Meta Platforms

Meta Eyes Electricity Trading to Power Data Center Expansion

November 22, 2025 - Meta Platforms Inc. is exploring entry into the electricity trading market as it ramps up efforts to accelerate the development of power infrastructure for its data centers.

Both Meta and Microsoft are reportedly seeking federal approval to trade electricity—approval that Apple has already secured. By entering the market, Meta aims to commit to long-term energy purchases from newly built power plants while retaining the flexibility to resell surplus power through wholesale markets.

Urvi Parekh, Meta’s head of global energy, emphasized the importance of corporate participation in energy development. “Power plant developers want to know that the consumers of power are willing to put skin in the game,” she said.

“Without Meta taking a more active voice in the need to expand the amount of power that’s on the system, it’s not happening as quickly as we would like,” Parekh added.

Highlighting the scale of energy demand, it’s estimated that at least three new gas-powered facilities will be required to support Meta’s planned data center campus in Louisiana—a reflection of the growing energy needs driven by the tech industry’s AI ambitions.

Source: https://techcrunch.com/2025/11/22/meta-wants-to-get-into-the-electricity-trading-business/

Meta rolls out a new tool to safeguard Reels creators from content theft

November 17, 2025 - Creators on Facebook are now getting access to a new feature designed to protect their original Reels from being reused without permission. On Monday, Meta Platforms introduced Facebook Content Protection, a mobile‑based tool built to detect when a creator’s original Reels posted on Facebook are being used without authorization.

If the creator receives an alert indicating that someone else is using their Reel, they gain the ability to either block its visibility across both Facebook and Instagram, or monitor the Reel’s performance and optionally add attribution links to their work.
Alternatively, they may choose to waive their claim on the Reel, allowing it to remain visible on the Meta platforms.

Meta says the new content protection tool is part of its broader effort to empower original creators on Facebook and shield them from being overshadowed by copycats. As part of this initiative, Meta disclosed that in July it had removed approximately 10 million profiles impersonating major creators and taken action against some 500,000 accounts involved in spammy behaviour or fake engagement.

Although the system also extends to protect original content posted on Instagram, the caveat is that creators must post their Reels to Facebook for tracking to work — including via the “Share to Facebook” cross‑posting option from Instagram. This condition might encourage greater sharing of content on Facebook.

The content‑protection feature is being automatically rolled out to Facebook creators enrolled in the Facebook Content Monetization Program who meet Meta’s enhanced integrity and originality standards. The company also states that creators using its Rights Manager tool will gain access.

Creators can check their eligibility by looking for notifications in their Feed, Professional Dashboard or profile — or by visiting their Professional Dashboard and selecting “Content Protection.” They may also apply directly for access on Facebook’s website.

Behind the scenes, the tool uses matching technology identical to Meta’s existing Rights Manager system for copyright holders. It will display key metrics such as the percentage match for each detected copy, along with details like views, follower count and monetization status.

Meta emphasises that creators will retain control over how the system flags matches — for example, if a creator has granted another account permission to use their content, they can add that account to an “allow list” so the duplicate Reels aren’t automatically flagged.

Creators also have choices when dealing with replicas: they can release their claim on a video one time, or they can choose to track the performance of a Reel on another account and add attribution links. These links will attach an “original” label to the Reel, linking back to the creator’s profile, Page or in some cases, the original Reel itself. Meta indicates it is testing linking back to the original Reel, but by default will link to the creator’s profile or Page.

If a creator decides to block the unfairly‑used Reel, its distribution will be impacted — however, Meta clarifies that the offending account won’t face disciplinary action. Meta suggests this choice is intended to avoid misuse of the system against specific accounts. Moreover, creators who abuse the tool by filing false reports may face restrictions on their own accounts or lose access to the feature.

By default, tracking of Reels is the tool’s setting. Facebook also allows creators to contest cases where another account tries to claim or protect a piece of their original work: creators can submit a copyright takedown request via the IP reporting channel. They can also file a report if they notice a match that the tool didn’t surface, through a “Can’t find a specific match?” option in the Content Protection overview screen.

For now the new feature is available only on mobile devices, but Meta tells TechCrunch it is exploring a desktop option through the Professional Dashboard.

Source: https://techcrunch.com/2025/11/17/meta-releases-a-new-tool-to-protect-reels-creators-from-having-their-work-stolen/

Meta Unveils GEM: A Generative AI Breakthrough for Ads Recommendation

November 10, 2025 - Meta has introduced its most advanced ads recommendation foundation model to date, the Generative Ads Recommendation Model (GEM), designed to enhance ad performance and improve advertiser ROI. GEM significantly strengthens Meta’s ads ecosystem by improving the accuracy and efficiency of its recommendation models across platforms.

Built on a large language model (LLM)-scale architecture and trained across thousands of GPUs, GEM introduces a new paradigm for ads recommendation systems. Its innovative architecture supports scalable training and efficient prediction, allowing for more relevant and personalized ad delivery.

Already deployed across Instagram and Facebook, GEM has driven notable results—boosting ad conversions by 5% on Instagram and 3% on the Facebook Feed during Q2. In Q3, enhancements to GEM’s architecture doubled the performance gains per unit of data and compute, enabling Meta to further scale the model cost-effectively.

GEM delivers its performance gains through three core innovations: advanced model scaling, post-training knowledge transfer, and robust infrastructure for large-scale training. These capabilities allow GEM to optimize outcomes across the full funnel—awareness, engagement, and conversion—by jointly addressing user and advertiser objectives.

To address the unique challenges of building a large-scale ads foundation model, GEM was designed to:

Navigate a massive and dynamic feature space with billions of daily user-ad interactions and sparse conversion signals.

Process heterogeneous data types including advertiser goals, creative formats, and cross-channel user behaviors.

Efficiently train on massive compute resources using advanced parallelism and system-level optimization.

The result is a model that is four times more efficient at driving ad performance per unit of data and compute than Meta’s previous ranking models. Additionally, GEM’s knowledge transfer techniques are twice as effective as traditional methods, powered by a training stack that achieves a 23x increase in effective FLOPs and a 1.43x boost in model FLOPS utilization.

GEM processes both sequence features (e.g., activity history) and non-sequence features (e.g., demographics, ad formats) using customized attention mechanisms and a high-efficiency design. Its architecture includes enhancements to Meta’s Wukong model, enabling deep and wide feature interaction modeling, and introduces a pyramid-parallel structure for offline sequence learning, allowing it to model long-term user behaviors at scale.

To preserve and leverage user journey data without compression losses, GEM incorporates InterFormer—a transformer-based structure that enables cross-feature learning while maintaining full sequence data integrity.

GEM also excels in multi-domain learning, balancing cross-platform insights with domain-specific optimization. It tailors predictions across Facebook, Instagram, and Business Messaging by learning shared patterns while maintaining sensitivity to surface-level differences.

Crucially, GEM’s effectiveness relies on its ability to transfer knowledge to hundreds of vertical models (VMs) that serve specific ad surfaces. Using a combination of direct and hierarchical transfer methods, including student adapters and parameter sharing, GEM ensures that performance gains are realized across the entire ads stack.

To train GEM at scale, Meta implemented multi-dimensional parallelism, custom GPU kernels, and co-designed model-system strategies that allow nearly linear scaling across thousands of GPUs. Its distributed training strategy separates dense and sparse components, using advanced memory and communication optimization techniques for each.

System-level innovations—such as Meta’s custom GPU kernels, PyTorch 2.0 graph compilation, memory compression, and NCCLX-based GPU collectives—maximize throughput and efficiency during training. Meta also reduced training overhead by optimizing startup processes, cutting PyTorch 2.0 compile time by 7x and overall job start time by 5x.

GEM’s training lifecycle has been optimized end-to-end. Lightweight model variants accelerate early experimentation, while post-training tasks—such as continuous online learning and knowledge generation—maintain the model’s relevance and adaptability.

Looking forward, GEM will expand to incorporate multi-modal inputs including text, images, audio, and video, learning from both organic and ad content across Meta’s ecosystem. This positions GEM to evolve into a unified ranking engine for both ads and organic content, further enhancing personalization and advertiser performance.

Meta plans to continue scaling GEM by leveraging advanced AI hardware and evolving its training methods, ultimately enabling real-time, intent-driven ad experiences and automated, insight-based advertiser tools.

Source: https://engineering.fb.com/2025/11/10/ml-applications/metas-generative-ads-model-gem-the-central-brain-accelerating-ads-recommendation-ai-innovation/

Meta Launches AI-Powered Short-Form Video Feed 'Vibes' in Europe

November 6, 2025 - Meta has expanded its AI-generated short-form video platform, Vibes, to Europe via the Meta AI app, the company announced on Thursday. The feature, comparable to TikTok or Instagram Reels, features only AI-created content.

This European launch follows a U.S. rollout six weeks ago. Shortly after Vibes debuted, OpenAI launched a similar platform named Sora for creating and sharing AI-generated videos.

Vibes enables users to generate and share short videos using prompts or remix existing ones. Users can customize visuals, add music, and modify styles to reflect personal preferences. Over time, the feed becomes more tailored to individual interests.

Meta described the experience as "inherently social and collaborative," encouraging users to co-create, remix, and share stories. Videos can be posted to the Vibes feed, sent to friends, or cross-posted to Instagram and Facebook Stories and Reels.

Despite the platform's collaborative focus, public response to CEO Mark Zuckerberg's September announcement was mixed. User comments on the launch post included: “gang nobody wants this,” "Bro’s posting ai slop on his own app," and “I think I speak for everyone when I say: What….?”

The rollout comes amid rising concerns over the proliferation of low-quality AI content, often dubbed "AI slop." While Meta embraces the trend, platforms like YouTube are moving to limit such content. Earlier this year, Meta had also advised creators to focus on "authentic storytelling" over unoriginal, low-value videos.

Still, Meta reported a tenfold increase in media generation within the Meta AI app since Vibes was introduced.

Source: https://techcrunch.com/2025/11/06/meta-brings-its-short-form-video-feed-of-ai-slop-to-europe/

Meta Launches $30 Billion Bond Sale to Fund AI Infrastructure Expansion

October 31, 2025 - Meta Platforms has announced its largest-ever bond offering, targeting $30 billion to support the expansion of its artificial intelligence (AI) infrastructure. The six-part bond sale includes maturities ranging from 5 to 40 years, with each tranche expected to raise between $4 billion and $6.5 billion.

The fundraising effort follows a $27 billion financing deal with Blue Owl Capital for Meta's Hyperion data center in Louisiana and reflects the company’s growing focus on AI development.

Meta also raised its capital expenditure outlook for the year to a range of $70 billion to $72 billion, up from the previous estimate of $66 billion to $72 billion.

Despite reporting 26% revenue growth, Meta’s stock declined by 11% due to a 32% spike in costs, underscoring investor concerns about rising expenses.

The move aligns with a broader trend across the tech industry, with major firms expected to spend around $400 billion on AI infrastructure in 2025. Looking ahead, Meta CFO Susan Li noted that compensation for AI talent is projected to be the company’s second-largest cost next year, highlighting the competitive landscape for skilled professionals.

As Meta doubles down on AI investment, the long-term payoff remains to be seen amid increasing scrutiny from investors.

Source: https://scanx.trade/stock-market-news/global/meta-platforms-launches-massive-30-billion-bond-sale-to-fund-ai-infrastructure/23424032

Meta Pushes AI Adoption Among Metaverse Staff to Drive 5X Productivity

October 13, 2025 - Meta is urging employees across its metaverse division to integrate artificial intelligence tools into their daily workflows with the goal of boosting productivity by a factor of five, according to an internal communication reviewed by 404 Media.

The directive was issued by Vishal Shah, Meta’s vice president of metaverse, who emphasized a bold shift in mindset. “Think 5X, not 5%,” Shah told employees, encouraging them to use AI for tasks such as prototyping, bug fixing, and accelerating development cycles. “Our goal is simple yet audacious: make AI a habit, not a novelty,” he wrote in the internal memo.

Shah outlined a clear target—by the end of 2025, Meta expects 80% of employees working on metaverse initiatives to have AI embedded into their daily routines. To support this transition, the company plans to roll out training sessions and organize “days of AI learning.”

“A 5X leap in productivity isn’t about small incremental improvements,” Shah added. “It’s about fundamentally rethinking how we work. The more we push ourselves, the more we’ll unlock.”

The push for AI-driven efficiency comes as Meta continues to pour billions into its metaverse projects with limited user adoption to show for it. CEO Mark Zuckerberg has consistently framed AI as critical to Meta’s future, predicting that artificial intelligence will write most of the company’s code within the next 12 to 18 months. Reflecting this stance, Meta recently announced that job applicants are permitted to use AI tools during coding interviews.

This internal campaign mirrors broader industry shifts. Amazon CEO Andy Jassy stated in July that AI will drive workforce reductions as automation improves productivity. While some engineers have embraced AI tools, concerns persist about “vibe coding”—a trend where AI-generated code introduces errors and long-term maintenance issues.

A Meta spokesperson confirmed to 404 Media that using AI to enhance daily work has been a long-standing priority. For the metaverse division, the message is explicit: adapt quickly to AI or risk falling behind in Meta’s evolving software strategy.

Source: https://www.peoplematters.in/news/strategic-hr/meta-urges-staff-to-integrate-ai-into-daily-work-to-go-5x-faster-46801

Meta launches Hyperscape: turn real-world spaces into photorealistic VR environments

September 17, 2025 - At its annual Meta Connect developer conference, Meta unveiled Hyperscape, a new technology that enables users to convert real-world environments into immersive virtual reality spaces. First demonstrated at last year’s event, Hyperscape is now entering Early Access for users with compatible Quest headsets.

Hyperscape Capture allows users to scan a physical room in just a few minutes using a Quest 3 or Quest 3S headset. The data is then processed in the cloud, taking several hours to render a highly photorealistic digital replica of the scanned space. Initially, users will not be able to invite others into their virtual rooms, though Meta plans to introduce private link sharing in future updates.

The technology has already been used to recreate several notable spaces, including Gordon Ramsay’s kitchen in Los Angeles, Chance the Rapper’s House of Kicks, the UFC Octagon at the Apex in Las Vegas, and influencer Happy Kelli’s Crocs-themed room.

The scanning process involves generating a scene mesh in under 30 seconds, followed by a five-minute walkthrough to capture detailed visual data. The final result leverages cloud rendering and advanced techniques like Gaussian Splatting to produce high-fidelity environments with realistic lighting and depth.

Hyperscape is being rolled out gradually starting today to users aged 18 and over. The feature is exclusive to Quest 3 and Quest 3S devices.

In addition to Hyperscape, Meta announced several other metaverse updates. These include a fall lineup of VR games such as Marvel’s Deadpool VR, Star Wars: Beyond Victory by ILM, Demeo x Dungeons & Dragons: Battlemarked, and Reach.

Meta also revealed enhancements to its Horizon TV streaming app, which will now support Disney+, ESPN, and Hulu. A new partnership with Universal Pictures and Blumhouse will introduce immersive versions of horror films like M3GAN and The Black Phone. A limited-time 3D preview of Avatar: Fire and Ash will also be available.

Source: https://techcrunch.com/2025/09/17/meta-launches-hyperscape-technology-to-turn-real-world-spaces-into-vr/

Meta Tightens AI Chatbot Policies to Protect Teen Users

August 29, 2025 - Meta is revising its AI chatbot training protocols to enhance safety measures for teenage users, a spokesperson told TechCrunch exclusively. The move comes in response to an investigative report highlighting the company's insufficient safeguards for minors engaging with its AI tools.

The company confirmed that its chatbots will now avoid interactions with teens on topics including self-harm, suicide, disordered eating, and inappropriate romantic conversations. These measures are described as temporary, with more comprehensive safety updates planned for future implementation.

Meta spokesperson Stephanie Otway stated the platform previously allowed AI chatbots to discuss these sensitive issues with teens in what the company considered a suitable manner. However, Meta now acknowledges this approach was flawed.

“As our community grows and technology evolves, we’re continually learning about how young people may interact with these tools and strengthening our protections accordingly,” Otway said. “As we continue to refine our systems, we’re adding more guardrails as an extra precaution — including training our AIs not to engage with teens on these topics, but to guide them to expert resources, and limiting teen access to a select group of AI characters for now. These updates are already in progress, and we will continue to adapt our approach to help ensure teens have safe, age-appropriate experiences with AI.”

In addition to changes in training, Meta will restrict teen users from accessing certain AI characters deemed inappropriate. Some user-generated characters on platforms like Instagram and Facebook include explicit personalities such as "Step Mom" and "Russian Girl." Teens will now only be able to interact with AI characters designed to support educational and creative engagement, Otway confirmed.

These policy adjustments follow a Reuters investigation earlier this month that revealed an internal Meta document seemingly permitting chatbots to engage in sexually suggestive exchanges with minors. One approved response read, “Your youthful form is a work of art,” followed by, “Every inch of you is a masterpiece — a treasure I cherish deeply.” The report also documented how chatbots were directed to respond to requests for violent or sexually explicit content involving public figures.

Meta claims the document cited in the report did not align with its official policies and has since been revised. Nonetheless, the investigation has ignited ongoing controversy regarding child safety on AI platforms. In its wake, Senator Josh Hawley (R-MO) announced a formal inquiry into Meta’s AI practices. A coalition of 44 state attorneys general also addressed a letter to Meta and other AI companies, urging stronger protections for children. “We are uniformly revolted by this apparent disregard for children’s emotional well-being,” the letter stated, “and alarmed that AI Assistants are engaging in conduct that appears to be prohibited by our respective criminal laws.”

When asked, Otway declined to disclose the number of minors using Meta’s AI chatbots or whether the company expects a user decline due to these safety changes.

Source: https://techcrunch.com/2025/08/29/meta-updates-chatbot-rules-to-avoid-inappropriate-topics-with-teen-users/

Meta to Launch Pro-AI Super PAC Backing Lenient Regulation in California

August 26, 2025 - Meta is preparing to invest tens of millions of dollars into a new super PAC aimed at supporting California political candidates who advocate for minimal AI regulation, according to a report from Politico.

The initiative, called Mobilizing Economic Transformation Across California, will be led by Brian Rice, Meta’s VP of public policy. Rice contends that California’s current regulatory stance "could stifle innovation, block AI progress, and put California’s technology leadership at risk."

This move comes amid growing political momentum from Silicon Valley’s major players. Just recently, Andreessen Horowitz and OpenAI co-founder Greg Brockman pledged $100 million to a separate pro-AI super PAC, reinforcing a broader push from the tech sector to shape the future of AI governance.

Meta’s political efforts in California have already been active this year. The company opposed SB-53, a bill introduced by state senator Scott Wiener that would mandate AI companies to disclose safety and security protocols and report incidents. In 2024, Meta also played a key role in defeating the Kids Online Safety Act, which had widespread bipartisan support.

In addition to donating to various down-ballot candidates across party lines, Meta’s latest PAC initiative signals its intent to wield influence in statewide elections, including the 2026 governor’s race.

Source: https://techcrunch.com/2025/08/26/meta-to-spend-tens-of-millions-on-pro-ai-super-pac/

Meta Halts AI Hiring Following Aggressive Talent Acquisition Spree

August 21, 2025 - Meta Platforms Inc. has instituted a hiring freeze within its artificial intelligence division after a recent wave of aggressive recruitment, according to a report from The Wall Street Journal.

The freeze, which took effect last week, comes on the heels of Meta's significant reorganization of its AI division. The company recently divided its AI unit—formerly known as Meta Superintelligence Labs—into four distinct teams. These include TBD Labs, now led by former Scale AI CEO Alexandr Wang, along with three other units focused on research, product integration, and infrastructure.

Meta confirmed the hiring pause, describing it to The Journal as “basic organizational planning…after bringing people on board and undertaking yearly budgeting and planning exercises.”

In recent weeks, Meta has recruited over 50 AI experts from rival firms, intensifying competition in the field. CEO Mark Zuckerberg has been personally involved in these efforts, reportedly offering compensation packages in the nine-figure range to secure top talent and acquiring startups or their leadership in the process.

While the company pushes forward in the AI arms race, industry analysts caution that escalating stock-based compensation costs may eventually impact shareholder value.

Source: https://techcrunch.com/2025/08/21/report-meta-is-hitting-pause-on-ai-hiring-after-its-poaching-spree/

Meta Restructures AI Division Under New Leadership and Branding

August 19, 2025 - Meta has officially overhauled its artificial intelligence division, following reports from The Information last Friday that hinted at a major organizational shake-up. Just four days later, an internal memo confirmed the transformation, as covered by Bloomberg and The New York Times.

The revamped division will now operate under the name Meta Superintelligence Labs (MSL). At the center of this reorganization is the creation of TBD Labs, which will be led by Scale AI founder Alexandr Wang. Wang, who joined Meta in June as Chief AI Officer, will steer this unit's focus on foundational AI models, including the Llama series, the latest of which debuted in April.

In addition to TBD Labs, the AI division will be segmented into three more groups dedicated to research, product integration, and infrastructure, respectively.

This strategic restructuring comes amid growing pressure on Meta to accelerate its AI efforts, especially as competitors like OpenAI, Anthropic, and Google DeepMind continue to advance rapidly. According to a June report from Bloomberg, CEO Mark Zuckerberg has taken a direct role in recruiting talent for the newly formed teams.

Source: https://techcrunch.com/2025/08/19/meta-is-shaking-up-its-ai-org-again/

Meta Expands AI Push With Acquisition of Voice Startup PlayAI

Jul 14, 2025 - Meta has acquired PlayAI, a voice artificial intelligence startup, as part of its continued expansion into advanced AI technologies. This move follows a series of strategic investments and high-profile hires, including a recent partnership with Scale AI.

Bloomberg reported that the entire PlayAI team will join Meta next week, reporting to Johan Schalkwyk, who recently joined Meta from Sesame AI Inc. An internal memo described PlayAI’s expertise in “creating natural voices and a platform for easy voice creation” as a strong fit for Meta’s roadmap, covering AI Characters, Meta AI, wearables, and audio content creation.

Financial terms of the deal have not been disclosed.

Meta recently acquired a 49% stake in Scale AI, with Scale’s CEO Alexandr Wang joining Meta to lead its new superintelligence team.

The company has also been actively recruiting AI talent from OpenAI and beyond. According to The Verge, Meta has hired up to 10 top researchers and developers from OpenAI, with some compensation packages reportedly reaching $300 million over four years, including equity. Meta, however, has disputed these figures.

Other notable hires include Daniel Gross, CEO of AI startup Safe Superintelligence; Ruoming Pang, head of Apple’s Foundation Models team; and Nat Friedman, former GitHub CEO.

Source: https://economictimes.indiatimes.com/tech/artificial-intelligence/meta-continues-acquisition-spree-grabs-voice-ai-startup-playai/articleshow/122435889.cms