According to a report from Reuters this morning, Meta Platforms, Inc. announced yesterday its plan to construct a major data center in central Alberta, Canada. This facility will be the company's first in the country and is part of a rapid expansion of computing capacity to support the global artificial intelligence boom.

The company stated this 1GW data center, located in Sturgeon County, can be scaled up to 1.8GW. The total investment is estimated at 13 billion Canadian dollars, approximately 92 to 100 billion US dollars depending on the exchange rate. Its power will be sourced from natural gas generation, with future supplementation from clean energy.

This development provides clear answers to two prevalent market concerns.

Is Meta Halting Data Center Construction?

Is Meta Preparing to Exit the AI Race?

Today's Reuters report delivers a distinctly different and very clear signal: Meta Platforms, Inc. is not slowing down its AI infrastructure build-out; instead, it continues to expand its global network of AI data centers.

In fact, over just the past week or so, Meta Platforms, Inc. has sent multiple signals. It appointed renowned scholar Dawn Song to lead AI Research at its Superintelligence Labs. If the company intended to exit the AI competition, why would it continue recruiting top-tier academic leaders? Now, it is adding a new $10 billion-scale data center in Alberta, Canada.

This contradicts the market logic that previously inferred "Meta will stop AI investment" from speculation about "Meta potentially leasing out computing power." A more reasonable explanation is therefore:

If a "Meta Compute" service for leasing some computing capacity is eventually launched, it is more likely a business model aimed at improving GPU utilization—by renting out H100 GPUs purchased during the metaverse era—and increasing the return on invested capital, rather than a move to cut AI capital expenditures.

Notably, H100 rental prices have been rising recently. The price decline in 2024 was due to a significant increase in H100 shipments, shifting supply from "severe shortage" to "relative abundance." The current price increase suggests demand remains robust. Despite the H100 being a computing card launched four years ago and far less efficient than newer models, the growing demand, likely from rapidly expanding inference workloads, indicates a thriving market.