Meta's infrastructure chief Santosh Janardhan is stepping down

Janardhan is Meta's head of infrastructure and co-head of engineering. He joined Facebook in October 2009, more than 17 years ago, and rose to become one of the people behind the hardware, data centers and networks that Meta's apps run on.

He is not walking out the door today. He is expected to leave in early 2027, and he stays through the first months of the handover. A Meta spokesperson said: "Santosh will stay on for transition and he is moving into an advisory role on Meta Compute."

The catch: nobody has said why. His memo to staff did not give a reason, and Meta has not tied the move to any dispute. The memo reportedly spoke of the need to "pass the baton."

Brown joined Meta earlier this year as vice president of infrastructure. Before that he spent 19 years at Amazon, where he started in 2007 as one of the early engineers on EC2. That is Amazon's service for renting servers by the hour.

By the end, he was a senior vice president at Amazon Web Services, running its computing and machine-learning services. Those are the products that let other companies rent computing power for AI.

• 2007: joins Amazon as an early engineer on EC2.
• 2026: leaves AWS after nearly 19 years.
• Later in 2026: joins Meta as vice president of infrastructure.
• November 2026: named to take over as head of infrastructure.

When Brown was hired, reports said he would work under Janardhan. Now the reporting line flips: the newcomer runs the department his boss built.

Think of Meta's apps as a restaurant. The apps are the dining room. Infrastructure is the kitchen, the gas lines and the power grid, and it has to keep running while a few billion people order at once.

In practice that means three things:

• Data centers: giant warehouses full of computers.
• Networks: the cables and switches that link those computers together.
• Custom chips: processors Meta designs itself, so it does not have to buy every one from other companies.

Over the years, Janardhan's remit grew to cover all of that. In his memo he described infrastructure as having grown from a support function into a central competitive advantage for Meta.

Because AI turned the kitchen into the main attraction. Training and running AI models takes enormous amounts of computing power. Whoever can add that power fastest can ship new AI features fastest.

Meta's numbers show how big the kitchen has become. In its third-quarter results, the company raised its full-year 2026 capital spending forecast to between $135 billion and $145 billion. Capital spending means money put into long-lasting things like buildings, chips and power equipment.

By the numbers: Meta spent $34.91 billion on capital expenses in the third quarter alone. That is more than many large companies earn in a year.

In January 2026, Mark Zuckerberg announced Meta Compute, a top-level initiative to oversee the data centers and computing power behind Meta's AI plans. Janardhan led it, together with Daniel Gross, who joined Meta from the AI startup Safe Superintelligence.

Gross leads long-term capacity planning, supplier partnerships and business modeling. Dina Powell McCormick, Meta's president and vice chair, works on partnerships with governments and other large backers to build and finance big projects.

Zuckerberg said Meta is planning to build "tens of gigawatts this decade, and hundreds of gigawatts or more over time." A gigawatt is a billion watts, roughly what a large power plant produces. Meta has also said it plans to invest up to $600 billion in US infrastructure and jobs by 2028.

In real life When you ask Meta AI a question on your phone, the answer comes from a computer in one of these data centers. More of them means more people can ask at once without waiting.

Two stand out. Prometheus, in New Albany, Ohio, was planned to come online in 2026 as a 1 gigawatt "supercluster," a huge group of computers wired together to work as one machine.

Hyperion, in Louisiana's Richland Parish, is planned to start at 2 gigawatts by 2030 and could scale up to 5 gigawatts over several years. Meta has also signed nuclear power agreements to feed its data centers.

These are the kinds of projects the next head of infrastructure inherits: build them on time, keep them powered and fill them with chips.

The bill is large compared with what comes in. In the third quarter Meta reported $64.6 billion in revenue. It also had negative free cash flow of $2.12 billion, which means it spent more cash than its business generated in those three months.

Meta also guided to total expenses of $175 billion to $179 billion for 2026. Capital spending is the biggest swing factor in that number, so the person running infrastructure is effectively the person steering where much of it goes.

Why it matters: compute is now rationed like a scarce resource. Janardhan was also involved in deciding where Meta's computing power gets allocated, alongside Powell McCormick and Daniel Gross. Whoever takes over that seat decides which projects get machines first.

Reports from earlier this year said Brown's mandate at Meta could include scaling data centers and building out Meta Compute as a possible cloud service. A cloud service is one that rents computing power to outside customers, the way Amazon does.

Those reports were not confirmed by Meta, and nothing in this week's announcement mentions a cloud product. Still, hiring one of Amazon's most senior cloud engineers fits the idea that Meta wants to run its machines at the scale of a cloud provider.

Brown takes over the operational running of the division. Janardhan used his memo to say Brown has the experience and technical judgment to manage Meta's fast-growing computing capacity and help product teams ship new technology faster.

What's next: Janardhan's advisory role on Meta Compute keeps his knowledge close during the handover. Meta has not announced any other leadership changes, and the date of his final day has not been published beyond early 2027.

• Nothing changes in your apps today. Executive handovers do not change Facebook, Instagram or Meta AI overnight.
• Watch the spending. The size of the budget shows how seriously Meta takes AI features, and how much it needs to pay for them.
• Expect more AI features over time. More computing power is what lets Meta add them for billions of users.