Meta Compute Takes Shape as Zuckerberg Poaches AWS Veteran Dave Brown

by Zaira Varghese 19. July 2026 AI 0

Meta Compute just went from boardroom whisper to a concrete plan with a marquee hire attached. Meta is bringing on Dave Brown, one of Amazon Web Services’ most senior computing executives, to accelerate its AI data center buildout and study whether it should rent spare capacity to outside customers, according to a Wall Street Journal report that lit up tech circles this week.

Brown spent nearly 19 years at Amazon, most recently running the AWS compute and machine-learning group behind EC2 — the backbone of the modern cloud. At Meta he will report to infrastructure chief Santosh Janardhan and is expected to start in the coming weeks, after staying at AWS through the end of July to hand off his duties. His reported mandate is to scale Meta’s data centers and help stand up a service internally referred to as Meta Compute.

Why Meta poached a top AWS executive

The timing is not subtle. Meta is pouring tens of billions of dollars into AI infrastructure to train its Llama models and power recommendation systems across Facebook, Instagram, and WhatsApp. Hiring the person who scaled the world’s largest public cloud sends a clear signal: Zuckerberg wants operators who have already built hyperscale systems, not just researchers. Brown’s departure was announced by Amazon days earlier, with a 27-year Microsoft veteran, Dave Treadwell, tapped to take over the AWS compute unit on August 1. You can read GeekWire’s full report on the move here.

What Meta Compute could actually become

The phrase Meta Compute hints at something bigger than internal plumbing. Zuckerberg told shareholders in May that companies regularly approach Meta to pay for access to its AI models or its spare computing capacity. Turning that inbound demand into a product would put Meta in the business of selling GPUs by the hour — the same model that made AWS Amazon’s profit engine. Meta has not committed to becoming a public cloud provider, but the hire suggests the option is very much on the table.

Rent idle GPU capacity to AI startups and enterprises between training runs.
Offer managed access to Llama and other Meta models as a hosted service.
Build a developer platform that competes directly with Bedrock and Azure AI.

Cutting the Nvidia cord

The buildout also dovetails with Meta’s push to reduce its dependence on Nvidia. The company recently moved its custom Iris inference chip toward production, and a homegrown cloud would give Meta more leverage over its silicon roadmap and its margins. Owning the full stack — chips, data centers, and a customer-facing service — is exactly the flywheel that turned Amazon, Microsoft, and Google into cloud giants.

A direct shot at AWS, Azure and Google Cloud

If Meta Compute launches as an external service, it would drop a fourth hyperscaler into a market that AWS, Microsoft Azure, and Google Cloud have carved up for a decade. Meta already operates one of the largest server fleets on the planet, so the raw capacity exists. The harder part is the boring, mission-critical work of billing, uptime guarantees, security, and enterprise support — precisely the muscle Brown spent two decades building at AWS.

What to watch next

For now, Meta Compute remains an internal ambition rather than a shipping product, and Meta has been careful not to over-promise. But poaching the executive who ran EC2 is not the kind of move a company makes to keep its infrastructure private forever. Watch for capital-expenditure guidance on Meta’s next earnings call and any hint of a developer preview — either would confirm that Zuckerberg intends to sell compute, not just consume it.