Apple’s first chip for servers is already underwayassures Information cite sources with direct knowledge of the matter. Designed specifically for AI processing, it is being developed under the codename ‘Barta’ and is expected to go into production in 2026.

The project would unite Apple with other major technology companies that have developed their own chips to support new AI services and their high resource consumption, as well as reduce the current dependence on the global giant of AI chips: NVIDIA.

Apple’s first chip for servers

Apple has been designing the chips it uses in its mobile devices for years, and recently overhauled the hardware architecture of its entire line of Mac personal computers, where it was able to replace x86 processors in three years. to Intel in favor of its own designs within the ARM-licensed RISC architecture. It is so far the only manufacturer in the world that has achieved real viability in building PCs with ARM.

Although servers and data centers have little in common with PCs, Apple has shown enough ability to design its own hardware engines. Additionally for this project partnered with Broadcom for development and with the TSMC foundry for mass production in advanced manufacturing processes, N3P, (3 nanometers).

Apple already announced this at its annual developer conference in June planned to use its own chips to power the artificial intelligence functions of its devices. So the news isn’t exactly new, although the connection to Broadcom, which soared (up more than 5%) on the stock market after the news broke, is.

It should be recalled that the push by major cloud service providers to diversify their supply chains has made Broadcom one of the biggest beneficiaries of the rise of generative artificial intelligence. Its shares have rallied 54% in 2024, after nearly doubling in value in 2023.

OpenAI and Google are other big players who are also working with Broadcom to develop this type of highly specialized chip where NVIDIA’s accelerators shine. All the tech giants want to reduce this dependency, but given the speed at which large AI models are being implemented, it doesn’t seem possible. its huge need for resources. Apple’s server chip will be fully focused on these types of tasks and is expected to enter mass production in 2026.