Eager to establish dominance in artificial intelligence, Meta has signed up to use cloud-computing services from Google Cloud in a deal worth at least $10 billion over six years, according to a report from Bloomberg.
The deal would expand Meta’s capabilities as it continues to push AI tools and services across its platforms, including Facebook, Instagram and WhatsApp. AI requires an enormous amount of computing resources, demanding more data center bandwidth. Those data centers need an increasing amount of electricity and water to run and cool the hardware inside.
Meta is also spending big to build its own AI data centers, including a 4-million-square-foot facility in Louisiana called Hyperion, which it plans to open by 2030.
Meta and Google declined to comment on the specifics of the deal or confirm that an agreement has been made.
Meta’s big AI push
In an effort to hire a superstar team to work on its AI projects, including its Llama AI platform, Meta has been on a hiring spree to the tune of hundreds of millions of dollars. Recent reports, however, suggest that AI-related hiring may be paused as the company reevaluates its strategy.
Regardless, Meta continues to roll out new AI tools and features, including AI translations for Facebook and Instagram.
Meanwhile, a deal with Meta could boost Google Cloud’s status as what Bloomberg describes as a “one-stop shop” for AI services. In June, OpenAI joined forces with Google Cloud to use its data centers.
Meta has its own data center build-outs and has even invested in nuclear power to provide energy for its data goals, but its needs appear to be growing faster than it can keep up.
“We are seeing across the industry that AI workloads are creating elevated power and compute capacity requirements that are outpacing even the most aggressive data center buildout timelines,” according to Chuck Marvin, chief commercial officer at Thunderhead Energy Solutions, which works closely with data centers.
“Meta is one of the most sophisticated data center operators in the world, but the scale and speed of AI deployment today means that even the hyperscalers are looking for creative solutions to bridge capacity gaps,” Marvin told . “Rather than signaling any weakness, this deal shows Meta is being strategic about meeting immediate AI infrastructure needs while [its] own large campus projects come online.”
Though AI workloads are part of why the data center business is booming, Marvin said that’s not the whole story.
“Yes, AI workloads are very power-intensive, particularly training large language models and running inference at scale, but that’s happening simultaneously with significant growth in enterprise cloud adoption, edge computing deployment, and digital transformation initiatives that were already straining data center capacity,” Marvin said.
AI uses more power-hungry GPU clusters, but other cloud services, like video streaming, are also growing globally. Those demands are converging and pushing utility grids to their capacity.
“The result is that data center operators are scrambling to secure power and capacity across all their use cases, not just AI,” he said.
What’s next for Google Cloud?
While AI isn’t the only factor in data center growth, it’s an important one. Google Cloud is in a particularly good position to provide those types of services, Marvin said.
Specifically, the company designed its own TPU chips, and has deep AI expertise from running search and YouTube at scale. It also built its infrastructure specifically for the massive data movements required by AI, Marvin said. “Google Cloud has been gaining ground by focusing on [its] core strengths in AI and machine learning… offering enterprise customers access to the same cutting-edge infrastructure that powers Google’s own services,” he said.
There’s enough demand for multiple winners, including competitors such as Microsoft and Amazon, but Google has done well by creating an AI-first infrastructure. “For AI-intensive workloads specifically, Google offers technical advantages that are tough to compete with,” Marvin said.
