,” a Commission spokesperson told CNBC, adding that fines would only relate to the conduct of firms’ operations in Europe that breach EU rules.
Donald Trump’s administration takes a different view.
It’s stepped up its criticism of the bloc, accusing it of over-regulating its tech firms and jeopardising Europe’s ability to benefit from the rise of AI.
U.S. administration interventions
In February, Trump signed a memorandum stating the U.S. would consider tariffs to “combat digital service taxes (DSTs), fines, practices, and policies that foreign governments levy on American companies.”
Fines against U.S. companies are the biggest source of friction on the economic relationship between the EU and the U.S., Under Secretary of State for Economic Growth Jacob Helberg told journalists last week, Reuters reported.
“If the European Union is going to participate in the AI economy…They’re going to need data centers, data and access to the United States AI hardware stack, and you can’t overregulate and move the goal post on regulations and hit companies with huge fines,” U.S. ambassador to the EU Andrew Puzder told Ian King on CNBC’s “Europe Early Edition” on March 27.
When approached for comment on how EU Big Tech fines were impacting U.S.-Europe relations, a U.S. Department of Commerce spokesperson referred CNBC to a November interview with Secretary Howard Lutnick. “Let’s settle the outstanding cases,” he told Bloomberg. “Let’s put them behind us.”
Europe fights back
There’s a difference in opinion on the other side of the Atlantic.
“Fines imposed under EU competition law, the Digital Markets Act and the Digital Services Act serve, first as a penalty for breaking EU laws, and second as a deterrent to ensure that those EU laws are respected, both as a deterrent against re-offending for the company in question and to deter breaches by other market operators,” a Commission spokesperson told CNBC.
Fines are a “last resort” when attempts at an amicable outcome fail, the spokesperson added.
Many changes had been achieved without fines, they said. Apple allowed competitors’ connected devices like smartwatches to work more seamlessly with iPhones after the EU launched formal proceedings in March 2025 under the Digital Markets Act (DMA) without resorting to a fine, the Commission spokesperson added.
When asked to comment, Apple pointed to previous statements, saying that the DMA discourages innovation, weakens privacy protections, delays or degrades product launches and increases security risks. It did not comment on the EU claim that it had changed its processes in response to the DMA proceedings.
Fines
Companies sometimes change their behaviour “only after receiving a fine,” a Commission spokesperson told CNBC.
Meta changed its “pay or consent” offer to users of Facebook and Instagram in 2025 after a DMA non-compliance decision imposed a 200-million-euro fine, they said. The company would begin offering the new service to users at the start of 2026, the Commission said in a December statement.
When asked for comment, Meta directed CNBC to comments from Chief Global Affairs Officer Joel Kaplan.
Kaplan said at the time that the EU’s fine was an attempt to “handicap successful American businesses,” adding that it “effectively imposes a multi-billion-dollar tariff on Meta while requiring us to offer an inferior service.”
Because the 6 billion euros in fines are being contested in court, the EU has not collected all of the money from companies in question, but fines are required by law to be covered by provisional payments or financial guarantees.
There are also several ongoing investigations by the European Commission into U.S. Big Tech companies.
In February, the Commission told Meta it intended to impose “interim measures” to stop it from excluding third-party AI assistants from WhatsApp as part of an ongoing investigation into the company.
The EU also opened formal proceedings in March to investigate whether social media platform Snapchat, owned by Snap, is in compliance with the Digital Services Act over online child safety.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
Amazon CEO Andy Jassy on Thursday released his annual shareholder letter and once again made the case to Wall Street investors that the company’s huge investments in artificial intelligence are worthwhile.
“We’re not going to be conservative in how we play this — we’re investing to be the meaningful leader, and our future business, operating income, and [free cash flow] will be much larger because of it,” Jassy wrote.
The company disclosed in February that it expects to spend roughly $200 billion this year on capital expenditures, with the lion’s share going toward AI infrastructure, including data centers, chips and networking equipment.
That’s more than any of its tech peers, and a nearly 60% increase from last year.
Amazon shares have struggled so far this year as investors question the company’s aggressive AI spending plans and grow increasingly impatient about when the investments will pay off.
Amazon shares closed up 5.6% on Thursday. The stock is up more than 1% year to date.
Jassy has said that Amazon needs the capital to go after “a once-in-a-lifetime opportunity” and to keep pace with “very high demand” for the company’s AI compute.
He reiterated that argument on Thursday and also disclosed for the first time that AI revenue in its cloud computing segment has hit a $15 billion annual run rate.
“We’re not investing approximately $200 billion in capex in 2026 on a hunch,” Jassy wrote.
He noted specifically the over $100 billion commitment from OpenAI, adding that Amazon has received customer commitments for “a substantial portion” of the capex spend and expects to monetize most of it next year and in 2028.
Amazon’s custom chip business, which includes Graviton processors, Trainium AI chips and Nitro architecture, has notched an annual revenue run rate of more than $20 billion, and is “growing triple digit percentages” year over year, Jassy said.
Amazon announced separately Thursday it plans to spend $12 billion on new data centers in central Mississippi, bringing its total investment in the state to $25 billion. It said it plans to cover “all expenses for new energy infrastructure” and any upgrades to local power grids.
Jassy, who became CEO in 2021 when founder Jeff Bezos stepped down, called back to his predecessor’s message to Wall Street decades earlier, when Amazon remained unprofitable for many years.
Bezos argued that long-term growth was more important than short-term profits, testing investors’ patience. In the process, Amazon invested significant sums in cloud computing, warehouses and devices.
Amazon eventually churned out sizable profits and grew to dominate new markets.
Jassy said Amazon is seizing on opportunities that could become big “pillars,” or growth engines, for the company over time. He pointed to the chips business, which is “on fire,” and highlighted growth in its grocery unit, rapid delivery service and nascent Leo satellite internet offering.
“We are willing to make large capex investments and endure short-term FCF headwinds for the substantial medium to long-term FCF surplus,” Jassy said.
Amazon year-to-date stock chart.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
Mark Zuckerberg, chief executive officer of Meta Platforms Inc., wears a pair of Meta Oakley Vanguard AI glasses during the Meta Connect event in Menlo Park, California, US, on Wednesday, Sept. 17, 2025.
David Paul Morris | Bloomberg | Getty Images
Almost 10 months after Meta spent billions of dollars to bring in Scale AI’s Alexandr Wang as the centerpiece of Mark Zuckerberg’s AI overhaul, the company finally revealed its first new model on Wednesday. One big question is — will users pay for it?
While rivals like OpenAI, Anthropic and Google have spearheaded the artificial intelligence boom with powerful models and popular chatbots as well as other services, Meta has been a hefty spender on AI but has yet to show any new revenue streams from it.
In June, Meta shelled out more than $14 billion to hire Wang and some of his top engineers and researchers, soon creating Meta Superintelligence Labs as a new elite unit. And in January, the company told Wall Street it plans to pour between $115 billion and $135 billion this year into capital expenditures, nearly double its 2025 capex figure.
“It’s been a year of basically no releases and a lot of hiring, and then the capex worries for this year are pronounced,” said Morningstar analyst Malik Ahmed Khan, in an interview. “I think Meta had to show investors and operators they have been working on something of substance. That’s the first step.”
Meta’s second step, Khan said, is making the model work and figuring out how to monetize it.
Muse Spark, Meta’s newly released model, is proprietary, a sharp change from its predecessor family of models called Llama, which consisted of open-source offerings, though the company said it does plan to eventually release some open-source versions. Zuckerberg shook up his company’s strategy after the April release of Llama 4, which failed to captivate developers.
Alexandr Wang speaks on CNBC’s “Squawk Box” outside the World Economic Forum in Davos, Switzerland, on Jan. 23, 2025.
Gerry Miller | CNBC
Arun Chandrasekaran, an analyst at Gartner, described the move as a “major shift” and said it “signals an intention to move away” from the Llama brand.
Taking a cue from other frontier AI labs, Meta aims to eventually offer third parties paid API access to Muse Spark after an initial “private API preview” with “select parties.”
But Meta is very late to the game. OpenAI and Anthropic are collectively valued at well over $1 trillion, thanks to the popularity of their models and services, and Google has embedded Gemini across its portfolio of apps and products, while also selling access to the Gemini models via its cloud unit.
Meta’s AI technology, to succeed, has to be good enough to compete with top models while also providing a novel business opportunity.
‘Crown jewel’
Andrew Boone, an analyst at Citizens, said Meta’s clear advantage is the more than 3 billion people who use Facebook, Instagram and WhatsApp every month. And the business opportunity for Meta has nothing to do with trying to attract developers, who currently swarm to OpenAI, Anthropic, Gemini and a host of Chinese models, but rather to focus on its core market: advertising.
“That’s the crown jewel, that’s what needs to continue to improve,” said Boone, who recommends buying the stock.
Khan shares that sentiment.
“I believe that would be the killer use case from Meta’s perspective,” Khan said, with the goal being to “make ads more engaging and improve targeting.”
Advertising accounted for 98% of Meta’s $200 billion in revenue last year. The company has made numerous efforts to diversify its business, most notably spending tens of billions of dollars to try to make the metaverse happen. But Meta’s ad model is the one thing that’s consistently worked, and the company’s investments in AI have served to improve its targeting capabilities and provide better tools for marketers.
Khan said that as advertisers see returns on investment from their Meta spending, they reinvest that money back into more ads on the platform. So it makes sense that they’d be willing to pay for AI services if they can get even better results.
Meta declined to comment about its API plans beyond its initial announcement.
Based on the technical benchmarks Meta released comparing Muse Spark to rivals, the new AI model appears to excel in areas related to image and video processing, said Doris Xin, CEO of AI startup Disarray. Those are important characteristics for advertisers seeking to make dynamic campaigns for an audience that’s grown accustomed to viewing short-form videos on Reels or gawking at cat photos on Facebook and Instagram.
“Compared to like Claude and Gemini, I think it definitely feels like it has more of a consumer bent,” Xin said about Muse Spark.
Zuckerberg, however, has long had ambitions that go well beyond advertising. His approach with Llama was targeted at developers and getting the best and brightest minds in AI using Meta’s tools even if they weren’t paying for them.
With the switch to proprietary models, the pitch to developers becomes more difficult. Joseph Ott, CEO of AI startup Samu Legal Technologies, said he’s unsure about where he would find value.
“The only reason I would use Llama is that I could fine-tune it,” Ott said, referring to the practice of customizing AI models.
Many developers use so-called open-weight AI models, like those provided by Chinese tech companies, as a basis to train AI models to meet their specific use cases. Ott said it’s unclear what would make Meta’s Muse Spark stand out against free or cheaper alternatives and the leading proprietary AI models.
Ulrik Stig Hansen, co-founder of AI and data training startup Encord, said it’s important for Meta to develop its own AI foundation models to avoid any future dependencies on third parties. As one of the few companies with the resources and computing infrastructure necessary to create and maintain big AI models, Meta wants to ensure that it remains relevant in the hottest market on the planet.
“It is about AI sovereignty and being a player in the game,” Hansen said. “They want to be perceived and known as an AI company.”
As for Meta’s massive investment in Wang and his team, Boone said the latest benchmarks suggest that Zuckerberg got what he wanted, and now it’s “back on Mark.”
“We just gave you a state-of-the-art frontier model,” Boone said, referring to the team behind Muse Spark. “What are you going to do with it?”
WATCH: Meta unveils its new AI model: “Muse Spark.”
Correction: Advertising accounted for 98% of Meta’s $200 billion in revenue last year. An earlier version mischaracterized the figure.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
OpenAI’s Stargate project in the U.K. is being paused, with the company pointing towards the cost of energy and the country’s regulatory environment.
The U.S. AI startup announced plans for the major infrastructure project in September, saying it would deploy up to 8,000 GPUs in partnership with Nscale and Nvidia. Politico first reported on Wednesday that the project was on hold.
“We continue to explore Stargate U.K. and will move forward when the right conditions such as regulation and the cost of energy enable long-term infrastructure investment,” an OpenAI spokesperson told CNBC in a statement.
Industrial energy prices in the U.K. are among the highest in the world. Critics of the U.K.’s AI infrastructure buildout previously told CNBC the high cost of energy and delays in accessing the national grid were key stumbling blocks.
Lawmakers in the U.K are also looking to develop new regulations around how AI models use copyrighted work.
Nscale declined to comment when approached by CNBC about the project being paused. Nvidia has been approached for comment.
OpenAI and Nscale are still in discussions about the project in the future, a source with direct knowledge of the matter told CNBC.
Stargate UK
When announced, Stargate UK was seen as a driver of the country’s AI strategy. It followed OpenAI’s signing of a Memorandum of Understanding (MOU) with the U.K. government in July 2025.
The project was expected to be based across a number of sites, including Cobalt Park, which will form part of the newly designated AI Growth Zone in the North East, an OpenAI statement at the time read.
The company was aiming to explore offtake of up to 8,000 GPUs in the first quarter of 2026, with the potential to scale to 31,000 GPUs over time.
That capacity would enable OpenAI’s models to run on local computing power for specialist use cases like critical public services, regulated industries like finance and national security partnerships.
Regulation
The U.K. was set to delay changes to its copyright rules that would’ve made it easier for AI companies to use media content following backlash from the creative sector in the country, the Financial Times reported in March.
Later that month, the government published a report on copyright and AI, which stated that the majority of respondents to its public consultation “rejected the originally preferred proposal in our consultation: a broad exception with opt-out.”
“Many responses were from the creative industries, who were concerned a broad exception would allow generative AI to learn from their works, without compensation, and in direct competition to them,” the report reads.
“We see huge potential for the U.K.’s AI future,” the OpenAI spokesperson added. “London is home to our largest international research hub, and we support the Government’s ambition to be an AI leader.”
“In the meantime, we are investing in talent and expanding our local presence, while also delivering on the commitments under our MOU with the Government to adopt frontier AI in UK public services,” the statement continued.
— CNBC’s Arjun Kharpal also contributed to this report.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
Meta is debuting its first major artificial intelligence model since the costly hiring of Scale AI’s Alexandr Wang nine months ago, as the Facebook parent aims to carve out a niche in a market that’s being dominated by OpenAI, Anthropic and Google.
Dubbed Muse Spark and originally codenamed Avocado, the AI model announced Wednesday is the first from the company’s new Muse series developed by Meta Superintelligence Labs, the AI unit that Wang oversees. Wang joined Meta in June as part of the company’s $14.3 billion investment in Scale AI, where he was CEO.
Meta is desperate to regain momentum in the fiercely competitive AI market following the disappointing debut of its latest open-source models last April. The release failed to captivate developers, leading CEO Mark Zuckerberg to pivot his strategy.
“Over the last nine months, Meta Superintelligence Labs rebuilt our AI stack from the ground up, moving faster than any development cycle we have run before,” Meta said in a blog post on Wednesday. “This initial model is small and fast by design, yet capable enough to reason through complex questions in science, math, and health. It is a powerful foundation, and the next generation is already in development.”
Meta isn’t positioning Muse Spark as a top-of-the-line model, but is instead highlighting its efficiency and “competitive performance” on various tasks.
The new Muse Spark will be proprietary, instead of open source, with the company saying there is “hope to open-source future versions of the model.” The company had been taking an open-source approach to AI with its Llama family of models.
Meta said in a technical blog about the new model that that improved AI training techniques along with rebuilt technology infrastructure has enabled the company to create smaller AI models that are as capable as its older midsize Llama 4 variant for “an order of magnitude less compute.”
“Muse Spark offers competitive performance in multimodal perception, reasoning, health, and agentic tasks,” Meta said in the post. “We continue to invest in areas with current performance gaps, specifically long-horizon agentic systems and coding workflows.”
Meta is also experimenting with a new AI model revenue stream by offering third-party developers access to Muse Spark’s underlying technology via an API. Currently, only unspecified “select partners” can access the AI model’s “private API preview,” but Meta said that it plans to eventually offer paid API access to a wider audience at a later date.
The new model now powers the company’s digital assistant in the standalone Meta AI app and desktop website. Muse Spark will debut in the coming weeks inside Facebook, Instagram, WhatsApp and Messenger, as well as in the company’s Ray-Ban Meta AI glasses. Meta also plans for Muse Spark to eventually power the company’s Vibes AI video feature in the Meta AI app. That service currently uses AI models from third-parties like Black Forest Labs.
With Muse Spark, users of the standalone Meta AI app and related website will now be able to alternate between certain modes depending on the sophistication of their prompts. Users can get use one mode quick answers to simple questions, and another for more complicated queries related to tasks like analyzing legal documents or gleaning nutritional information from photos of grocery store products.
With Muse Spark, users of the standalone Meta AI app and related website will now be able to alternate between certain modes depending on the sophistication of their prompts. With Instant mode, users can get quick answers to simple questions whereas Thinking mode lets them input more complicated queries related to tasks like analyzing legal documents or gleaning nutritional information from photos of grocery store products.
Additionally, a Contemplating mode “will be rolling out gradually” in the Meta AI app and site for the most complicated queries and tasks, Meta said in the technical blog. In this mode, the Muse Spark model utilizes a squad of AI agents to help “reason in parallel,” thus helping it “compete with the extreme reasoning modes of frontier models such as Gemini Deep Think and GPT Pro,” the technical blog said.
The revamped Meta AI with Muse Spark will also contain a Shopping mode that the company said will be able to help people buy clothes or decorate rooms.
“Shopping mode draws from the styling inspiration and brand storytelling already happening across our apps, surfacing ideas from the creators and communities people already follow,” Meta’s blog post said.
This is breaking news. Please check back for updates.
WATCH: Alphabet, Meta, Microsoft all down as data center spending rises.
Gen Z girl looking at smartphone screen feeling upset scrolling on social media.
Mementojpeg | Moment | Getty Images
Governments around the world are making efforts to crack down on teen social media use amid mounting evidence of potential harms, but critics argue blanket bans are an ineffective quick fix.
Australia became the first country to enforce a sweeping social media ban for under-16s in December, requiring platforms like Meta’s Instagram, ByteDance’s TikTok, Alphabet’s YouTube, Elon Musk’s X, and Reddit to implement age verification measures or face penalties.
Several European countries are now looking to follow Australia’s lead, with the U.K., Spain, France, and Austria drafting their own proposals. Although a national ban in the U.S. looks unlikely, state-level legislation is underway.
It comes after Meta, the parent company of Facebook, Instagram and Threads, faced two separate defeats in trials related to child safety and social media harms in March.
A Santa Fe jury found Meta misled users about child safety on its apps. The next day, a Los Angeles jury ruled that Meta and YouTube designed platform features that contributed to a plaintiff’s mental health harms.
Meta’s stock drops almost 8% as 2 court defeats add to Zuckerberg’s recent woes
These developments are set to “unleash a lot more legislation,” Sonia Livingstone, social psychology professor and director of the London School of Economics’ Digital Futures for Children center, told CNBC.
However, Livingstone said a social media ban for teens is a slapdash solution from governments that have failed to properly police tech giants for years.
“I think the argument for a ban is an admission of failure that we cannot regulate companies, so we can only restrict children,” she said, explaining that the U.S. and Europe already have a lot of legislation in the books that isn’t being enforced.
“When are governments really going to enforce, raise the stakes on fines, ban the companies if necessary for not complying,” she added.
Enforce existing laws
Experts argue the sector has for too long escaped accountability and the rigid requirements faced by other industries.
“[Governments] should be implementing the law [and] big tech companies should be facing a slew of regulatory interventions that forbid a whole series of practices that they currently do,” Livingstone said.
She highlighted the U.K.’s Online Safety Act, which “requires safety by design” — this means features such as Snapchat’s “Quick Add” that invite teens to befriend others should be stopped, according to Livingstone.
Livingstone believes that a blanket ban wouldn’t even be under discussion if social media companies had undergone appropriate premarket testing to establish if their features are safe for their target audience.
“There are lots of areas where we have a well functioning market that requires testing to establish it meets the standards…[before products] can go into the market,” she said. “If we did that for AI and for social media, we would be in a whole different place and we’d not be having to talk about banning children from anything.”
Josh Golin, executive director at Boston-based non-profit Fairplay for Kids, told CNBC that he’d like to see “privacy and safety by design legislation rather than blanket bans” across the U.S.
This includes passing the Children and Teen Online Privacy Protection Act to put a stop to personal data-driven advertising towards children, so there’s “less financial incentive for social media companies to target and addict kids.”
Golin added that passing the Senate’s version of the Kids Online Safety Act (KOSA) is also key to ensuring platforms are held legally responsible for design features that can cause addiction or other harms.
He added that Meta has already successfully lobbied to stop KOSA even though it passed the Senate in 2024. But, if it continues to block legislation further, Golin thinks this could see further pressure “line up behind bans because addictive and unsafe is not OK.”
A ban is ‘lazy’ and ‘unfair’
A sweeping social media ban only punishes a generation of young people who have become increasingly dependent on online means of interaction, according to Livingstone. She said bans are a “lazy” solution from governments and an “unfair” outcome for young people.
“It’s the 15 years in which we don’t let our children go outside and meet their friends. It’s the 15 years in which we stopped funding parks and youth clubs for them to meet in,” she said.
“So a ban now is to say to ‘Children, we can’t make the regulation work. We can’t update it fast enough. We haven’t built you anything else to do, but that’s just tough. We’ve terrified your parents into feeling that there’s nothing they can do, and we’re going to take you away from the service where you hoped you would feel some sociability and entertainment.”
Dr. Victoria Nash, associate professor and senior policy fellow at the Oxford Internet Institute, described social media bans as an “extreme” measure that alienates young people from the benefits these platforms provide.
“We know children and young people get their news online and through apps, so you cut that off,” she said. “My view would be that I don’t think this justifies a ban. To me, what this justifies is more responsible behavior by social platforms to cut down on their most harmful features.”
She said that bans could drive young people and children to less regulated corners of the internet, which don’t have the same protections.
Many Australian teens flouted the social media ban when it first came into force in December. A BBC report found that downloads of VPNs, which hide users’ locations to avoid country-specific restrictions, increased before the ban.
Additionally, downloads of some apps that weren’t yet affected such as Lemon8, Yope and Discord also surged in the days after the law came into effect, per the report.
“I think it [a ban] certainly gets rid of all the harmful aspects, but it gets rid of the good ones too and I’m just not yet sure if that’s proportionate,” Nash added.
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
Broadcom CEO Hock Tan speaks at the digital X event in Cologne, Germany, on September 13, 2022.
Ying Tang | Nurphoto | Getty Images
Broadcom said Monday that it’s agreed to produce future versions of artificial intelligence chips for Google, and signed an expanded deal with Anthropic that will give the AI startup access to about 3.5 gigawatts worth of computing capacity drawing on Google’s AI processors.
Shares of Broadcom rose 3% in extended trading.
The disclosure in a securities filing underscores the surging demand for infrastructure that can run generative AI models. Anthropic’s popularity has soared this year, with its Claude app becoming the top free U.S. app listed in Apple’s App Store in February after a dispute between the company and the Pentagon became public.
On an earnings call last month, Broadcom CEO Hock Tan said that “for Anthropic, we are off to a very good start in 2026” in providing 1 gigawatt of compute from Google’s homegrown tensor processing units (TPUs). Broadcom helps Google make its TPUs.
“For 2027, this demand is expected to surge in excess of 3 gigawatts of compute,” he said.
In a note following the earnings call, analysts at Mizuho led by Vijay Rakesh estimated that Broadcom would pick up $21 billion in AI revenue from Anthropic in 2026 and $42 billion in 2027. The filing on Monday did not contain a dollar amount.
Meanwhile, Broadcom is also collaborating with Anthropic rival OpenAI on custom silicon for AI. Both model builders currently rely heavily on graphics processing units from Nvidia through cloud providers such as Amazon, Google and Microsoft. OpenAI has also committed to drawing on six gigawatts of AMD’s GPUs, with the first gigawatt set to come in the second half of this year.
WATCH: Final Trades: Broadcom, Spotify, Applovin and Uber
Choose CNBC as your preferred source on Google and never miss a moment from the most trusted name in business news.
