Google is ramping up its investment in cloud and artificial intelligence (AI) infrastructure with plans to allocate $75 billion toward expanding capacity in 2025. The increased spending will primarily focus on enhancing technical infrastructure, including cloud servers and data centers.
During the company’s Q4 2024 earnings report, CEO Sundar Pichai confirmed the investment, which marks a sharp increase from the $52.5 billion Google spent on capital expenditures last year.
Ashkenazi emphasized that much of the funding will address ongoing capacity constraints, ensuring the company can meet rising demand for its AI and cloud services.
Google Cloud is also continuing its upward trajectory, reporting a 30% year-over-year revenue increase in the fourth quarter, reaching $12 billion. For the full year, Google Cloud’s revenue rose 31% to $43.4 billion.
Looking ahead, the tech giant said it expects to allocate between $16 billion and $18 billion in capital expenditures in the first quarter of 2025, with a focus on expanding its data center footprint and AI computing capabilities.
Scott Wheeler, cloud practice lead at Asperitas, said the billions of dollars spent on AI infrastructure create an ever-increasing barrier for competitors who want to produce large language models (LLMs) for public use.
“We can see the financial value of this barrier when DeepSeek came out with a more efficient AI inference engine that resulted in an immediate market value reduction of $1 trillion in AI-related companies,” he said.
He added DeepSeek has shown that there is still a lot to squeeze out of AI models’ training and running, and all major AI firms are putting enormous effort into model efficiencies.
“These efficiencies will reduce the pressure on the physical infrastructure demands, although they will not remove all the pressure,” he said.
Wheeler added the good news is that the industry has already been through a massive data center expansion with the rise of the public cloud where both AWS and Azure grew by more than 50% from 2014 to 2018. “This is a path that has been travelled before,” he said.
John Dinsdale, chief analyst at Synergy Research Group, explained these investments represent not so much a radical change but a continuation of what has been happening over recent years.
“The numbers have got a lot bigger, but the story is essentially the same – not only are they investing to capture huge new revenue streams, but by default they are also constructing huge moats that act as a barrier to new entrants challenging their market position,” he said. “This is a game of scale. Serious scale.”
He added the capacity constraints are really around two issues: Availability of power to fuel the data centers and availability of new GPUs in sufficient volume to support generative AI (GenAI) services.
“The power issue is about having the right amount of power available at the desired geographic location,” Dinsdale said. “There are ways around this and the hyperscale operators are managing this reasonably efficiently.”
He predicted GPU supply will catch up with demand, and in the meantime, he noted GPU driven services are growing like gangbusters. “While there are constraints, it is hardly preventing the market from taking off,” Dinsdale said.
Wheeler explained innovations will drive AI to the most efficient and cost-effective platforms, which may not be the specialized TPU and GPU hardware today.
“One example is DeepSeek’s inference performance, which could perform better on the unified memory architecture of Apple CPUs than on a GPU,” he said.
Meanwhile, chip development will continue to be done by hyperscalers, which started in 2015 when AWS acquired Annapurna Labs.
This purchase led to the development of AWS Graviton, Inferentia, and Trainium chips in 2018, 2019, and 2020 respectively.
“Given how volatile the AI market is, there is tremendous risk in placing a single AI vendor bet,” Wheeler said. “The AI market is highly volatile, as we have seen with the disruption caused by DeepSeek’s latest release, and this will not be the last market-shaking innovation we will see.”
