Talk of the end of Intel's GPU business has been around for a long time, almost since before the debut of the first generation of Arc cards based on the Alchemist architecture. At the time, it was widely believed that Intel had invested heavily in the discrete GPU market with limited success, and that the end result would be branding cuts, program shrinkage, and team disbanding. Over the past two years, Intel has indeed conducted several rounds of reorganization of the relevant organizations, but this is not the same as withdrawing from the GPU track. On the contrary, recent signals indicate that the product line is being redefined, not abandoned.

At the second annual Artificial Intelligence Summit Chen Liwu gave a clear and positive answer to the question of whether Intel will continue to produce GPUs in the future. The occasion itself dictated that the focus of the discussion was on AI computing, and such a statement could easily be interpreted to mean that Intel is still hoping to make a renewed push in the field of AI gas pedals. But what's really worth noting is the information he added: Intel has hired a new chief GPU architect and is explicitly treating GPUs as equal in importance to CPUs, and is no longer limiting architectural choices to Xeon or x86, emphasizing more flexible design tradeoffs for different workloads.

The new chief GPU architect is none other than Eric Demers The new chief GPU architect is Eric Demers. The name is not unfamiliar to those who have been in the hardware industry for a few years. He was formerly a member of ATI Technologies He was an important technical backbone of ATI Technologies, and participated in leading the development of the Radeon 9700 Pro. He was involved in the development of the Radeon 9700 Pro, which changed the landscape of high-end graphics cards with its architecture and performance that first suppressed NVIDIA, and he continued to be involved in GPU design even after ATI merged with AMD. After ATI merged with AMD, he continued to work on GPU design, and when ATI sold its Imageon business to Qualcomm, Demers moved to Qualcomm and was involved in the early development of the Adreno graphics architecture, the name "Adreno" being a direct homage to Radeon.

Demers joined Intel at this time, releasing a fairly clear signal. Intel's current presence in the AI gas pedal field is not strong, especially in the software ecosystem and system-level integration, the gap with NVIDIA is obvious. Bringing in a technical leader who has long been involved in GPU architecture design and has worked across PC and mobile graphics is more like laying the groundwork for the next phase of GPU products than tinkering with the existing Arc roadmap. Realistically, he's more likely to be involved in the AI compute line than directly leading gaming graphics, but that doesn't mean the two routes are completely separate; NVIDIA has already demonstrated that architectures designed for AI compute can be decentralized and adapted to consumer GPUs as well.

More relevantly, Lip-Bu Tan repeatedly mentioned Intel's manufacturing strategy on the same occasion. He made it clear that the company will continue to produce both CPUs and GPUs, and rely on its own fabs to complete manufacturing, while opening up foundry capabilities to the outside world. Intel is currently focusing on the 14A process, and while that node won't be ready for mass production until 2028, the PDK is already planned to be open to external customers before then. This means that GPUs are no longer just a matter of product planning, but are directly integrated into the long-term layout of Intel's foundry system.

When pressed to ask "whether Intel will really produce GPUs in its own factories", Chen Liwu gave a response that did not leave room for maneuver: Intel will produce CPUs and GPUs, and the foundry will not only serve its own products, but will also be open to other manufacturers. This stance, the GPU clearly put into the core list of Intel's future process and capacity planning, and is no longer a dispensable exploration project.

The remaining uncertainty is instead focused on the consumer market. The tipster Jaykihn mentioned that the Arc Pro B70 is on the way, and it's likely that this type of product will come with up to 32GB of video memory, and will be more oriented towards professional and workstation positioning rather than mainstream gamers. Whether or not Intel puts enough competitive GPU resources back into the gaming market in the short term will still depend on cost, memory availability, and overall product pacing. But at least from the current information, Intel GPU business is not towards the end of the line, but is being pulled back to a more long-term, more systematic track.