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
Meta’s AI app has seen a major jolt in downloads since launching its Vibes feed of AI-generated videos, giving investors a glimpse of the company’s artificial intelligence strategy ahead of Wednesday’s third-quarter earnings.
Since releasing Vibes on Sept. 25, the Meta AI app’s downloads on both iOS and Android are up 56% month-to-month to a total of 3.9 million downloads as of Oct. 18, according to data provided to CNBC by mobile research firm Appfigures.
“That’s what I’d call standout growth,” said Appfigures Head of Insights Randy Nelson, adding that “the month’s not even over.”
Still, OpenAI’s rival Sora app, which launched Sept. 30, appears to have more momentum despite only being available on Apple’s iOS platform and requiring an invitation for use. From Sept. 30 through Oct. 18, Sora saw 2.6 million downloads on iOS, compared to the 1.1 million downloads of the Meta AI app, according to Appfigures.
Sora’s early popularity isn’t surprising to nearly a dozen creators, marketers and brand agencies that spoke with CNBC about about their interest and use of the AI tools.
Meta declined to comment.
Creators and marketers say they generally find Sora easier to use. Among the key distinctions is Sora’s capability of producing more realistic looking videos that show people talking, while Vibes only lets users choose select songs as accompanying audio.
The Vibes feed is filled with surreal, playful content, ranging from four lions sharing a pizza in the jungle to a hedgehog singing karaoke.
Meta has been paying creators to produce AI-generated videos for Vibes as a way to boost the visibility of the app, according to people with knowledge of the matter. Meta has sought out influencers who specialize in generative AI tools provided by the startup Midjourney, and is requiring them to sign non-disclosure agreements, according to the people, who asked not to be named because they weren’t authorized to comment on the matter.
Meta’s evolving AI strategy
The Facebook parent debuted the Meta AI app in late April, confirming an earlier CNBC report. The app was originally called Meta View and was used by owners of the Ray-Ban Meta AI glasses to manage their device settings, import photos and perform other utility functions.
The app, which is still used to manage Ray-Ban Meta glasses, was relaunched as Meta AI to serve as the company’s hub for users to interact with the social media firm’s ChatGPT-like AI assistant, but the September launch of Vibes added a video feed component similar to that of TikTok. The key difference is that all of the content found on the Vibes feed has been created entirely by AI generators.
Unlike Sora, which is powered by OpenAI’s proprietary model, Meta AI’s Vibes feature relies on models provided by third parties like Midjourney and Black Forest Labs, according to a Threads post last month from Alexandr Wang, Meta’s chief AI officer and the former CEO of Scale AI. Meta is also developing its own internal generative AI technology, Wang wrote.
Meta’s reliance on third-party AI models to power Vibes underscores the company’s newfound willingness to seek outside help as it tries to get its AI technology back on track. The underwhelming launch of Llama 4 in April spurred CEO Mark Zuckerberg into spending billions of dollars to shake up Meta’s AI organization and install new leaders like Wang, CNBC previously reported.
That overhaul continued last week, when Meta laid off 600 employees in its AI organization. The company spared Wang’s core TBD Labs group, which now overseas touchstone AI efforts like Llama.
Meta’s AI strategy is sure to be a major topic in the company’s third-quarter earnings report and investor call on Wednesday. For the quarter, analysts expect revenue growth of 22% from a year earlier to $49.4 billion, according to LSEG. Wall Street expects revenue growth for the full year of 19% to $196.2 million.
Meta Vibes
Cfoto | Future Publishing | Getty Images
Some of the influencers that Meta is paying to populate Vibes are based in India, where Sora is not yet available and TikTok is banned.
Company executives have previously said that Meta AI usage, largely accessed via the company’s WhatsApp service, is the highest India, the world’s most-populous country, even though not all influencers there get paid.
“One of the biggest reasons that got me excited is because I’ve been in the creator ecosystem for almost 15 years now,” said Gaurav Bisen, a creator from India who posts 10 to 15 times a day on Vibes but isn’t paid by Meta. “Creating content on Instagram, TikTok, YouTube takes a lot of time, energy and investment. But here, it’s so easy. You just type a prompt.”
Bisen said he’s found early traction on Vibes, where his following has grown to more than 25,000. He said he’s learned that short, animated dance clips perform best.
“People who didn’t have the confidence to come in front of a camera can now create from their imagination,” Bisen said. “You can produce creative content without even being in it.”
Still, engagement on Vibes remains limited compared to Instagram or TikTok. Most creators told CNBC that their posts average fewer than 10 likes and almost no comments.
Instead, Vibes users measure success through “remixes” – a feature that lets creators take an AI-generated video, edit the prompt and repost it as their own.
“Remixes kind of seem like the currency on the app,” said Dylan McIntyre, who runs the AI video account JunkBoxAI on Instagram, where he has more than 1.3 million followers but just 22,000 on Vibes. “People are liking what they’re seeing and want to make their own and repost it.”
Meta is leaning heavily into generative AI as it courts creators and competes with OpenAI. Its AI Studio tool lets users build customizable AI characters and chat with them across Messenger, Instagram and WhatsApp.
Investors will also be listening for commentary on the company’s Metaverse strategy.
Meta confirmed on Monday that Vishal Shah, who spent the past four years leading metaverse initiatives, is now helping run AI products as part of the company’s Superintelligence Labs division, which is led by recent high-profile hires, including former GitHub CEO Nat Friedman.
Shah, who will now report to Friedman, was previously a vice president of product at Instagram. The Financial Times first reported on Shah’s new position.
Neurodiverse professionals may see unique benefits from artificial intelligence tools and agents, research suggests. With AI agent creation booming in 2025, people with conditions like ADHD, autism, dyslexia and more report a more level playing field in the workplace thanks to generative AI.
A recent study from the UK’s Department for Business and Trade found that neurodiverse workers were 25% more satisfied with AI assistants and were more likely to recommend the tool than neurotypical respondents.
“Standing up and walking around during a meeting means that I’m not taking notes, but now AI can come in and synthesize the entire meeting into a transcript and pick out the top-level themes,” said Tara DeZao, senior director of product marketing at enterprise low-code platform provider Pega. DeZao, who was diagnosed with ADHD as an adult, has combination-type ADHD, which includes both inattentive symptoms (time management and executive function issues) and hyperactive symptoms (increased movement).
“I’ve white-knuckled my way through the business world,” DeZao said. “But these tools help so much.”
AI tools in the workplace run the gamut and can have hyper-specific use cases, but solutions like note takers, schedule assistants and in-house communication support are common. Generative AI happens to be particularly adept at skills like communication, time management and executive functioning, creating a built-in benefit for neurodiverse workers who’ve previously had to find ways to fit in among a work culture not built with them in mind.
Because of the skills that neurodiverse individuals can bring to the workplace — hyperfocus, creativity, empathy and niche expertise, just to name a few — some research suggests that organizations prioritizing inclusivity in this space generate nearly one-fifth higher revenue.
AI ethics and neurodiverse workers
“Investing in ethical guardrails, like those that protect and aid neurodivergent workers, is not just the right thing to do,” said Kristi Boyd, an AI specialist with the SAS data ethics practice. “It’s a smart way to make good on your organization’s AI investments.”
Boyd referred to an SAS study which found that companies investing the most in AI governance and guardrails were 1.6 times more likely to see at least double ROI on their AI investments. But Boyd highlighted three risks that companies should be aware of when implementing AI tools with neurodiverse and other individuals in mind: competing needs, unconscious bias and inappropriate disclosure.
“Different neurodiverse conditions may have conflicting needs,” Boyd said. For example, while people with dyslexia may benefit from document readers, people with bipolar disorder or other mental health neurodivergences may benefit from AI-supported scheduling to make the most of productive periods. “By acknowledging these tensions upfront, organizations can create layered accommodations or offer choice-based frameworks that balance competing needs while promoting equity and inclusion,” she explained.
Regarding AI’s unconscious biases, algorithms can (and have been) unintentionally taught to associate neurodivergence with danger, disease or negativity, as outlined in Duke University research. And even today, neurodiversity can still be met with workplace discrimination, making it important for companies to provide safe ways to use these tools without having to unwillingly publicize any individual worker diagnosis.
‘Like somebody turned on the light’
As businesses take accountability for the impact of AI tools in the workplace, Boyd says it’s important to remember to include diverse voices at all stages, implement regular audits and establish safe ways for employees to anonymously report issues.
The work to make AI deployment more equitable, including for neurodivergent people, is just getting started. The nonprofit Humane Intelligence, which focuses on deploying AI for social good, released in early October its Bias Bounty Challenge, where participants can identify biases with the goal of building “more inclusive communication platforms — especially for users with cognitive differences, sensory sensitivities or alternative communication styles.”
For example, emotion AI (when AI identifies human emotions) can help people with difficulty identifying emotions make sense of their meeting partners on video conferencing platforms like Zoom. Still, this technology requires careful attention to bias by ensuring AI agents recognize diverse communication patterns fairly and accurately, rather than embedding harmful assumptions.
DeZao said her ADHD diagnosis felt like “somebody turned on the light in a very, very dark room.”
“One of the most difficult pieces of our hyper-connected, fast world is that we’re all expected to multitask. With my form of ADHD, it’s almost impossible to multitask,” she said.
DeZao says one of AI’s most helpful features is its ability to receive instructions and do its work while the human employee can remain focused on the task at hand. “If I’m working on something and then a new request comes in over Slack or Teams, it just completely knocks me off my thought process,” she said. “Being able to take that request and then outsource it real quick and have it worked on while I continue to work [on my original task] has been a godsend.”
Over 95% of international data and voice call traffic travels through nearly a million miles of underwater communication cables.
These cables carry government communications, financial transactions, email, video calls and streaming around the world.
The first commercial telecommunication subsea cable was used for telegraphs and was laid across the English Channel between Dover, England and Calais, France in 1850.
The technology then evolved to coaxial cables that carried telephone conversations, and most recently, fiber optics that ferry data and the internet as we know it.
“About ten years ago, we saw the advent of another big category, which is the webscale players and the likes of Meta, Google, Amazon, etc., who represent now probably 50% of the overall market,” said Paul Gabla, chief sales officer at Alcatel Submarine Networks.
Alcatel is the world’s largest subsea cable manufacturer and installer, according to industry trade magazine Submarine Telecoms Forum.
Demand for subsea cables is increasing as tech giants race to develop computation-intensive artificial intelligence models and connect their growing networks of data centers.
Investment into new subsea cable projects is expected to reach around $13 billion between 2025-2027, almost twice the amount that was invested between 2022 and 2024, according to telecommunications data provider firm TeleGeography.
A map of the world’s undersea communication cables.
CNBC | Jason Reginato
Big Tech, big cables
“AI is increasing the need that we have for subsea infrastructure,” said Alex Aime, vice president of network investments at Meta. “Oftentimes when people think about AI, they think about data centers, they think about compute, they think about data. But the reality is, without the connectivity that connects those data centers, what you have are really expensive warehouses.”
In February, the company announced Project Waterworth, a 50,000km (31,000-mile) cable that will connect five continents, making it the world’s longest subsea cable project.
Meta will be the sole owner of Waterworth, which the company says will be a multi-year, multi-billion-dollar project.
Amazon also recently announced its first wholly-owned subsea cable project called Fastnet.
Fastnet will connect Maryland’s eastern shore to County Cork, Ireland, and capacity will exceed 320 terabits per second, which is equivalent to streaming 12.5 million HD movies simultaneously, according to Amazon.
“Subsea is really essential for AWS and for any connectivity internationally across oceans,” Matt Rehder, Amazon Web Services vice president of core networking, told CNBC in an interview about Amazon’s subsea cable investments. “Without subsea you’d have to rely on satellite connectivity, which can work. But satellite has higher latency, higher costs, and you just can’t get enough capacity or throughput to what our customers and the internet in general needs.”
A ship belonging to Alcatel Submarine Networks deploys a plow to install subsea telecommunications cables.
Alcatel Submarine Networks
Google is another large player, having invested in over 30 subsea cables.
One of the company’s latest projects is Sol, which will connect the U.S., Bermuda, the Azores and Spain.
Microsoft has also invested in the infrastructure.
“You’ve seen this huge growth in submarine cables over the past 20 years. And this is driven by just a voracious demand for data,” says Matthew Mooney, director of global issues at cybersecurity firm Recorded Future.
Cut cables
Disruptions due to cable damage can be quite significant, particularly in areas served by few internet connections.
“If you cut a cable, you can cut off multiple countries from internet access, and that includes financial transactions, banking, e-commerce and basic communications,” said Erin Murphy, a senior fellow at the Center for Strategic and International Studies, a nonprofit national security research organization.
That very thing happened to Tonga, an island nation east of Australia.
In 2022, debris from an underwater volcanic eruption severed the island’s only subsea communication cable, cutting the island off from the rest of the world.
In September, cuts to subsea cables in the Red Sea caused disruptions to Microsoft’s Azure cloud service. The company was able to re-route traffic, but users in Asia and the Middle East still faced increased latency problems and degraded performance.
Experts have said that the majority of subsea cable damage is accidental, usually due to fishing activity or a ship accidentally dropping its anchor on a cable. But lately, these cables are becoming the suspected targets of sabotage.
A subsea cable being manufactured at Alcatel Submarine Networks factory in Calais, France.
CNBC
“When you have so many vessels in international waters that are highly trafficked by lots of commercial vessels or fishing vessels, the likelihood of accidents is fairly high,” Murphy said. “But if you’re a hostile actor, you know that as well. So if you’re sending out the so-called Russian ghost fleet, or if you have a Chinese fishing vessel and a cable is accidentally cut, you could just say, ‘Oh, well, it was an accident.’ But it could be intentional. So it’s really hard to discern sometimes when an act of damage is actually intentional or accidental.”
Mooney and Recorded Future have been tracking some of these cases of suspected sabotage.
“I would say that we have seen a significant uptick in what we would consider intentional damages,” Mooney said. “In 2024 and 2025, [we] saw a notable increase in incidents that occurred in the Baltic Sea and around Taiwan. And so it is difficult to be able to determine with 100% validity that these are intentional. However, the fact patterns that emerge from these events does give you cause to be suspicious that they could all be considered accidental.”
Mooney said the increase in suspected sabotage has corresponded to increased tensions between Russia and Ukraine and China and Taiwan.
Despite there being a lack of concrete evidence of subsea cable sabotage, governments are taking the threat seriously.
In January, NATO launched the “Baltic Sentry” following several incidents of cable cuts in the Baltic Sea. The operation involves deploying drones, aircraft and subsea and surface vessels to safeguard the subsea infrastructure in the region.
“As a result, I don’t believe we’ve seen any instances of cable severing since late January 2025, in the Baltic Sea,” Mooney said.
A picture taken on February 4, 2025 shows a Helicopter 15 (HKP15) (L) on the flight deck of patrol ship HMS Carlskrona (P04) on open water near Karlskrona, Sweden, as part of the NATO Baltic Sea patrol mission, the Baltic Sentry, aimed to secure critical underwater infrastructure. The patrol ship HMS Carlskrona (P04) set off from the naval port in Karlskrona on February 4, 2025 to become part of NATO’s Baltic Sentry operation as one of several Swedish ships that are part of Standing NATO Maritime Group One (SNMG1). This is the first time the ship has hoisted the NATO flag on board. The purpose of NATO’s Baltic Sentry operation is to demonstrate presence and secure critical underwater infrastructure. (Photo by Johan NILSSON / TT NEWS AGENCY / AFP) / Sweden OUT (Photo by JOHAN NILSSON/TT NEWS AGENCY/AFP via Getty Images)
Johan Nilsson | Afp | Getty Images
U.S.-China tension
In the United States, the Federal Communications Commission, which is responsible for granting licenses to anyone wishing to install or operate subsea cables connecting to the U.S., has introduced tighter rules on foreign firms building this infrastructure, citing security concerns.
“One area we’ve been particularly focused on are threats that come from the Chinese Communist Party as well as from Russia,” FCC Chair Brendan Carr told CNBC. “So we’re taking actions right now to make it difficult or effectively prohibiting the ability to connect undersea cables directly from the U.S. to a foreign adversary nation.”
Carr said the FCC is also taking steps to make sure the hardware itself isn’t compromised, not allowing Huawei, ZTE or other questionable “spy gear” to be used in undersea cables.
In July, three House Republicans sent a letter to the CEOs of Meta, Amazon, Google and Microsoft asking if the companies have used PRC-affiliated cable maintenance providers.
In response to CNBC’s question about the letter, Meta’s Aime said, “We do not work with any Chinese providers of cable systems on systems that we’ve announced, and we are in full compliance with U.S. policy regulations around partners in the ecosystem and the supply chain.”
Amazon also told CNBC it does not work with Chinese companies.
Microsoft and Google did not return CNBC’s request for comment on the letter.
To understand how subsea cables work, CNBC visited Alcatel Submarine Networks subsea cable manufacturing facilities in Calais, France and Greenwich, England. We also spoke to government officials and tech giants to find out why subsea cables are crucial to keeping us connected and what we can do to protect this critical infrastructure.
Palantir co-founder and CEO Alex Karp attends meetings at the U.S. Capitol in Washington on Oct. 18, 2023.
Jonathan Ernst | Reuters
With Palantir’s stock plummeting more than 11% this week despite a better-than-expected earnings report, CEO Alex Karp took aim at investors betting against the software company.
Karp, who co-founded Palantir in 2003, went after short sellers in two separate interviews on CNBC this week. After “Big Short” investor Michael Burry revealed bets against Palantir and Nvidia, Karp on Tuesday accused short sellers of “market manipulation.”
He repeated that message on Friday in an interview with CNBC’s Sara Eisen, again knocking Burry’s wager against the stock.
“To get out of his position, he had to screw the whole economy by besmirching the best financials ever … that are helping the average person as investors [and] on the battlefield,” Karp said.
Even with Palantir’s slide this week, the stock is up 135% in 2025 and has multiplied 25-fold in the past three years, an extended rally that’s lifted the company’s market cap to over $420 billion. While revenue and profit are growing rapidly, the multiples have shot up much faster, and the stock now trades for about 220 times forward earnings, a ratio that rivals Tesla’s.
Nvidia and Meta, by contrast, have forward price-to-earnings ratios of about 33 and 22, respectively.
In August, Citron Research’s Andrew Left, a noted short seller, called Palantir “detached from fundamentals and analysis” and said shares should be priced at $40. It closed on Friday at $177.93 after late-day gains pushed the stock into the green.
Palantir, which builds analytics tools for large companies and government agencies, reported earnings and revenue on Monday that topped analysts’ estimates and issued a forecast that was also ahead of Wall Street projections.
But the stock fell about 8% after the report and then slid almost 7% on Thursday. Karp told Eisen that the recent boom in Palantir’s share price isn’t just for Wall Street.
“We’re delivering venture results for retail investors,” he said.
While Palantir has in the past faced a fairly heft dose of short interest, there are currently relatively few investors placing big bets against it. The short interest ratio, or the percentage of outstanding shares being sold short, peaked at over 9% in September and is now at a little over 2%, which is about as low as its been since the company went public in 2020.
Still, calling out the doubters is a common occurrence for Karp, who has previously said on CNBC that people should “exit” if they “don’t like the price.”
In May, after the stock plummeted following earnings, Karp said ,”You don’t have to buy our shares.”
“We’re happy,” he said. “We’re going to partner with the world’s best people and we’re going to dominate. You can be along for the ride or you don’t have to be.”
The company has also faced backlash over its work with government agencies like U.S. Immigration and Customs Enforcement, and Karp has admitted that his strong pro-Israel stance led some people to leave the company.
The boisterous CEO has been particularly vocal this week. On Monday’s earnings call, he questioned how happy the people are who didn’t invest in the company, and told them to “get some popcorn.”
And on CNBC he aimed much of his ire at Burry after the investor revealed his short positions in Palantir and Nvidia.
“The two companies he’s shorting are the ones making all the money, which is super weird,” Karp told CNBC’s “Squawk Box” on Tuesday. “The idea that chips and ontology is what you want to short is bats— crazy.”