Meta’s newly established Meta Superintelligence Labs (MSL) could surpass Google at the forefront of artificial intelligence within the next six months, according to a report by research firm SemiAnalysis, which argues that the social media giant’s aggressive investment in talent, infrastructure and proprietary data has dramatically altered the competitive landscape.
‎The report says Meta has shifted away from relying on publicly available datasets by developing an internal reinforcement learning (RL) ecosystem.
‎‎The company has reportedly reassigned around 3,000 engineers to build a large-scale RL environment factory that generates proprietary training data for next-generation AI agents.
‎‎SemiAnalysis believes this approach gives Meta a significant competitive advantage by creating a unique data pipeline that rivals cannot easily replicate through commercial data providers.
‎‎The assessment comes as Meta expands its AI product portfolio. On Thursday, the company released developer access to its upgraded Muse Spark 1.1 model, positioning it as a competitor to the paid application programming interface (API) models offered by OpenAI and Anthropic.
‎‎Meta said the new model is designed to perform complex coding and agentic tasks more effectively and forms part of its broader ambition to develop what it describes as “personal superintelligence” capable of carrying out multi-step tasks with minimal human input.
‎‎Beyond software, the report highlights Meta’s ambitious infrastructure programme. SemiAnalysis projects that the company will overtake both OpenAI and Anthropic in total AI computing capacity by the end of the year through the construction of five gigawatt-scale AI data centre clusters.
‎The facilities will be connected through Meta’s proprietary AI-Backbone networking architecture, enabling the company to distribute large-scale AI training workloads efficiently across geographically dispersed locations.
‎‎The infrastructure expansion aligns with reports that Meta plans to invest as much as US$145 billion in AI infrastructure this year as part of a broader global rollout.
‎‎According to Reuters, citing an internal company memo, Meta intends to deploy seven gigawatts of AI computing capacity in 2026 before doubling that figure to 14 gigawatts in 2027.
‎‎The company is also preparing to begin production of its custom AI chip, codenamed Iris, in September. The chip, developed in partnership with Broadcom and manufactured by TSMC, reportedly completed bug testing within six weeks. Reuters said the programme is supported by long-term supply agreements with Samsung, SanDisk and Sumitomo Electric.
‎‎Meta has complemented its infrastructure expansion with an aggressive recruitment campaign, investing billions of dollars to attract leading AI researchers from OpenAI, Anthropic and Scale AI. The company’s US$14.3 billion investment in Scale AI forms part of a broader strategy to assemble a specialised team capable of maximising the performance of its expanding AI infrastructure.
‎‎Investors welcomed the latest developments, with Meta Platforms shares closing around 4% higher after recovering from earlier losses during Thursday’s trading session. By contrast, Alphabet shares declined about 1% as investors weighed the implications of Meta’s accelerating AI ambitions.
‎‎Although Muse Spark’s initial benchmark results have trailed some competing open-source models, SemiAnalysis argues that current performance metrics do not fully reflect Meta’s long-term prospects.
‎‎The firm contends that the pace of improvement is more significant than present capability, suggesting that if Chief Executive Mark Zuckerberg maintains the company’s current level of investment and strategic focus, Meta could establish itself as the leading AI hyperscaler while Google’s position at the industry’s frontier comes under increasing pressure.
