Oracle has been watching competitors steal the artificial intelligence (AI) spotlight recently, but it too has been busy. The company has been building an AI portfolio that goes up and down the stack, from cloud infrastructure to middleware to applications.
Oracle makes a compelling case. But will the company catch up to competitors with far larger market share in the cloud and (notably Microsoft and Google) brighter reputations for AI innovation?
The Oracle AI platform starts with hardware: Oracle says it has been working for two years on building a cloud infrastructure with the power, speed and low cost for training large AI models.
Oracle Cloud Infrastructure supports technology that the tech company calls “Superclusters” — large clusters that can be devoted to a single training job. Superclusters support up to 16,000 NVIDIA Hopper GPUs, or 32,000 Ampere A100s, connected by a high-speed Remote Direct Memory Access (RDMA) networking, which enables exchanging information in main memory without relying on the processor, cache, or operating systems, for performance rivaling on-premises purpose-built supercomputers, Karan Batta, Oracle VP of product management, said in an interview.
Superclusters enable the performance needed for AI training. “You need a massive cluster to inject the data, run the deep learning job and output a model. Those require massive amounts of GPUs,” Batta said. “You’re not training a chunk of it. You’re training the entire data set. That’s why you need an infrastructure cluster to rival most supercomputers on the planet.”
Winning startup hearts
That infrastructure has attracted high-profile AI startups such as Adept AI Labs, CharacterAI and MosaicML (recently acquired by Databricks for $1.3 billion), which demand large infrastructure investments and need that infrastructure spending to be optimized, Batta said. More than 30 AI development companies have recently signed contracts to purchase more than $2 billion worth of capacity in Oracle’s cloud, Larry Ellison, Oracle chairman and CTO, said in the company’s June 12 earnings statement.
As for software: With Oracle’s recent investment in and partnership with Cohere, it’s building AI models delivering security, privacy and specialized information that serve industry needs, particularly in healthcare.
Enterprises can bring their own data to specialize models for their own industries, Batta said. That distinguishes Oracle from general-purpose conversational AI like ChatGPT, which is one-size-fits-all.
Batta declined to provide a timetable on when these industry-specific models will be available but said they’re a priority. “It’s a critical piece of our platform that we need to have available in some form very soon,” he said.
Oracle’s AI strategy is focused on privacy, security, data access and data control for enterprise customers rather than general-purpose use cases. “When we train base models with your data, we deploy the models in your account, your tenancy, your subscription, so only you have access to it and you can put controls on it,” Batta said. Dirty data is emerging as a problem for AI, as some companies limit the technology’s use over security concerns.
The company's strategy for industry-specific models is similar to IBM’s Watsonx strategy. But while Oracle declined to provide a timetable for making its models available, IBM said its models will be available next month.
Prioritizing healthcare
“Oracle already has two specialized large language models available, one for medical professionals and one for first responders,” Ellison said during the earnings call earlier this month. "Specialized large language models will be instrumental in helping highly trained professionals use their precious time more efficiently,” he added.
Healthcare is a priority for Oracle, driving its biggest-ever acquisition of electronic medical records firm Cerner for $28.3 billion last year. At the time of the acquisition, Ellison said the purchase was motivated by nothing less than a drive to completely transform the U.S.’s broken healthcare system.
However, that merger has been troubled. The company has been struggling with a contract with the U.S. Department of Veterans Affairs to revamp that agency’s healthcare records, and Cerner laid off hundreds of employees, rescinded job offers and cut back open positions, according to reports in early June.
Additionally, Oracle is building AI capabilities into its own applications, including supply chain and retail, where the vendor plans to inject intelligent assistants to take action automatically or make recommendations.
And historically, the company’s strategic Autonomous Database and Autonomous Linux owe their automatic, low-cost, low-maintenance operations to artificial intelligence, the company said.
Strong potential
Oracle has a tiny share of the cloud infrastructure market. Amazon leads with 32% market share, followed by Microsoft (23%) and Google (10%), according to an April report from Synergy Research Group. Oracle barely gets a namecheck in that report, cited among the “Next 20 Companies,” which have a combined 26% market share. However, Synergy does mention Oracle as among those with the highest year-on-year growth rates (without naming specifics).
Oracle shows strong potential in the AI market, Roy Illsley, Omdia chief analyst, said. Specific AI solutions tailored for company data will give organizations competitive advantage, requiring processing power, software and access to foundational models. All hyperscalers and OEMs are competing in this area; winners will be determined by who has the best models, easiest-to-modify software and computation power delivered in a single package, which will require partnerships, he said.
The embedded AI market, with AI included in domain-specific tools and solutions, is more mature and will require company-specific data, Illsley said.
Oracle has been late to the cloud market, but that could work to its advantage, as it’s had an opportunity to learn from competitors’ experience, Illsley said. And he’s not the only one who thinks that.
Oracle faces competition in the enterprise AI Market. Amazon Bedrock and Google Cloud’s Vertex AI take a similar approach, while Azure ML uses Snowflake data warehousing.
Global spending on AI software, hardware and services will likely top $150 billion in 2023 and exceed $300 billion in 2026, IDC predicted. The top AI solutions enterprises are looking to adopt in the next two years include AIOps, augmented intelligence, discovery and analysis, intelligent task and process automation and preventative maintenance, according to a report from IDC.