Their consumer GPU segment def does look a bit shaky at the moment. The Arc gen 2 discrete GPUs will only compete current-gen to current-gen with mid tier Nvidia and AMD chips. However, gen 3 is on an accelerated release schedule, and will likely be arriving late next year (supposedly, with capabilities that will bring it closer to the top tier chips).
I think where Intel really shows promise for GPUs and accelerators is in the price-point area.
The Gaudi 3 accelerators that entered general availability two months ago manage to outdo Nvidia's hopper chips when comparing precision-to-precision, and retail for half as much. The overall success of that line really depends on whether work to overcome the lock in of CUDA will pan out sufficiently. Personally, I think it will, as AMD, Google, Amazon, and others are all making (and successfully rolling out) chips which depend on breaking the industry out of the walled garden of CUDA via direct integration with the most common higher level tools used by ML engineers.
Given the availability bottleneck for Blackwell, there is a growing market for alternatives.
And, again, half the price of Hopper chips, for better performance, for a product that's already on the market, is a pretty attractive proposition that will certainly see more orgs coming through the door. Might need to put one of these $14k 128GB intel hopper-peers in my workstation...