In my experience, large enterprises that have embraced multiple public clouds will typically use Google Cloud for analytics or any service related to “data.” The recent Google Cloud AI and Data Summit, which I have attended for years allows us to understand what is new in generative AI, databases, data analytics and business intelligence.
It is still early days yet
One interesting but not surprising takeaway from the summit is that we are still in the early days of the data revolution. According to Google, most organizations are investing in data and AI but are struggling with only twenty percent succeeding with models in production. This may seem odd with all the ways we experience AI, but more times than not, these are the largest technology companies with the most resources with research groups.
The companies that are succeeding span all industries and scenarios, from back-office supply chain innovation to new customer experience design in the front office. The common thread is that organizations are investing in machine learning (ML) and AI to speed up the innovation cycle.
That said, Google noted a recent dramatic increase in ML predictions (output of an algorithm trained on a historical dataset) and ML evaluations (different evaluation metrics to understand a machine learning model's performance)—perhaps a precursor for more companies succeeding with models in production.
BigQuery gets a makeover
BigQuery is an enterprise data warehouse with integrated BI, machine learning and AI for real-time analytics, part of the Google Cloud Platform (GCP) suite. With BigQuery, users can run complex queries on large volumes of data in seconds. I have written several articles tracking the evolution of BigQuery.
Google said it received customer requests to have the ability to both optimize stable and predictable data workloads and more efficiently address unpredictable data workloads. The result is BigQuery autoscaling and compressed storage. This makes a ton of sense, particularly as customers use a lot more of a service they get better at know exactly what they want and want it their way.
Autoscaling will dynamically adjust capacity to usage demands, delivering 40% gains in compute efficiency over the current fixed-capacity offering, which is essentially pay-for-use. Autoscaling creates real-time optimal capacity allocation for each workload. No more need to plan for over- and under-provisioning: resources will be there when needed.
The second new feature in BigQuery is the availability of a compressed storage option. Resounding feedback from customers indicated the need for more storage at a lower cost. In response, Google has now made a multistage compression model available within BigQuery to achieve a 30-to-1 compression rate. Customers now have the option to pay for highly-compressed data storage.
Also new: BigQuery editions with three pricing tiers—Standard, Enterprise and Enterprise Plus—that customers can mix and match for the correct price-for-performance based on individual workload needs. For example, the Standard edition is best for ad-hoc, development and test workloads, while Enterprise has increased security, governance, machine learning and data management features. Enterprise Plus is for mission-critical workloads that demand high uptime, availability and recovery requirements or that have complex regulatory needs.
As we have seen with any product or service over time, if it is successful, it segments. Whether it’s smartphones, PCs, cars, Three Bears beds or data warehouses, over time, they will segment to better meet the needs of narrower customer sets.
Generative AI is a hot topic
Generative AI refers to AI algorithms that use training data to generate or create an output, such as text, photo, video, code, data or 3-D renderings. Generative AI creates content, whereas traditional AI performs functions such as analyzing data or controlling a self-driving car. The most famous generative AI program recently is OpenAI's ChatGPT.
The basis of generative AI is a foundation model. There are three attributes of a foundation model. First it multitasks rather than single-tasks, performing various predictive and generative tasks right out of the box. Second, it is generative and multimodal, generating high-quality content, ranging from text to images to speech to code, using natural language prompts. Third, it requires zero or minimal training to tune for custom tasks with low amounts of task-specific data.
While foundation models are compelling and capable, Google believes that to build enterprise-grade generative AI applications, enterprises need more than just foundation models. By combining the power of the foundation model, enterprise search and conversational AI, enterprises can define how generative AI will apply to applications and customer experiences.
For example, every enterprise has a sizable internal data repository and will want to generate specific responses based on that repository rather than on general knowledge from the foundation model. Enterprise search should therefore provide much more targeted and relevant results in this example.
Recently Google made two announcements to support generative AI. The first is extending Vertex AI, its end-to-end unified platform for AI and ML. Generative AI support for Vertex AI now includes APIs to access many foundation models as well as APIs to build applications or services. Data scientists have a full suite of tuning options to optimize foundation models for specific usage, incorporating data to train the model further.
Meanwhile, Gen App Builder is a new development platform for developers, combining enterprise search, conversational AI and the foundation model to enable data scientists to develop new generative APIs for specific use cases.
Moor Insights & Strategy Principal Analyst for AI and Quantum, Paul Smith-Goodson, wrote an in-depth piece on what Google is doing in AI. You can find it here.
What came across loud and clear at this event was that Google is very focused on advancing products and technology while at the same time enabling customers to be competitive. I believe another reason customers choose Google Cloud is the company’s commitment to being open. I think that is critical as companies embrace multi-cloud and hybrid cloud deployments. Enterprises want hybrid, multi-cloud solutions. As companies continue to innovate, Google ensures customers can deploy when and where needed—something that now extends to AI capabilities.
For me, the BigQuery segmentation into three different versions indicates a maturation of Google Cloud and a nod to its success. Every time you see a company do segmentation, it means the business has become big enough that “one size fits all” is no longer valid. Adding the flexibility of three different versions meets customers' demands while allowing Google to optimize its offering.
Even though it is still early in the data revolution, Google quoted some impressive numbers, with over ten million monthly active BI users on the platform. Google Cloud also has hundreds of BigQuery customers that have data warehouses larger than one petabyte. BigQuery customers analyze over 110 terabytes of data per second, equivalent to scanning the Library of Congress four times in a second. While all of this is impressive, it does raise the question of what will happen when the other 80% of companies start being successful with their data deployments!