Source: Groupon Doesn’t Have The Cash to Build Its New Data Center

> What are they doing that <insert name of successful web company here> isn't?

Managing cash apparently. There have been questions about their accounting since the S-1 was filed, which now seem to be even more justified.

Actually,I think your initial impression is right. They are are an expensive business because of their army of sales associates. It is precisely because of this expense that they can't afford the technology they would prefer.

I've always wondered why Groupon didn't just provide a platform for well-connected local people to act as completely independent sales reps who work purely on commission. Why bother with actual outright hires? They're most likely going to have to lay a bunch of them off anyways.

Absolutely. It's headlined as if the situation is some kind of financial failure, but there are no numbers on the premium they're currently paying or the cost of a new center. 1/4 of the piece quotes prospectus boilerplate. No named attributions, nothing from analyst types following the industry.

Not even clear cash is the key here anyway, I would think they could finance a data center with some kind of construction mortgage, and maybe some leasing deal.

Now, pulling things out of my hat, I understand that companies can fully deduct their IT investments made in 2011, that's a huge tax advantage. I don't even know if Groupon can make use of that (are they paying taxes yet?). But if Groupon is missing that window because they _cannot_ get financing in place, _that_ would be a story, one with implications for Groupon, their model, the banks, and bank valuation of corporate IT capital assets.

This, on the other hand, is not a story of any sort.

More importantly, IF they have that many pageviews, it ought to be scaling in proportion to deals with sales of groupons. So either their margins are horrible, or their tech cost per unit sold is horrible, or something else we don't know about is causing the issue.

It could be they're trying to estimate based on handling their growth for the next three years, for example, in which case of course the cost could be horribly out of whack with current revenues.

Hey - I'm the reporter who wrote this story.

I wrote the headline to make it clear that this was coming from a single source. I am not an expert in hosting or infrastructure.

That being said, there is an interesting discussion going on over at Betabeat. To quote on comment = "At one side is platform as a service or managed platforms on top of EC2 -- Heroku, EngineYard (which Groupon appears to use), etc. These are easy to set up, and have lower overhead for small or medium organizations. They have the highest per-unit cost, but for rapidly growing or small organizations, can be worth it."

I think its fair to say Groupon is a large company and past the point of its most rapid growth. To me that fits with the idea that they are looking to make a transition but have not accomplished it yet, whether that change is moving to EC2 or building their own data center.

You might be surprised; I know Alexa isn't exactly an infallible oracle, but they list groupon.com as averaging between the 200th and 400th most visited site globally, and 53rd in the US. "ranking.com", whoever they are, says between 150 and 75. I'd feel confident saying they're at least in the top 500, and that requires a fair bit of horsepower. Now, they could probably solve a lot of their problems with smarter engineering before building a data center, but they certainly have traffic.

Joking aside, I think this might be a good idea.

I have absolutely no idea what Groupon's load actually looks like, but as for actual datacenter cost, $1,000/sqft would be a starting minimum, and that's for a decent sized datacenter where some economies of scale kick in. Google supposedly spends as much as $3000.

One of the smaller datacenters I've ever been in was around 20-30,000 square feet. Replicating it probably would have cost closer to $3000/sqft than $1000, so you're probably talking about $40 million minimum.

A rule of thumb is $4000 or $4500 per kilowatt.

So 100KW of power, with N+1 power feeds, generators, full UPS backup system, etc. would cost about $500K to build. Would give you 10 to 20 racks' worth of power.

However, that is too small to be worth it, so they would probably build something quite a bit larger than that.

Figure 200 racks, each with 8KW-10KW per rack, that means 2 megawatts of power, thus 2000 * 4500 = $9 million for the building.

Figure $40K per rack to fill it with hardware (at a low estimate), half full of 100 racks is another $4 million.

Plus they have to manage construction, sign various contracts, pay permit fees, buy land, pay for fiber, etc. At a minimum they would need $20 million all in before serving the first byte of traffic.

All this is my estimate, could be entirely wrong.

Depends on the data center of course.

There are four components to a 'data center':

1) The structure. These can be pretty simple, a concrete slab, then 'tilt up' walls tied together by steel joists. Internal structure is 'UPS area' / 'power ingress/conditioning' / 'data area' / 'office area' / 'fire systems area'. If its mostly data operations the bulk will be the 'data area'. The cost of land figures into it as well but you probably try to build it where land is cheap, connectivity is high, and power is cheap.

2) Power and Cooling infrastructure. This stuff is dictated by both the climate around the data center and the municipality. Santa Clara California used to have a bunch of semiconductor fabrication plants (fabs) which have all since closed. That left them with excess power capacity, so lots of folks are building data centers there to get access to the cheaper power. (land is still expensive). Cooling can be one of a number of forms, evaporative works well in the Bay Area generally.

3) Staff - Generally an electrician, a network guy, a security lead (+ some number of contract security personnel)

4) General 'stuff' associated with building a data center (raised floor tiles, sprinkler systems, power conditioners, big switches, UPSes and their batteries, chillers, more big power switches and transformers, generators (back up power), lights, etc.

I would be surprised if it cost less than $5M to build or more than $15M.

I actually put together a plan for a 'pocket' data center (one that could be put into an urban area easily, 1 - 3MW of power) but not surprisingly cannot find a bank to fund it. (its around $10M all up depending on initial land and fiber costs).

Are they allowed to do this as a public company? Would their growth be attractive enough for someone to take them up on the offer?

Don't they already use EC2?

It isn't the answer to everything. In fact, it's the answer to far fewer things than people think. EC2 is great because, as you say, it has no big up-front cost. But when you start scaling it isn't actually very affordable.

A recent blog post by Engine Yard claims that they host Groupon's site right now.

http://www.engineyard.com/blog/2011/groupon-makes-history-in...