Developing artificial intelligence (AI) models requires a substantial amount of computing power, which most personal computers and devices can't provide. Therefore, most of these workloads are processed inside large data centers that are filled with thousands of advanced chips from suppliers like Nvidia (NVDA 2.17%), Advanced Micro Devices (AMD 0.79%), Broadcom (AVGO 2.66%), and Micron Technology (MU +0.59%).
Oracle's (ORCL 2.05%) cloud data centers are among the fastest and most cost-efficient in the world, thanks to the company's recent investments in automation and innovative networking technologies. As a result, leading AI software developers like OpenAI, Meta Platforms, and Elon Musk's xAI are lining up to use them, which is driving a surge in revenue for Oracle's cloud infrastructure business.
On Wednesday, Oracle released its operating results for its fiscal 2026 first quarter (which ended on Aug. 31). According to one key number in the report, the company is going to need significantly more data centers, which could be fantastic news for chipmakers.
Image source: Getty Images.
Oracle's order backlog just skyrocketed
Developers can tap into remarkable amounts of processing power through Oracle's cloud data centers. Its superclusters allow them to scale their workloads up to 131,072 of Nvidia's advanced Blackwell graphics processing units (GPUs) -- enough computing muscle to train and deploy the industry's most powerful AI models. Oracle is also building new GPU superclusters that feature 131,072 of AMD's latest MI355X processors, which will give developers additional options.
Oracle's data centers use a proprietary random direct memory access (RDMA) networking technology that moves data between chips and devices faster than traditional Ethernet networks. Since developers usually pay for computing capacity by the minute, more rapid data transfer can lead to substantial cost savings.
Scalability is another major cost saver. Oracle has invested heavily in automation that reduces the need for human workers, so its data centers are incredibly cheap to run, and it passes some of those saving on to its customers. Plus, since staffing requirements per location are minimal, the company can bring more infrastructure online at a rapid pace.
Therefore, even though Oracle's data centers use the same chips from Nvidia and AMD as those managed by other cloud infrastructure providers, Oracle Chairman Larry Ellison says his company's data centers are faster and far more cost-efficient than the competition's.
During the fiscal 2026 first quarter, Oracle's cloud infrastructure unit generated a record $3.3 billion in revenue, up 55% from the year-ago period. However, the headline-grabbing number was the company's remaining performance obligation (RPO), which soared by an eye-popping 359% year over year to $455 billion. RPO reflects the value of services that customers have signed firm contracts for, but that have not yet been delivered, so the metric can be viewed similarly to an order backlog.
NYSE: ORCL
Key Data Points
What it means for Nvidia, AMD, Broadcom, and Micron
Demand for Oracle's data center capacity has outstripped supply for the last couple of years, but never by this magnitude. Oracle will have to build new cloud infrastructure at a rapid pace to convert its enormous RPO into revenue. That will require it to make substantial capital expenditures, which will mean billions of dollars flowing to companies such as Nvidia, AMD, Broadcom, and Micron.
In fact, in a conference call with investors last week, Oracle CEO Safra Catz said her company's capex could top $35 billion during fiscal 2026. That was up substantially from the forecast of $25 billion that Oracle issued just three months earlier, highlighting how quickly demand is ramping up.
Nvidia and AMD are two of the world's top suppliers of data center GPUs, which are the primary chips used to provide parallel processing power for AI development. Therefore, they will be among the biggest beneficiaries of Oracle's exploding capital budget. But Broadcom's AI accelerators -- more specialized types of parallel processors -- are rapidly growing in popularity because they can be customized to suit the needs of each customer, so they offer more flexibility than GPUs.
Broadcom has at least three hyperscaler customers that each plan to deploy up to 1 million AI accelerators by 2027, creating a $90 billion market opportunity for the company. There is no indication that Oracle is one of those customers, but the two do collaborate on certain enterprise hardware solutions for data centers already.
Micron, on the other hand, supplies high-bandwidth memory solutions that get directly embedded into GPUs, including Nvidia's Blackwell and Blackwell Ultra lines. Therefore, it too will be a direct beneficiary of Oracle's rapidly increasing capital expenditures.
