Google wants to be the home of your AI infrastructure investment, suggesting it can host the hardware and offer more services than if companies try and go it alone.
At its Google Cloud London Summit this week, the company set out its stall as a major player in the cloud AI space.
Though the company has often lagged behind the likes of Amazon Web Services and Microsoft Azure in terms of revenues, Google is pitching itself as the place to be when it comes to AI infrastructure investments.
Tara Brady, Google’s EMEA president, claimed it was a “fact” Google created generative AI – likely referring to the 2017 Google Research paper Attention Is All You Need, which introduced the transformer concept underpinning many of today’s genAI models.
Brady said that today 90 percent of AI unicorns (startups valued at more than $1 billion) use GCP, as do 60 percent of all funded AI startups – he said some 2 million developers are using the company’s generative AI tools.
But the company also pitched itself as the home of enterprise AI. This week has seen Google announce AI deals with the likes of Vodafone, online marketplace OnBuy, marketing firm Incubeta, BioCorteX, and UK retailer Dunelm. Pfizer, Hiscox, Toyota, and Monzo were also named on stage as AI customers.
Similar deals were announced with Warner Bros. Discovery in September. At its Cloud Next event in April, Google announced cloud AI deals with Mercedes-Benz, Bayer, WPP, IHG Hotels & Resorts, Best Buy, Orange, and PwC.
The event also saw the likes of Lloyds Bank, Bupa, and Screwfix owner Kingfisher take to the stage to share their stories.
Ranil Boteju, chief data and analytics offers at Lloyds Bank, said the company, which joined GCP around 2020, was using Google’s AI services for internal such as code translation and document scanning first before it looks to release such technology to the customer-facing side.
He said the company uses Google partly because it can have access to more models to suit the needed use case – as well as its own Gemini, Google offers access to models from the likes of Meta, Hugging Face, and others. The efficiency that the search company’s cloud offers in terms of cost and energy use – often known as FinOps – was also a lure, he said.
Mohsen Ghasempour, group AI director at Kingfisher, said the company uses all three major cloud providers, but relies 100 percent on Google for its AI needs.
Nik Sharma, co-founder and CEO at drug discovery firm BioCorteX, said that while the company is largely cloud agnostic, its Carbon Knowledge knowledge graph was hosted in Google Cloud. He said the speed and cost at which the firm could deploy new knowledge graphs was a major benefit.
AI driving the last data center refresh?
Unsurprisingly, Google wants you to use its AI hardware in the cloud.
The company went so far in one session as to say that the industry is now past the last cycle of enterprise data center refreshes. Reasons for this included access to AI hardware and services – along with the with the costs and energy demands. One presentation slide suggested even large enterprises were no longer ‘cloud first’ but aiming for ‘cloud only’. Various headlines about cloud repatriation may argue otherwise.
In his own keynote and subsequent interview with DCD, Google Cloud GM for infrastructure, Sachin Gupta, was less ebullient that enterprise data centers are no more and conceded AI will live in multiple places – just ideally on Google infrastructure.
“AI forces a decision; new infrastructure investment has to be made,” he said. “The best place is inside Google Cloud, but that’s not going to work for all customers.”
In lieu of every enterprise going cloud-only, the company pitched its on-premise hardware – known as Google Distributed Cloud – as a way to gain the benefits of AI hardware on-premise in a cloud-like way.
GDC shares the same interfaces as the search company’s public cloud and can bridge legacy on-premise systems with infrastructure in the cloud. The company also talked up its air-gapped version for customers who can’t or won’t expose systems to the Internet but still want some of the benefits of cloud and AI.
These on-premise offerings scale from single servers up to hundreds of racks, depending on the customer's need and use case. Multiple governments were said to be using the air-gapped version of GDC, while McDonald’s was said to be rolling out GDC servers to thousands of stores.
The company said its focus on tightly coupling its data center designs, IT hardware (including its TPU chips), and software offers much greater efficiency than companies might be able to achieve on their own.
Google does not currently offer its TPUs in any of its on-premises hardware; Gupta told DCD that Nvidia’s A100 and H100 hardware was “good enough” for current on-premise use cases such as translation, speech-to-text, and document search.
The issue of sovereignty in the context of AI was mentioned several times; when asked if data sovereignty demands were driving some of the company’s investments, he said it “absolutely” was. The company, like many of its rivals, is heavily reinvesting in markets where it has had a longstanding footprint, including the UK, with billions of dollars set to be pumped into new infrastructure in the coming years.
Whether in the cloud or on-premise, however, the message is clear: AI is driving the next wave of investment in data centers.