Sure, the smaller players might be able to run GLM 5.2 on 4 sparks or on runpod but I'm not sure that solves the asymmetry for every business which might be a future target of an agentic attack.
Having a model running in case of attack might sound reasonable, but I've worked long enough with systems to know if have something running for a 'just in case event' and not actively used, the moment you need it, you might find all kinds of issues.
Worse than this, very few businesses would have the resources or skills to run their own frontier or near-frontier model. GLM 4.2 on a few sparks or runpod might sufficient today, but in six months, a year?
Ensuring this capability is available to defenders when needed, with appropriate guards that defenders can identify themselves and be transparent about what they are actually doing and why, seems like a strong case for business services to provide over every business developing this capability for themselves.
