Popular imagination places artificial intelligence in chatbots and digital assistants. But the next story of AI is taking shape in the data centres and the databases that hold the world’s knowledge. The announcements at Oracle AI World in Las Vegas this week offered little that dazzled the eye, but plenty that reveals the shape of the next phase of the AI era.
The headline grabber was the unveiling of what Oracle calls a zettascale cluster: a cloud-based supercomputer built from tens of thousands of graphics processors wired together to perform calculations at extraordinary speed. Instead of a government or global research institute spending billions to build such a machine, a company can now rent a slice of it by the hour.
This is the scale of power needed to train vast AI models like ChatGPT. Most businesses will never do that themselves, but the fact that such raw capacity is for hire changes who can compete at the top table of AI.
Just as revealing was the focus on where AI meets data. Oracle’s latest version of its database, called AI Database 26ai, introduces features like vector search, a way of finding information by meaning rather than by keywords. Alongside it Autonomous AI Lakehouse combines the flexibility of a data lake, where information of any type can be stored, with the structure of a warehouse, where it can be organised and queried.
The significance is that AI can now work directly with the information companies already store, rather than requiring separate, complex systems to be bolted on. For banks, hospitals or retailers, that makes the difference between experimental AI and AI that becomes part of everyday operations.
AI World also suggested a new realism about the multicloud world. Until recently, cloud providers acted as if customers would give them all their data and applications and never look back. That dream has crumbled. Enterprises spread workloads across several platforms to avoid being locked in or caught short when capacity runs out.
Oracle’s decision to run its databases inside AWS, Azure and Google Cloud is an acknowledgement of this shift. Even the launch of multicloud credits, which can be spent across providers, signals that financial flexibility now ranks alongside technical performance as a top priority.
Hardware strategy is also shifting fast, as OpenAI’s recent deal to buy AMD chips has already shown. Oracle joined the club this week, and announced an expanded partnership with AMD, committing to tens of thousands of its new AI processors. Until now, the AI industry has revolved around NVIDIA, whose chips dominate the training of large models. With demand outstripping supply and prices climbing, big cloud players are hedging their bets. For customers, this won’t mean instant price cuts, but it does widen the ecosystem. It hints at a future where several chip makers compete to power AI, each optimising for different workloads.
Then there is the question of adoption. Building AI at industrial scale is one thing; making it usable by businesses without armies of data scientists is another. Oracle’s AI Factory and assistants inside business applications are an effort to address this challenge. The AI Factory is essentially a set of prebuilt templates and tools to help organisations deploy AI faster. The assistants are built into existing software so that employees can query and analyse data in natural language. These may sound like a sideshow to zettascale clusters, but in practice could be more important. Without tools that ordinary staff can use, all the compute power in the world remains locked away.
Taken together, these announcements mark four shifts in the direction of AI:
- First, scale is being commoditised, as supercomputer-class systems move into the rental market.
- Second, data gravity is pulling AI closer to the heart of enterprise databases, cutting out the friction of moving information into separate systems.
- Third, multicloud pragmatism is replacing lock-in, allowing businesses to hedge their risks and balance costs.
- And fourth, usability is becoming the deciding factor in whether AI finds its way into daily business practice.
For South African organisations, the deciding factor in whether local firms thrive in the AI age will not be driven by algorithms, but by choices about infrastructure. Where data is stored, which cloud providers are trusted, and how governance is built into systems will outweigh the popularity of chatbots.
Oracle’s announcements should not be read as one company’s triumphs, but as signposts for where the industry is going. AI is moving away from surface froth and into the scaffolding underneath. The breakthroughs of today are about giving humans access to the machinery that makes such thinking possible at scale.
That said, none of the big announcements pointed to African organisations already using these tools. That silence highlights the gap between the global marketing pitch and the reality on the ground here: infrastructure is uneven, and many enterprises are still wrestling with the basics of data readiness.
* Arthur Goldstuck is CEO of World Wide Worx, editor-in-chief of Gadget.co.za, and author of “The Hitchhiker’s Guide to AI – The African Edge”.