The article is unclear so I assume they're fabless, exploring designing and owning the IPR but the usual fab lines will actually make them.
Which means they compete for bandwidth in that generations stepping, and all of TSMCs 3nm is taken by Apple right now and NVidia is also fabless and depends on .. TSMC so this is an interesting "where in the queue are you" problem.
Up to a point, TSMC could replicate or increase the throughput of its fabs if demand increased systematically, which is plausible as ASIC designs become more common. Solving the current problems of a company like OpenAI requires only a very small fraction of time on the fab compared to the time Apple needs to produce a billion chips for phones, iPads, laptops, or the time that NVidia needs for printing GPUs, so it probably these small additions by newcomers do not yet warrant rapid expansion of the latest fabs of TSMC. I hope that TSMC are planning for it anyways.
They are definitely fabless. But in reality, fabs are not loaded by 100%. Usual practice, fabs loaded to 60-90% depend on local things, and due to incompatibilities.
Slightly more details. Usual fab cannot just take masks and copy them to other fab.
Rumors said, Intel achieved so high level of fabs compatibility, that they could manufacture any chip on any their fab.
Other manufacturers work on iterative methodology, so first 1-3 batches used to test and improve masks and tune technology, and at next batches give good product output (moderate percent of defects).
Other important thing, different fabs of even one manufacturer, usually have different tools, and could do different resolution. So in reality, when we hear about chips deficit, it really means, that some ONE resolution used by 90% of clients is overloaded.
So, if fabless manufacturer is not tightly entangled to some resolution, it could theoretically use all free resources of all fabs, and got production at any moment, even when all others suffer deficit.
Which means they compete for bandwidth in that generations stepping, and all of TSMCs 3nm is taken by Apple right now and NVidia is also fabless and depends on .. TSMC so this is an interesting "where in the queue are you" problem.