Thanks to the rise of AI, Google Cloud has an opportunity to differentiate itself from Microsoft Azure and AWS. While it has much to say about the technology du jour, Google highlights that it is innovating extensively in many other areas.
It is no secret that the Benelux (Belgium, Netherlands, Luxembourg) is a Microsoft region par excellence. Where AWS (still) has the largest global market share, Google Cloud (GCP) is in a clear third place. Yet the hyperscaler has been gaining a foothold locally for some time. With a €1.6 billion investment in data center expansion and deals with Proximus, Clarence and Lux Connect, Google’s commitment is significant. It has already led to 1,000 jobs in the Benelux region, with 600 to be added in Belgium. In Winschoten, Netherlands, Google is also putting €600 million into a new data center.
The narrative around infrastructure is clear: Google Cloud is here to stay in the Benelux. However, it’s ultimately about the product. In this case, it’s intended to be a full-fledged “Best-of-Suite” package that persuades companies to house everything from word processing to email infrastructure to AI development in one place. Joris Schoonis, Benelux Country Manager for Google Cloud, talks about more than 1,000 product advances within GCP and Workspace since October, when the previous Cloud Summit Benelux took place. Much of it is AI-related. But, “While we are often going to mention AI, it is important to deliver on promises now as well as in the near future,” Schoonis said.
AI from Google and within Workspace
Still, AI gets a separate segment during the Cloud Summit keynote. Floor Eigenhuis, AI Field Solutions Architect at Google Cloud Benelux, emphasizes that Google is rapidly innovating with its Gemini models. As such, it has already achieved stunning breakthroughs, such as a giant context window with Gemini 1.5 Pro. This allows Google’s AI model to consider a much larger amount of data to generate an output than, say, OpenAI’s GPT-4.
In addition to open-source and third-party options, it will soon expand its choice of models to include Gemma 2 and PaliGemma. Gemma 2 promises to perform as well as Meta’s recent Llama 3-70B with a model containing only half that amount of parameters, i.e. circa 35 billion. PaliGemma, meanwhile, is a VLM (Vision Language Model). It supports users with captions to images and short videos, can answer visual queries, recognize text in images, detect objects and segment them from the background.
In Workspace, Google’s AI integration is already well advanced. This is in line with what Microsoft has accomplished within the 365 suite via an army of Copilots. Whether it’s slideshows, word processing, or spreadsheets, an AI assistant assists users. Eigenhuis indicates that people are overwhelmingly positive about this assistant’s input. In fact, in Google Docs and Slides, most already approve AI suggestions more often than not.
John Stone, who holds the high-profile title of Chaos Coordinator Office of the CISO at Google, argues that AI is also excellent at protecting AI. The Google Secure AI Framework (SAIF) has been around for a year. It defines best practices around sensitive data, protecting data/models and AI deployment to detect fraud or cyber threats. For example, Google’s paid security offering explains what a file does before the user executes it. Meanwhile, Gemini in Security Operations can see if a suspicious file has occurred before. Gemini in Security Command Center goes a step further by explaining how a hacker’s attack path is altered when a particular security measure becomes active.
Starting the GenAI journey yourself
After controversial incidents surrounding faked ads, the generation of malicious a-historical AI images and the recommendation of glue on pizza, one may wonder if Google itself shouldn’t let go of the accelerator pedal a tiny bit and release their products a bit less haphazardly. Still, there is little to criticize about what Google Cloud offers its customers to start their GenAI adventures. Thanks to the comprehensive Responsible Generative AI Toolkit, they can also quite easily compare model performance when deployed for their business needs. Automatic Side-by-Side makes it easy to instantly see how two LLMs respond differently to the same input.
Eigenhuis sees that AI adoption among Google Cloud users is significant. The past year saw a 900 percent growth in the use of GPUs and TPUs in Google Cloud’s so-called AI Hypercomputer. Google has its own hardware ready in the form of Axion and Trillium, significantly reducing reliance on Nvidia GPUs. The increased efficiency of new chips simultaneously provides a much-needed boost for hitting sustainability targets, something that has only become more difficult with the rise of AI.
GenAI will continue to develop at lightning speed. Eigenhuis envisions a cloud of connected AI agents, intelligent entities all with a specific purpose, a concept which has enormous potential. In three steps, the Vertex AI Agent Builder should turn this potential into practical deployment. First, Gemini Pro provides the ability to have human conversations with a model. Next, the customer determines which topics are relevant to their own use case. Finally, “grounding” ensures that the AI agent has insight into verified data, such as medical data, academic literature or financial results.
Optimize existing processes
These are respectable AI ambitions, but impossible to achieve for many organizations in the short term. What can AI offer now? “You can also use AI to optimize existing processes”, Schoonis points out. He mentions practical examples such as targeted advertising with AI assistance and the addition of writing suggestions in e-mail messages. These are necessary to implement if a company has any desire to be competitive. All these 30-second time savers added up can lead to dozens of hours of extra productivity.
Schoonis does caution against a directionless search for some way to deploy AI. “Please start from the business strategy. AI is supportive.” According to him, Google Cloud should make it possible to get AI deployed easily, no more, no less. “You don’t have to be a big company to do that.” That’s why Google Cloud has already organized 1 million GenAI training sessions in order for this new discipline to become well established in the developer community. That same 1 million figure, by the way, is how many developers Google serves on its own various platforms.
Schoonis says AI must be done responsibly. “Using AI is not a competition. You have to know what you’re doing and how you’re going to do it.” Examples include selecting the right data to avoid bias or keeping sensitive data in-house.
For that reason, Google should not be able to access your data. The aforementioned partnerships with Proximus, Clarence and Lux Connect enable the so-called Sovereign Google Cloud. This offers nearly the same functionality as GCP’s public cloud. It comes with the added bonus of not having to fear that Google L.L.C. will be told by the U.S. government to make your data viewable to security agencies. Incidentally, Schoonis argues that this is mostly an emotional issue. With the right paid options, the data is well enough protected that no one, not even Google, can access it.
While the goal, according to Schoonis, the goal is to give Sovereign Google Cloud feature parity with GCP, but the latest features appear on the latter first. The distributed, air-gapped cloud is more suitable for smaller AI models, but Schoonis rarely recommends going in that direction. After all, it’s not even Google Cloud running the service in that instance. A “connected” version of the non-public cloud does talk to GCP, but does not offer the same data shielding as the air-gapped variant. Google Cloud almost always recommends the public cloud, but offers customers several security features. For example, there is the option of simply not giving Google the encryption key to access customer data. Washington won’t get to your data then, as it doesn’t have the key either and can’t just demand it from European companies.
Google Cloud as a product
The reality is that the Google Cloud Summit is overwhelmingly about AI. Still, we should take a brief look at the competitiveness of GCP versus Azure and AWS in a general sense. What does GCP have to offer that Azure, for example, does not? “We think choice is important for customers. We offer an open cloud. You can use a number of other clouds, you can integrate with us very easily. We see ourselves as a somewhat newer cloud.” A lot of customers appreciate that stance, Schoonis argues. “Customers who started with another cloud see a differentiator with GCP through data management, AI and security, among other things.” Over the past five years, Google Cloud’s global market share has increased fivefold partly for these reasons, he points out.
While Schoonis and his colleagues would obviously like a customer to make full use of Google Cloud, he acknowledges that this is often not realistic. AWS and Microsoft possess a certain”stickiness” that, across many years, has driven customers to a partial or complete lock-in. Schoonis emphasizes that the customer determines whether the various vendors’ setups make sense. Given that the market is gradually turning more toward GCP and Azure and bit by bit further away from AWS, change is clearly possible.
Also read: Revenue from Microsoft and Google cloud services grows significantly