AI infrastructure spending is no longer a line item, it’s the whole budget. Meta has just announced a multi-year partnership with Nvidia, giving it access to the chip giant’s latest processors as it gears up to spend between $115bn and $135bn on AI infrastructure in 2026 — nearly double the $72.22bn it spent in 2025. If you thought the AI arms race was slowing down, think again.
The Deal: What Meta and Nvidia Actually Agreed To
Meta will roll out millions of Nvidia’s Blackwell and Rubin GPUs across its infrastructure for AI training. The partnership covers chips, networking, and software — essentially a full-stack collaboration designed to squeeze every bit of performance out of Meta’s systems.
Nvidia CEO Jensen Huang put it plainly: no one deploys AI at Meta’s scale. With billions of users across Facebook, Instagram, and WhatsApp, Meta’s personalisation and recommendation engines are arguably the largest in the world. This deal is about keeping that machine running and making it faster.
CPUs Are Coming Too
Beyond GPUs, both companies are preparing for a large-scale rollout of Nvidia’s new CPUs from 2027. Nvidia says these have been engineered for greater energy efficiency, which matters a lot when you’re running data centres at the scale Meta is targeting.
WhatsApp First, Then Everywhere Else
Meta has already started deploying Nvidia’s computing tech inside WhatsApp’s private processing framework. The setup lets AI-powered features run while keeping user data secure — a meaningful detail given ongoing scrutiny around data privacy.
The plan is to extend the same approach to Instagram and Meta’s broader product suite. Think smarter recommendations, faster content moderation, and more capable AI assistants — all powered by Nvidia silicon.
The Infrastructure Bill Is Enormous
Meta has guided for between $115bn and $135bn in capital expenditure for 2026 — nearly double the $72.22bn it spent in 2025 — with the vast majority earmarked for AI-related infrastructure. Here’s where the money is going:
- A $27bn joint venture data centre campus (the Hyperion campus in Richland Parish, Louisiana, co-developed with private equity firm Blue Owl Capital, which holds an 80% stake)
- Agreements with four nuclear energy companies — Constellation Energy, Vistra, TerraPower, and Oklo — supporting up to 6.6GW of clean energy by 2035
- A new ‘Meta Compute’ initiative targeting tens of gigawatts of computing capacity by the end of the decade
Senior executives have reportedly been handed specific mandates around long-term capacity planning and government engagement to support the build-out. This isn’t a tech project — it’s an infrastructure programme.
Why This Matters Beyond Meta
Meta’s 2026 spending plans signal where the entire industry is heading. When one company is willing to nearly double its capital expenditure on AI infrastructure year-on-year, it sets expectations — and competitive pressure — for everyone else. Microsoft, Google, and Amazon are all in the same race, and Nvidia sits at the centre of it all.
For investors watching the AI infrastructure theme, this deal reinforces Nvidia’s position as the picks-and-shovels play in the AI gold rush. It also raises legitimate questions about when — or whether — these investments will generate the returns to justify the spend.
The Bottom Line
Meta and Nvidia’s partnership is one of the most significant AI infrastructure deals of 2026, and the $115bn–$135bn capital expenditure commitment makes clear that Zuckerberg isn’t blinking. Whether the returns match the ambition remains to be seen — but the build-out is accelerating fast.
Want to stay on top of the AI infrastructure story? Keep an eye on Meta’s quarterly earnings — that’s where the real scorecard will emerge.
FAQ
Q1: Why is Meta spending so much on AI infrastructure in 2026?
A: Meta announced at its Q4 2025 earnings call that it expects to spend between $115bn and $135bn on AI capex in 2026 — nearly double the $72.22bn it spent in 2025. The investment is aimed at powering its recommendation systems, ad targeting, and new AI products across Facebook, Instagram, and WhatsApp.
Q2: What are Nvidia Blackwell and Rubin GPUs?
A: These are Nvidia’s latest generations of graphics processing units, designed specifically for demanding AI training workloads. Meta will deploy millions of them as part of its expanded infrastructure.
Q3: What is the Meta Compute initiative?
A: It’s Meta’s long-term plan to build tens of gigawatts of computing capacity by the end of the decade, supporting future AI systems at a scale that current infrastructure can’t handle.
Q4: Why is Meta investing in nuclear power?
A: AI data centres are extraordinarily energy-intensive. Meta has signed agreements with four nuclear energy companies — Constellation Energy, Vistra, TerraPower, and Oklo — targeting up to 6.6GW of clean energy capacity by 2035.
Q5: What does this mean for Nvidia’s business?
A: It further cements Nvidia’s dominance as the go-to supplier for AI compute. Multi-year, multi-generational deals with hyperscalers like Meta provide long-term revenue visibility and reinforce its competitive moat.
DISCLAIMER
Effective Date: 15th July 2025
The information provided on this website is for informational and educational purposes only and reflects the personal opinions of the author(s). It is not intended as financial, investment, tax, or legal advice.
We are not certified financial advisers. None of the content on this website constitutes a recommendation to buy, sell, or hold any financial product, asset, or service. You should not rely on any information provided here to make financial decisions.
We strongly recommend that you:
- Conduct your own research and due diligence
- Consult with a qualified financial adviser or professional before making any investment or financial decisions
While we strive to ensure that all information is accurate and up to date, we make no guarantees about the completeness, reliability, or suitability of any content on this site.
By using this website, you acknowledge and agree that we are not responsible for any financial loss, damage, or decisions made based on the content presented.
