According to a post on Threads, Meta CEO Mark Zuckerberg announced on Monday that the company will launch its first AI data center, called Prometheus, in 2026.
Meta is also building a second supercluster, called Hyperion, which is expected to grow to five gigawatts over several years. Zuckerberg noted that Meta is working on several additional data center clusters, one of which will span a portion of Manhattan.
This data center expansion is part of Meta’s broader effort to develop superintelligence, a category of AI that the company differentiates from general artificial intelligence.
What Zuckerberg announced about Prometheus, Hyperion and Meta’s AI investment scale
Zuckerberg’s post outlined a series of upcoming developments. Prometheus is expected to come online in 2026, while Hyperion will be expanded to five gigawatts over several years.
META is building several superclusters, one of which is planned to cover a large portion of Manhattan’s footprint. The company intends to finance these projects through its internal cash flows and business operations, avoiding external funding sources.
Zuckerberg noted, “We’re also building several more Titan clusters. Just one of these covers a significant portion of Manhattan’s footprint.” The context of Manhattan-scale infrastructure underscores the huge physical and energy requirements required for AI training and inference these days.
Meta has announced plans to invest hundreds of billions of dollars in computing infrastructure with the aim of developing superintelligence. The company explained that the funding will come from its business operations, leveraging capital generated from the existing strong performance of its core business. The scale of investment reflects Meta’s estimate of the computing power required to achieve superintelligence.
Alexander Wang, Meta’s chief AI officer, said, “For our superintelligence effort, I’m focused on building the most elite and talent-dense team in the industry. We’re also going to invest hundreds of billions of dollars in computation to build superintelligence.”
How Meta Frames Superintelligence vs General AI and Its Strategy
Meta distinguishes his work on superintelligence from normal AI. According to Wang and company, general AI refers to systems trained to perform a wide variety of tasks, while superintelligence describes systems that surpass human ability in a wide range of domains. This difference affects how Meta justifies its investments in infrastructure.
The designation of superintelligence also aligns with Meta’s competitive stance against companies like OpenAI, Anthropic, and Google, each of which uses different terminology to describe the cutting-edge of AI development.
The Prometheus and Hyperion announcements complement other Meta AI initiatives, including Muse Spark 1.1, a coding-focused model now in US public preview with competitive pricing; Muse Image, an image creation model that defaults to using public Instagram profiles; Meta Superintelligence Labs, the internal team that developed these models; and Ambient Meta AI, which provides assistive features in Meta’s apps.
Meta’s infrastructure expansion supports its contributions to agentic AI, coding tools, image generation, and consumer-facing AI features. Without extensive training and inference capabilities, Meta cannot compete at the forefront with companies like OpenAI, Anthropic, and Google.
Talent incentives, energy and infrastructure concerns, and what users should look forward to
Meta is actively recruiting AI talent, offering compensation packages for senior engineers reportedly reaching hundreds of millions of dollars. This talent acquisition strategy supports the company’s infrastructure investments. Wang said Meta aims to build the most elite and talent-rich team in the industry.
Meta’s approach differs from that of OpenAI, Anthropic and Google, which have historically relied on small, mission-focused teams rather than mass hiring to build comprehensive AI research and engineering capabilities.
The mention of the Manhattan-scale project and the five-gigawatt Hyperion cluster raises important questions about energy use and infrastructure, including where the electricity will come from, how grid stability will be maintained, whether local communities will absorb rate increases to finance the buildout, and how environmental impacts will be assessed.
These concerns are consistent with recent regulatory actions. New York has halted permits for data centers 50MW and larger while developing new rules. Maine attempted to enact a full statewide moratorium earlier this year, but it was vetoed. President Trump has called on Big Tech to take responsibility for the power consumption of data centers.
Meta’s planned five-gigawatt Hyperion cluster is 100 times larger than New York’s 50MW limit for data center permits, reflecting the scale of the infrastructure the company will develop. How Meta secures the necessary permits, energy agreements and community benefits will impact whether their timeline remains viable.
For users following Meta’s AI development, with Prometheus expected to go live in 2026, its impact on Meta’s AI capabilities will become evident through advanced product features. Hyperion will be launched gradually over several years, with a gradual increase in capacity.
There are additional superclusters in development, but their names and timelines have not been announced. As construction continues, community-level impacts such as grid stress and environmental impacts are likely to emerge.
For users comparing Meta’s AI products to competitors: Meta’s investment in infrastructure indicates that its AI offering will continue to expand on the technology front. Unlike many AI companies that rely on infrastructure sharing, Meta builds its own data centers.
This approach can allow rapid iteration and widespread access to models, as seen with releases such as Muse Spark 1.1 under the Apache 2.0 license, which can benefit from META’s substantial infrastructure investment.
Meta’s data center growth comes amid changes in the industry, where infrastructure has become a key competitive factor. Major players such as Microsoft, Google, Amazon and Meta operate large-scale data center networks. Meanwhile, emerging competitors like Anthropic and OpenAI rely on partnerships with cloud providers like Amazon, Nvidia and Microsoft for access.
The race for infrastructure capacity increasingly determines which companies can operate at the highest levels. Meta’s Hyperion, designed to scale up to five gigawatts, puts the company alongside the largest hyperscalers for AI training and inference.
Prometheus is expected to be ready in 2026. Construction of Hyperion is ongoing over several years, and additional superclusters are in development without specific timelines. Those interested can follow Meta’s official announcements for updates on the rollout of the infrastructure and its impacts on AI products. Local news sources where Meta is building the data center are also likely to provide information about community impacts and progress in permitting.
Thanks for being a Ghax reader. The post Meta has announced the first AI data center, Prometheus, coming online in 2026 with more supercluster plans appeared first on Ghacks.
