I was just thinking about this, but I reckon GPU plotting also increases the likelihood of people/companies to plot space when available/economic, instead of permanently reserving the space for Chia. Already now we see an uplift in netspace for the first time in a while now that the price has gone up a bit.
With plots being done in 2 minutes, we could see more rapid change to netspace along with the price.
It will simply come down to who can get storage at the cheapest price. Where is that in the world I wonder? If Chia’s price makes farming more worthwhile then prices of drives goes up again. The balance of nodes will centralize in a similar way to Bitcoin. The cheapest place to acquire anything is directly from the source. If Chia farming is deemed low power use then there won’t be the same pressure as the Bitcoin miners. The skin required in the game when (or if) Chia’s price makes this far more lucrative will become a cost burden. It would be like a new era of Chia farming. Like what happened to Bitcoin when small players were pushed out. I’m not filled with confidence on how this is progressing, but that may just be me. Aside from that? Have a great day everyone!
Good point - the space could be more transient. We saw that with GPU and CPU mining - people would just flip to another coin/pool when the gettin was good. That or mining rental (nicehash and the like).
No one knows for sure, bit imo that would likely push the price in the wrong direction.
Doesnt matter. There is no single person capable to sponsor filling up 18tb hdd in one day 
Not to mention there is no hdd capable storing each plot <4-12min. I wouldn’t worry much.
Do you earn 12000€/month just for hdds?
Netspace rises because chia suddenly, and without reason jumped 50%…one month later when chia drops down <30$ there will be depressed kids selling stuff with discount while blaming scalpers for high prices and rug pull.
I used to have gpus that I accidentally bought at right time. When I was selling 1000+, kids whined and called me names that I ripped them off.
Poor man always pays twice.
As you’d be limited by write speed you could easily fill up 18TB a day , multiple this by x for the amount of disks that a datacenter may have available at any given time and you could easily see bigish swings in netspace. Not to mention that a datacenter could have these disks in raid configuration.
These business would have the capacity sitting idle in any case, why not put it to use to possibly earn a little while it’s idle. Electrical costs for this would be minimum and wear and tear as well would be minimum. If a customer buys diskspace then you simply reallocate it and when a customer releases it, simply replot the space.
I read a report of a datacenter who had intended to do just this but ended up deciding that it was not feasible. With these new methods using less time and energy to plot it might just become feasible in which case we may see big swings in netspace.
BackBlaze have Petabytes of spare places, and they do not bother about Chia because it doesn’t make money.
I wouldn’t worry much guys…Chia is for hobby people
ROI 6-8 years for disks is no go in business world. Add your employees that costs at least 50$/hour…you do not make money.
Just calculate time you wasted on setting up farm, tweaking plotter…it is 20+ years of experience done in 100 hours. 100h x 100€ is 10k € just for your work. I haven’t seen anyone here mentioning/accounting for their time
Unless, Chia jumps back to 1000$ there will be no business/datacenter plotting because they earn more by their core business…if it happens, it will be quickly oversaturated…not making money again.
make all efficient, and hope your disks survive 10+ years to pay for your time, HW invested
I assume that you are talking about this article: Is Chia Farming in the Cloud Profitable? (backblaze.com)
The landscape has dramatically changed since this was published. There would no longer be SSD wear (they could easily only plot in RAM) and also plotting time will be much decreased. They also envisioned that the netspace would continue to grow which has been fairly stable the last months though I would expect this to increase if CHIA gains in price.
They initially calculated $5/TB costs to run which if we accept as still being current they would still lose money at the current price. 250PB returns 125k USD per month currently and their costs were approx 175k USD per month. A slight up tick in price to 80usd and it might (assuming also a slight corresponding increase in netspace) might make this worthwhile for them.
After initial trial and error they could easily automate most of this meaning their employee cost would probably be minimal after the preprod testing phase with only a few people “dedicated” to oversee its smooth running. Then your only cost would be electricity and since GPU plotting uses a fraction of the initial CPU plotting energy the above $5/TB could be cut a bit.
I am perhaps missing some things in my quick calculations but I don’t think the problems are as great today as they were in July 2021 to make this profitable. That said it would be better for the rest of us if it remains unprofitable for them
If you are speaking of a more recent report I would really love to read it.
If you do business, you do not dream as crypto kids do.
If you wanna plot large scale, and you have no people. Specialists easily costs 250$/hours…at least in central Europe.
Even guy with 20+ years knowledge would need easily 100hours to setup this operation - you are down 25k $.
If you wanna plot large scale, it won’t be done on tourist stuff kids use. Running 100 x consumer PC needs people if something goes wrong…and it always go wrong. You would be dreamer of perfect dream world of Matrix.
Or as you said, the RAM plotting way. I doubt they will be hunting bargains of DDR3 platforms. They already built their own servers before, and outsource now. Their machines lack power for plotting.
You buy EPYC multi core/CPU high efficiency platform that easily costs 50k $ per machine that can perhaps plot 500 plots/day…but you need another infrastructure to keep up with final plot storage…lets guess load balancers, NVMe cache…it would mean another thousands of $.
If you have so much crap around, you can either spend for state-of-art problem-free stuff…or you hire minions for managing this heap of crap. Either way, it will cost you fortune.
Let’s make it easy…you invest easily 100k $ to plot 250PB of plots. They have cheap storage for sure, they most likely buy it for like 10$/TB. That would be 10+ years ROI for them, assuming they resale plotting HW.
With their core business, they thrive…they are intelligent so they don’t do stupid things like trying add something they do not understand.
That is how profitable business works Or they can go Intel, IBM way…get arrogant…almost bankrupt, and then spend years to rebuild.
I agree with most of what you are saying and I see that I misunderstood Backblaze’s core business. I thought they were a datacenter when they are actually cloud storage providers so there would be a bigger cost outlay initially then I thought seeing as they would need to purchase the compute hardware. Maybe for Backblaze it wouldn’t be ideal but for a medium to large datacenter with plenty of spare capacity sitting idle it wouldn’t be a huge outlay.
Let’s do a thought experiment:
You are a datacenter with 250PB of storage buffer. While this storage buffer is not in use you decide to fill it with plots and delete the plots when the space is required for a client, no extra storage hardware outlay required nor cpu nor network infra as everything is already setup. Only thing I assume you will not have is the required Cuda core GPUs.
Now the next question is how much does your space requirements vary (how much is ordered and freed up per week) on average. I have no idea but let’s say that you might need to plot 5PB a week. How many GPU’s would you need?
Let’s say a 3060ti does a plot ever 4 minutes thus 15 plots an hour or 360 plots a day. Let’s now assume that you want to minimise cpu usage while harvesting so you plot at c1 or 90GB per plot. You therefore do approx 32TB a day or about 225TB (to keep things round) a week per gpu. You want to to plot 5PB a week so you would require around 23 GPUs. Let’s say 500 per GPU and your total hardware layout is then around 15000 rounded up to be pessimistic.
You now need to employ someone to prepare those GPUs and automate the plotting infrastructure. Let’s use your figure of 100hours or 25k. Your initial outlay is therefore about 40k to get things setup.
You now need to factor in electricity and upkeep. Now things get a little difficult to calculate as I have no idea how much power all the infrastructure would cost but the disks should be easy enough to calculate at consumer prices. Let’s assume 1 kwh is 0.25c (the price I pay as a consumer in Switzerland, a datacenter probably pays less than I would as they buy in bulk). Let’s also assume that they have 20TB disks which while farming chia use approx 5w an hour per disk. In 250PB you are looking at around 12,500 20TB disks so your daily hard drive electricity price is around 63kwh just to keep them spinning. Now multiply that by 30.4 (avg days in a month) and you are looking at 1.9MWh , let’s add another 1.1 MWh for general infra electricity and we are at 3MW per month farming or 750 a month. I’m ignoring plotting costs of 5PB per week but I guess that can be part of the 1.1MW of extra usage.
Now there will be hardware maintenance and plotting overheads so lets use a full time employee for that at 120k per year (to pay the employee salary, pensions etc) or 10k a month.
We therefore have:
Initial outlay of 40k to set things up to start with.
Monthly expenditure (employee and electricity) of around 11k a month.
We now assume that 230PB of the 250PB is more or less “plotted” permanently (including freeing up and replotting). How much does that make us a month and would it cover our expenses (once fully plotted).
Assuming current netspace and a price of 30usd we make approx 89k a month. Now obviously expenses would be higher due to having to replace more often some of the infra which I haven’t calculated butI think it would be doable. That said I don’t work in a datacenter so I am probably very wrong 
If nothing else an interesting thought experiment,
you still miss the main problem as most Chia dreamers
you can plot how fast you want - it still takes 4-12min to store plot to final HDD or average 8min/plot. Thus, you are limited to 180 plots/day/hdd. you can still fill just 1x 18TB/day if lucky
storing on SSD that cost 3-10x more per TB is fools game. If you have resources to buy 15$/TB, it is 6years ROI just hdd. Rich people are rich for a reason. No sane person invests for 18-60years in something with questionable future.
I have 48x hdd consuming 300W or 6.25W/disk…you are lucky, the part of Swiss I live in charges us 0.31CHF/kWh…and we have own hydro energy built by our taxes ROFL
We run most efficient farm/plotter so far we have seen there - based on data supplied by other guys in forum…yet, it is just a hobby to support future. You cannot make living from it…even, if you are poor guy from Swiss earning 5000CHF brutto
Assuming your wife, won’t kill you…and you are allowed to do your little hobby. You would need 18 PB to quit your job…you would also need 7.5kWh grid connectivity…usual house/flat have two circuits 230@10A or 4.6kWh. Not to mention 270k just for HDDs
As a hobby person, it may be possible if you can do it smart/efficient…can you imagine to manage 1200 HDDs?..40+ dBa noise…space
Based on your example…I have server hosing at Hetzner…they have enough resources. They pay 0.49€/kWh. If you have computing power, you do not waste it You will earn more money by renting computing resources/space.
Imagine what Google, MS, IBM, Amazon, BackBlaze, Pure Storage have…if it would make money…they would do it already, and we would be obsolete…The usual 500TB hobbyist would whine about poor earnings.
If you must pay employees, invest thousands into HW…it makes no economic sense. As a small guy, playing with few HDDs…it may be decent pocket money for kids
or if you are lucky one that have full control over everything.
own energy, building, connectivity…everything “free”…maybe. How many do you know?
If you have 250k CHF, you won’t waste it on kids’ games.
If you put this money in Intel stock…you would get over 12k just in dividends, “risk” and worry free…ROI 20 years
I think there is a language barrier here or you don’t really want to understand my point regarding data centers as it seems to be going over your head. I am not saying it’s profitable or worth doing considering the equipment is probably worth over 10 million just to earn around 50 k a month, but it is sitting idle in theory. Idle equipment costs money.
In any case, I never at any point mentioned any setup less than a data center and a thought experiment is just a bit of fun.
With regards to harddisk topping out at 180 plots a day. Here I’m afraid you are just wrong. When you are talking Enterprise grade equipment and the data itself is not important you would just throw your drives into a raid 0 array.
I think a raid controller tops out at about 1.5GBs so about a minute per plot for for standard sata drives times how ever many you need. Enterprise grade drives would probably be SAS so 12gbps (3GBs)in non raid or 200gbps in a SAN correctly configured which I guess would be our limit.
Data center mass storage would not be limited to 200MBs. Even if it was, with 12500 drives you could just transfer a plot to each disk at the same time and would in theory be 2.5 TBs if your network can handle all the traffic which, let’s be honest, it couldn’t.
I don’t know what my non existent wife has to do with this nor my salary. As an IT professional for a multinational ( not in the storage field or id much better able to debate this) I earn a bit more than you seem to think but again I don’t see the relevance. Chia is a hobby that I throw some money at now and again.
Anyway I’ve had my fun.
I’ve just read your replies and I now see why you mentioned a salary. Sorry I read it wrong, blame tired eyes and brain after a day at work.
As u said, I agree with you that it’s probably not feasible but I like hypothetical situations 