Apple spent years building chips for iPhones and Macs. Now it wants to take that advantage into AI servers and private computing. The move could push Apple beyond consumer devices and into the infrastructure running AI.
Apple widens its chip play
The new iPhone 18 Pro shows how far Apple’s chip strategy has come. Its A20 Pro chip includes a larger system built specifically for AI.
GeekBench results cited by Barron’s show the A20 Pro beating leading chips from Intel Corp., Advanced Micro Devices Inc., and Qualcomm Inc. in single-core performance. That speed matters for many everyday tasks performed on a device.
Apple’s edge comes from controlling the full system. Its chips are designed alongside the software, battery, cameras, security features, and services that depend on them.
That control also lets Apple run more AI features directly on its devices. Barron’s reported that the A20 Pro’s neural engine has doubled in size from the previous generation.
Processing requests on a device can reduce delays and keep more personal data private. It also means fewer everyday AI tasks need to be sent to a distant data center.
Servers enter the roadmap
Apple could now bring the same chip strategy into the server market. Bloomberg reported that the company is developing an enterprise AI server for developers, governments, and businesses.
The Information said Apple is considering two versions. One would use two planned M8 Ultra chips, while the larger system would use four.
The server is not expected before 2029, according to Reuters. The project could still be canceled or move forward without technology from Nvidia Corp..
A launch would mark Apple’s return to dedicated server hardware. The company discontinued its Xserve line in 2011.
Apple is targeting inference rather than model training. Inference is the work that happens when a finished AI model responds to a request. Demand for it rises as more AI tools reach phones, computers, apps, and workplaces.
Nvidia still owns the wiring
Apple may build the processors, but connecting them is another challenge. Large AI systems need chips to exchange data quickly as they work through the same task.
The Information reported that Apple has considered using Nvidia’s NVLink Fusion. The technology combines switches, chiplets, and software that connect processors inside data centers.
That reach makes Nvidia difficult to avoid. Even companies building their own AI chips may still need its networking technology to turn them into larger systems.
Networking equipment can account for 10% to 15% of AI data center hardware costs. It also generated one-fifth of Nvidia’s data center revenue in the April quarter.
That keeps Nvidia in the picture even when customers design their own processors. Greater control over chips does not remove the need for the technology connecting them.
The rotation is changing shape
The AI hardware trade is starting to move beyond the biggest data center names. The iPhone maker has recently traded more like an alternative to semiconductor stocks than part of the same bet.
Shares moved in the opposite direction from the iShares Semiconductor ETF on 19 of the 26 trading days since Aug. 11. Through Wednesday, Apple had gained 8%, while the chip ETF had fallen 5%.
The split does not turn the company into a pure AI infrastructure play. It shows that investors are starting to treat device-based AI and data center spending as separate trades.
With roughly $112B in annual net income, the company has the earnings power to fund years of chip and server development. Its valuation of around 38 times earnings leaves little room for delays.
Nvidia remains central because its networking systems are already built into AI data centers. The iPhone maker is taking another route by using private AI across its devices and servers to strengthen its ecosystem and reduce cloud dependence.
Private Cloud Compute already handles requests that are too demanding to run directly on iPhones and Macs. The planned enterprise server would take that strategy further.
The timeline remains the biggest risk. A launch targeted for 2029 asks investors to wait years before the product can generate meaningful revenue.
