What is Meta MTIA? Meta MTIA (Meta Training and Inference Accelerator) is Meta's in-house family of custom AI silicon, designed to run the company's AI workloads without relying exclusively on Nvidia GPUs. On its Q1 2026 earnings call, Meta guided 2026 capital expenditures to $115 billion to $135 billion, nearly doubling the $72.22 billion it spent in 2025, and confirmed that its first MTIA generation has entered data centers while three more chips are scheduled across 2026 and 2027. For commercial real estate investors, this is the most consequential chip and capex announcement of the month. For a broader view, see our guide to AI commercial real estate.
Key Takeaways
- Meta guided 2026 capex to $115 billion to $135 billion, up from $72.22 billion in 2025, placing it fourth among hyperscalers by absolute spend.
- MTIA 300 is already deployed in Meta data centers; MTIA 400 has completed testing and is being rolled in with liquid-cooled server systems designed around the chip.
- MTIA 450 and MTIA 500 arrive in 2027, targeting inference workloads, with the next-generation chip built on a 2-nanometer process node.
- Meta signed over $10 billion with Google Cloud, $14.2 billion with CoreWeave, $3 billion with Nebius, and is reportedly negotiating a $20 billion Oracle contract to supplement in-house capacity.
- Meta's capex intensity of roughly 55% to 67% of projected revenue is unprecedented for a profitable tech company with no cloud services business to monetize the spend.
- For CRE investors, the combination of hyperscale self-chip adoption and record neocloud contracts signals sustained power, cooling, and land demand through at least 2027.
What Meta MTIA Chips Mean for Data Center Real Estate
The MTIA program answers a simple question: can Meta reduce its dependence on Nvidia without slowing its AI build out? By shipping MTIA 300 in late 2025 and now rolling MTIA 400 into production, Meta has effectively proven that hyperscalers can manage GPU supply risk with in-house application specific integrated circuits (ASICs). Meta VP of Engineering Yee Jiun Song told reporters that inference demand is exploding, and the company is releasing a new MTIA chip roughly every six months, which is an unusually fast cadence for custom silicon.
For CRE investors in data center, power, and land near hyperscaler campuses, the implications are concrete. Meta's footprint already spans over 50 million square feet across 26 campuses in the United States and four international sites. That footprint is now being upgraded, not replaced, to accept liquid-cooled MTIA server systems. Liquid cooling changes floor plate requirements, chilled water infrastructure, and power density tolerances, and data center landlords with triple-net leases to Meta are seeing tenant-funded retrofits accelerate. Our deep dive on CoreWeave, Nebius, and Meta's $35 billion neocloud commitments walks through the leasing mechanics.
Why $115 Billion to $135 Billion in 2026 Capex Matters
Meta's guided capex of $115 billion to $135 billion is not simply larger than 2025, it is structurally different. Industry analysts estimate that every $10 billion of hyperscale capex translates into roughly 150 to 200 megawatts of incremental data center demand, depending on chip density and workload mix. At the midpoint of Meta's guidance, that implies close to 2 gigawatts of new power draw attributable to Meta alone in 2026, before counting overflow leases to CoreWeave, Nebius, and Oracle.
Crucially, Meta has no cloud services business, which means every dollar of this capex must be justified by internal workload economics. That constraint sharpens the signal for CRE investors: Meta will not build speculative capacity for third-party tenants. Any power or land adjacent to a Meta campus that clears market is backed by direct occupier demand, not pass-through cloud revenue. For investors underwriting data center acquisitions, this is a materially different credit story than a CoreWeave build-to-suit.
HBM Supply, 2-Nanometer Chips, and Power Density
The MTIA roadmap has three features that translate directly into data center underwriting assumptions. First, the chips use more high-bandwidth memory (HBM) to accelerate generative AI inference, and HBM supply is tight industry-wide. Meta says its 2026 HBM supply is secured, but the constraint continues to drive premium pricing for AI-optimized data center space with pre-negotiated supply chains. Second, MTIA 450 and the 2-nanometer MTIA 500 push power density higher per rack, which rewards facilities built for 50 kilowatts per rack and above.
Third, the six-month chip release cadence forces hyperscaler data center operators to design for refresh cycles that are shorter than the typical 10 to 15 year lease amortization window. CRE investors evaluating Meta-adjacent sites should pay close attention to lease flexibility, expansion rights, and power redundancy assumptions. Our note on Anthropic's 3.5 gigawatt Google Broadcom TPU deal details similar patterns for ASIC-driven CRE.
What CRE Investors Should Do Next
The MTIA and $135 billion capex story is not a reason to chase every hyperscaler submarket. It is a reason to underwrite three specific factors. First, power availability: Meta's capex assumes substations and transmission upgrades that take 24 to 60 months, and sites with pre-qualified interconnection queues trade at material premiums. Second, water access for liquid cooling: the MTIA 400 server system requires dedicated water loops, so sites without reclaimed or non-potable water at scale may not qualify. Third, talent and operations proximity: Meta's roughly 5,000 permanent data center workers cluster in specific metros, and CRE adjacency to that labor market supports pricing power.
The AI in real estate market is projected to reach $1.3 trillion by 2030 at a 33.9% CAGR, and 92% of corporate occupiers have initiated AI programs. Yet only 5% report achieving most of their AI program goals. Meta's willingness to spend $135 billion in a single year is the clearest single data point that hyperscalers view AI capacity as a multi-year moat, not a cyclical bet. CRE investors looking for hands-on AI implementation support in underwriting hyperscaler-adjacent deals can reach out to Avi Hacker, J.D. at The AI Consulting Network. For a broader view of 2026 concentration trends, see our Stanford AI Index 2026 data center concentration analysis.
Additional reporting from CNBC and Data Center Dynamics provides further context on the capex mechanics and chip roadmap.
Frequently Asked Questions
Q: What are Meta's MTIA chips and why do they matter for CRE investors?
A: MTIA stands for Meta Training and Inference Accelerator, Meta's in-house AI chips designed to reduce Nvidia dependence. They matter for CRE investors because MTIA power density and liquid cooling requirements shape which data center buildings qualify for hyperscaler leases and at what pricing. Meta has committed $115 billion to $135 billion in 2026 capex to deploy them at scale.
Q: How does Meta's 2026 capex compare to other hyperscalers?
A: Meta's $115 billion to $135 billion places it fourth among hyperscalers by total capex, behind Microsoft, Amazon, and Google. However, Meta's capex intensity of 55% to 67% of revenue is the highest in the industry because Meta has no cloud services business to monetize the spend. Every dollar must pay back through internal advertising and product returns.
Q: What data center features should CRE investors prioritize given the MTIA roadmap?
A: Investors should prioritize sites with pre-qualified power interconnection, 50 kilowatts per rack or higher density capability, reclaimed or non-potable water access for liquid cooling, and labor market proximity to hyperscaler operations hubs. Sites missing any of these three qualities rarely clear for Meta-caliber tenants without costly retrofits.
Q: Is Meta leasing from third-party data center landlords or building everything in-house?
A: Both. Meta's in-house footprint is over 50 million square feet across 26 U.S. campuses and four international sites. On top of that, Meta signed over $10 billion with Google Cloud, $14.2 billion with CoreWeave, $3 billion with Nebius, and is reportedly negotiating a $20 billion Oracle contract. For CRE investors, this means both direct-lease and colocation opportunities are growing in parallel.
Q: How soon could MTIA chips affect my current CRE underwriting model?
A: Immediately for data center and power-related assets, and within 12 to 24 months for adjacent industrial and office markets supporting hyperscaler workforces. If you have hyperscaler campus exposure, re-underwrite power and cooling capacity assumptions now. If you are ready to transform your underwriting process with AI, The AI Consulting Network specializes in exactly this.