Meta Gets Efficient, Even as it Ramps AI Compute Capabilities

For investors, the highlight of Meta’s fourth quarter 2022 earnings report was the improvement in profit margin. For others it was the new emphasis on “efficiency,” including better use of capital investment, flattening the organization and moving faster to cut projects that are not showing nearer term upside. 

“We expect capital expenditures to be in the range of $30-33 billion, lowered from our prior estimate of $34-37 billion,” said Susan Li, Meta CFO. “The reduced outlook reflects our updated plans for lower data center construction spend in 2023 as we shift to a new data center architecture that is more cost efficient and can support both AI and non-AI workloads.”

For some, that comment about “lower cost” data centers might have been interesting as well. 

One issue appears to be the difference between centers able to support artificial intelligence and all other workloads. 

Supercomputers might be one example of the change. Will we see more data centers that are optimized for AI workloads? And are such specialized facilities needed mostly to build the inference models? If so, that implies that the non-real-time training can be done at specialized facilities, while the actual applications run closer to end users and the edge. 

But Meta also is saying that the new data center architecture combines AI and standard workloads, which implies some new way of assigning functions and running the workloads. 

 Is a shift of compute towards the edge also part of the architectural shift? 

And some of the savings might come from a more modular approach to assets, which is not new, but might be operationally more important. 

“Along with the new data center architecture, we're going to optimize our approach to building data centers,” Li said. “So we have a new phased approach that allows us to build base plans with less initial capacity and less initial capital outlay, but then flex up future capacity quickly if needed.”