And logistics of sitting up a ton of GPU machins, and troubleshooting them.... I don't see anyone picking a industrial size gpu farm.
I wouldn't not worth it most im gonna do is 6 or 7 build them in PC sense i can all ways use one or two more PC's for other things and maybe rent a few form time to time sense the rent on them is not to bad you can actually make something renting not much but something .
there have been plenty of 100+ GPU setups in the past, but even 100 R9-380 cards would only draw ~20kW, which is a drop in the bucket for a facility that can handle at least 10x that much equipment. They would need to run 500+ GPUs to even remotely make proper use of the facility
BUT... if they did, it could be solid returns and resale/liquidation may be much easier than it is with bitcoin ASICs.
1) build a simple, functional, rackmounted system with cheap components (<$200 for mobo/ram/cpu/etc/crappy 350W ATX PSU) that can handle 5 GPUs (powered by a single 1250W server PSU, which they likely have hundreds of lying around from SP10/3x/50). costs beyond the GPU alone would be minimal, at least $50 savings over even the most budget home-made rigs due to bulk purchases and existing infrastructure
2) using either linux-on-a-USB or linux/windows running on a <$20 32GB SATA, set up the software side to start mining on boot and open a teamviewer or other remote-interface program. (remember they designed some pretty good interfaces on the ASIC hardware, a linux/windows remote monitoring app would be a breeze).
3) Replicate the system, with the exact components dozens of times (should take <30min/rig if the OS is programmed right) until you have at least 100 rigs, which could likely be fully assembled and functional by a small <5 person team within a week. They then mine at a decent profit because they pay ~$0.05/kwh, assuming r9-380x tats about $2-3/GPU/day in profit, or about $1000-1500 for a 500-card assembly. Within a year they could be looking at >$200,000 in revenue (after power costs and assuming a gradual decline in profitability), with ~$125,000 in initial hardware costs and probably $75,000 worth of costs related to their lease and minimal income for those involved in the initial assembly and ongoing operations.
notice that's effectively a $25,000 loss after 1 year, assuming that GPU mining maintains profitability. (brief price movements like in ethereum could add another $100,000+ of revenue, making it profitable). They also now have ~$60,000 worth of consumer hardware that could be liquidated without much difficulty, assuming they dont continue to mine for 1-2 more years before the 380 is obsolete.