Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchMeta has expanded its El Paso, Texas AI data center investment sixfold to $10 billion, targeting 1GW of capacity by 2028, while simultaneously preparing to launch prescription-ready Ray-Ban AI glasses to tap a $223B eyewear market.
AI, startups & growth insights. No spam.
Meta is placing two enormous bets in the same week — one measured in gigawatts, the other in diopters. On March 26, 2026, the company announced it was increasing its planned investment in a West Texas AI data center from $1.5 billion to more than $10 billion, a more-than-sixfold jump that underscores the relentless capital arms race in AI infrastructure. One day later, Bloomberg reported that Meta was preparing to launch its first Ray-Ban smart glasses models designed specifically for prescription wearers, targeting the roughly 1.5 billion people worldwide who require corrective lenses. Together, the moves reveal a company sprinting on two very different tracks: building the compute backbone to train its most powerful AI models ever, and racing to put that AI on the faces of everyone who walks into an optician's office.
When Meta broke ground on its El Paso data center campus in October 2025, the company pegged the project at $1.5 billion — a substantial figure by any normal measure, but modest by the standards of 2026 AI infrastructure spending. Five months later, that number is $10 billion.
The expansion was reported simultaneously by Bloomberg and CNBC on March 26, 2026, and confirmed directly by Meta on the company's data center blog. In a post titled "Big Things Are Happening, El Paso," Meta described the site as central to its long-term AI infrastructure roadmap.
The revision is not a sign of a project gone wrong — it is the opposite. Meta's AI ambitions have grown faster than any single construction timeline could anticipate. In January 2026, the company announced plans to spend up to $135 billion on AI across the full year. The El Paso expansion is one data point in a broader capital deployment story that spans dozens of sites, multiple GPU generations, and an unprecedented appetite for compute.
West Texas was chosen for multiple practical reasons: land is abundant and relatively inexpensive, the climate is dry (which matters enormously for cooling system design), and the region sits within reach of large renewable energy installations in the broader Texas grid. The original 1.2-million-square-foot site was already substantial; at $10 billion, it becomes one of the single largest data center investments ever made at a single location by any technology company.
The headline capacity figure — 1 gigawatt by approximately 2028 — deserves unpacking, because the number is extraordinary even in the context of 2026 hyperscaler ambitions.
To put it in perspective: a large traditional data center might draw 50 to 100 megawatts. Hyperscale facilities routinely target 200 to 500 megawatts. A single gigawatt is 1,000 megawatts — the output of a full-scale nuclear power plant, devoted entirely to AI computation. Meta's El Paso campus, when complete, would rank among the largest AI compute facilities on earth.
The 2028 timeline is significant. That is when NVIDIA's Rubin GPU architecture — the successor to the current Blackwell generation — is expected to be in full production deployment at scale. A facility coming online in 2028 is being designed now with next-generation hardware in mind, meaning the infrastructure investments being made today are targeted at training and inference workloads that do not yet fully exist.
Meta has committed to matching 100% of the facility's electricity consumption with renewable energy. The company also stated a goal of adding over 5,000 megawatts of clean power to the Texas grid over the life of the project — a commitment that extends well beyond the data center itself and into broader regional energy infrastructure.
Meta's investment carries direct economic consequences for El Paso, one of the largest cities on the US-Mexico border and a region that has historically lagged wealthier Texas metros in technology sector employment.
The company announced 300 permanent full-time jobs tied to the facility, alongside a construction workforce expected to peak at more than 4,000 workers during the build phase. The KVIA local news report noted that the original announcement had projected 3,000 construction workers; the revised $10 billion plan raised that figure.
Beyond direct employment, Meta committed $500,000 in grants for a workforce development partnership with El Paso public schools — a signal that the company intends to cultivate local technical talent rather than importing labor entirely from outside the region. Over $8 million is being invested in local infrastructure improvements.
Water usage drew particular attention, given El Paso's location in one of the driest regions of North America. Meta addressed this directly: the data center will use a closed-loop liquid cooling system that the company says consumes zero water for cooling during the majority of the year — a stark contrast to traditional air-cooled facilities that depend on evaporative cooling towers. Meta also partnered with nonprofit DigDeep to bring clean running water to more than 100 homes in the surrounding region, and committed to restoring 200% of the water the facility does consume back to local watersheds.
The company framed this as achieving water-positive status — a commitment that, if met, would mean Meta's presence in El Paso improves the regional water situation rather than straining it.
The El Paso facility is not being built around commodity hardware. It is being designed around NVIDIA's most advanced GPU architectures, in a partnership that NVIDIA's own newsroom described as multiyear and multigenerational.
Meta will deploy both Blackwell GPUs — the current flagship generation, which NVIDIA reports enables 3x faster training and nearly 2x better performance per dollar compared to the previous Hopper architecture — and Rubin GPUs, the next generation that began production in early 2026. The partnership, expanded and detailed in a February 2026 CNBC report, secured Meta a healthy allocation of both chip generations at a time when NVIDIA hardware remains constrained.
One detail that distinguishes Meta's approach: the company is becoming among the first hyperscalers to deploy NVIDIA's Grace CPUs as standalone processors — not paired with GPUs in the traditional NVL server configuration, but independently handling the CPU workloads that do not require GPU acceleration. This architectural choice allows Meta to right-size compute resources across different workload types rather than over-provisioning expensive GPU nodes for tasks that do not need them.
Engineering teams from both companies will collaborate in what NVIDIA described as "deep codesign to optimize and accelerate state-of-the-art AI models" for Meta's specific use cases. This is not simply a procurement relationship — it is a technical partnership in which chip design decisions and model architecture decisions influence each other.
The El Paso investment is not purely about serving existing products at greater scale. It is about training what comes next.
Meta has been developing a frontier model internally codenamed "Avocado," described as a successor to its Llama AI model family. While Llama 4 — released with the Maverick and Scout variants — represented a significant generational step, Avocado appears to be the company's push into the highest tier of frontier AI capability.
Training models at this level requires compute at a scale that even Meta's current infrastructure struggles to support. Llama 4 was already trained on a cluster of more than 100,000 NVIDIA H100 GPUs — what Mark Zuckerberg called "bigger than anything that I've seen." Avocado will require more. The NVIDIA partnership, the Rubin GPU allocation, and the 1GW El Paso campus are all pieces of the same infrastructure puzzle oriented around training that next-generation model.
This matters for the broader AI landscape because Meta's models are released under open-weight licenses that allow developers, researchers, and companies worldwide to use and build upon them. The investment in El Paso is therefore not just building capacity for Meta's own products — it is funding the open-source foundation that a large fraction of the AI ecosystem depends on.
The infrastructure story is about scale. The glasses story is about reach.
One day after the El Paso announcement, Bloomberg reported that Meta is preparing to debut two new Ray-Ban smart glasses models — codenamed Scriber and Blazer — designed specifically for prescription wearers. The launch is expected as early as early April 2026.
Both models were surfaced through FCC filings before any official announcement, a common pattern for consumer electronics launches. They are non-display glasses — no screen, no augmented reality overlay — but they carry the full suite of AI capabilities that has made the existing Ray-Ban Meta line a commercial success: camera, microphone, speakers, and full Meta AI integration for voice queries, real-time translation, and visual analysis.
What distinguishes Scriber and Blazer from the existing Ray-Ban Meta frames is their form factor and distribution strategy. Existing Ray-Ban Meta glasses can technically accept prescription lenses, but the frames were designed primarily as sunglasses and consumer electronics products. The new models are designed from the outset for optical retail channels — the LensCrafters, Sunglass Hut, and independent optician shops where the majority of prescription eyewear is purchased.
Both models include Wi-Fi 6 UNII-4 band support, an upgrade over the current generation. Blazer will be offered in regular and large sizes; Scriber appears to ship in a single-size option. Both will include charging cases.
Pricing has not been publicly confirmed ahead of launch. The Next Web's reporting on the FCC filings noted the rectangular and rounded frame styles respectively, chosen to appeal to a broader range of prescription frame preferences than the sport-adjacent Ray-Ban silhouettes that defined the first generation.
The prescription glasses announcement is easy to frame as a minor product update. It is not. It is a potential step-change in the addressable market for AI wearables.
Meta sold more than 7 million Ray-Ban and Oakley AI frames in 2025 — a genuinely impressive figure for a product category that barely existed three years ago. But that number looks different when set against the scale of the global eyewear market. Prescription eyewear accounts for roughly 69% of a $223 billion global market. Approximately 1.5 billion people worldwide wear corrective lenses. Against that backdrop, 7 million units is a rounding error.
The fundamental barrier for much of that population has not been price or interest — it has been practicality. For someone who depends on prescription lenses to see, a pair of smart glasses that cannot replicate their prescription is not a primary eyewear option. It is an accessory they would carry in addition to their regular glasses. That math rarely works in favor of adoption.
By designing frames specifically for optical retail channels and working with EssilorLuxottica — the world's largest eyewear company, which also manufactures the existing Ray-Ban Meta line — Meta is attempting to solve the distribution and fitting challenge at scale. Opticians already have the fitting tools, prescription data, and customer trust required to sell corrective eyewear. The companies are reportedly considering doubling production capacity to 20 million units annually, a figure that would only make sense if prescription adoption materially accelerates sales volume.
There are headwinds. Solos Technology has filed patent litigation against Meta seeking damages described as "multiple billions of dollars," which could complicate near-term operations. Opticians accustomed to selling passive eyewear will need training to position and sell connected devices effectively. And the non-display constraint means that Scriber and Blazer compete on AI utility alone — there is no visual overlay to differentiate them from simply wearing AirPods.
But if the strategy works, the prescription glasses market represents the largest untapped distribution channel in consumer AI hardware.
It would be easy to read the data center announcement and the glasses announcement as unrelated stories that happened to land in the same week. They are not.
Meta's hardware success and its AI model quality are deeply intertwined. The Ray-Ban smart glasses are compelling precisely because Meta AI — the voice assistant integrated into the frames — is genuinely capable. It can identify objects in the wearer's field of view, answer complex questions in real time, translate conversations, and navigate tasks through voice alone. That capability is the product of the training infrastructure Meta has been building for years.
As El Paso comes online and Avocado succeeds Llama, the AI models running inside those glasses will become meaningfully more powerful. A multimodal model trained at 1GW scale on NVIDIA Rubin hardware will deliver a qualitatively different assistant experience than one trained at current capacity. The infrastructure investment today is the glasses experience in 2028 and beyond.
The reverse is also true. Every pair of prescription Ray-Bans sold is a data point — a real-world interaction with Meta AI in a context the company has not previously had access to. Glasses-wearers using Meta AI throughout their daily lives generate feedback signals that inform what capabilities matter most, what failure modes are most frustrating, and what the next generation of frontier models should prioritize. The product and the infrastructure build each other.
This is the same logic that has guided Meta's overall AI strategy since the company restructured its teams and consolidated AI leadership — a move that eliminated redundant roles to focus execution on exactly this kind of deep integration between infrastructure, model development, and consumer products.
Meta's Q1 2026 has been defined by infrastructure commitments at a scale that would have seemed implausible three years ago. The $10 billion El Paso announcement follows a January declaration of up to $135 billion in full-year AI spending, multiple GPU supply agreements with NVIDIA, and a string of open-weight model releases that have established Llama as the dominant open-source AI family.
The next major milestones to watch:
Llama 4 derivatives and Avocado timeline. The current Maverick and Scout models represent Llama 4's capabilities at launch. Meta typically releases additional variants — including smaller, more efficient models — in the months following a major release. Avocado remains on an undisclosed training timeline, but the El Paso capacity target of 2028 suggests the most ambitious compute runs are still being planned.
Prescription glasses launch. Scriber and Blazer are expected as early as April 2026. Adoption through optical retail channels will be the key metric — not just unit sales, but whether the channel can be trained to sell connected hardware effectively.
El Paso construction milestones. The 4,000-worker peak construction workforce suggests a multi-year build-out. Progress milestones in late 2026 will indicate whether the 2028 capacity target is on track.
Broader AI infrastructure competition. Microsoft, Google, Amazon, and xAI are all making investments of comparable scale. The race for AI compute capacity is far from over, and the winner of that race is likely to have a structural advantage in the models that can be trained — and the products that can be built on top of them.
In a single week, Meta demonstrated the full range of its AI ambitions: a $10 billion bet on the physical infrastructure that will train its most powerful models, and a prescriptions-ready hardware product that could put those models on the faces of a billion additional users. Neither move is speculative. Both are grounded in real construction timelines, confirmed supplier relationships, and documented market gaps.
The El Paso data center and the Scriber and Blazer glasses are separated by years and by scale, but they are expressions of the same thesis: that AI is infrastructure, and that the companies building the deepest infrastructure today will have the widest reach tomorrow. Meta is building for both.
Sources: Bloomberg — Meta Increases Investment in El Paso Data Center to $10 Billion · CNBC — Meta to spend $10 billion on AI data center in El Paso, 1GW by 2028 · Bloomberg — Meta to Launch New AI Glasses Aimed at Prescription Wearers · The Next Web — Meta launches prescription Ray-Ban smart glasses · Meta Data Centers — Big Things Are Happening, El Paso · CNBC — Meta expands Nvidia deal to use millions of AI chips · NVIDIA Newsroom — Meta Builds AI Infrastructure With NVIDIA · KVIA — El Paso Meta data center investment grows to $10 billion
Arm announced the AGI CPU on March 24, its first production silicon in 35 years: a 136-core, TSMC 3nm processor built for agentic AI workloads and co-developed with Meta, promising 2x rack performance over x86 and up to $10B in data center CAPEX savings per gigawatt.
Meta will discontinue Horizon Worlds on Quest headsets by June 2026, marking the definitive end of its metaverse era and freeing billions for AI development under Zuckerberg's new vision.
Meta commits up to $100 billion for custom AMD Instinct MI450 GPUs and 6th Gen EPYC Venice CPUs across a 6-gigawatt deployment — the largest AI hardware partnership ever signed.