As published in The Next Platform, http://www.nextplatform.com/2016/02/18/the-joys-of-not-owning-a-supercomputer/
Shopping for high performance computing machinery can be a sobering experience. That’s because buying high-end equipment in bulk is an expensive proposition.
A single server costs only a few thousand dollars, but even a relatively modest-sized HPC cluster can run into the hundreds of thousands of dollars,
especially when software, support and maintenance costs are bundled in. For top tier supercomputers, those costs can climb to $100 million or more.
And even though the price of HPC componentry falls with each passing year, thanks mainly to Moore’s Law, system costs aren’t declining. That’s due
primary to the fact that HPC is a unique market, where users are a bottomless pit of desire when it comes to computational performance. As a result,
they want as big a system as their funding will allow for.
So it’s no surprise that HPC buyers tend to be a frugal bunch when it comes to system procurement season. What is surprising is that the majority of
them aren’t taking advantage of traditional financing and leasing arrangement to make best use of their cash flow and funding cycles.
According to our research, based on surveys of HPC sites from 2011 through 2015, the majority of users are buying their HPC gear outright. In fact,
80 percent of the HPC sites surveyed were not using any leased equipment at all; only 20 percent reported leasing at least some of their HPC hardware.
Although we didn’t specifically ask about loans, it’s reasonable to assume that participation rates would be similar.
There is some variability between sectors, with academic and government sites reporting the lowest rates of leasing: 14 and 20 percent, respectively.
By contrast, more than a third of commercial HPC sites, 36 percent to be exact, reported leasing some hardware. For those commercial sites that
did lease, they reported doing so with about half of their equipment, on average. Nevertheless, across all sites we surveyed, less than 10 percent
of all HPC hardware is being deployed under lease.
It’s not entirely clear why more of these users aren’t taking advantage of leasing or financing. No doubt some of this resistance is to avoid paying
interest or financing charges, which to HPC customers just seems like money that could have been spent on more CPUs or software. But as any decent
financial advisor will tell you, spent cash has certain costs associated with it, which may end up being higher than the cost of debt.
For government and academia users, these low rates of leasing are probably the result of procurement habits that encourages spending all allocated
funds by a specific deadline. This funding-then-spend model in the public sector does not lend itself to optimal decision-making. For commercial
users, habitual buying behavior is probably also involved, although, as noted before, their leasing rates were significantly higher than the government
and academia users.
Perhaps there is a general lack of awareness that these services are even available for large-scale IT purchasing. Traditional lenders are unlikely
to play in this space, but there are a handful of financial service providers that specialize in IT financing, and they tend to be associated with
the big OEMs – IBM, HP, Dell and others. The typical HPC buyer, however, may be unaware such institutions exist.
There’s probably also a sizeable chunk of HPC users who don’t see much advantage in going the lease or borrow route. At the organizational level, though,
there are some obvious advantages, namely, that leasing or financing preserves cash for expenses that cannot be leased or financed, such as paying
staff salaries and the data center’s electric bill. That preserved cash could also be used to get a bigger HPC machine than they would if they
had to purchase it.
Even when financing or leasing a system costs more than purchasing it up-front, the advantages of having additional cash around for other uses often
outweighs the extra expense. That’s especially true on the commercial side, where cash flow constrains nearly all capital expenditures. When it
comes to CAPEX or OPEX, most companies would much rather most expenses end up in the OPEX bucket. It’s just easier to manage.
But HPC buyers, in particular, have an even more compelling reason not to purchase their systems out-right: The average cluster is replaced or upgraded
about every three years, and leasing it over that timeframe can actually be less expensive than buying it. That particular insight was revealed
as the result of research and analysis we performed for a whitepaper that
explored how leasing and financing could be leveraged for HPC procurement. The paper focused on the various leasing and lending services provided
by IBM Global Financing (IGF), the financing services arm under IBM, and probably the largest such institution in the industry.
Using the online calculator provided by IGF, it became apparent that you can
spend less money on a three-year lease than on a system buy. In the scenario I played with on their site, I leased $250,000 worth of make-believe
IT equipment and financed $50,000 worth of make-believe software and services (which you can’t lease), using what IGF calls their Fair Market Value
option. Over the three-year term, the total cost was $281,664, or near $20,000 less than the $300,000 up-front purchase. If the leasing term is
extended to five years, the cost goes up to $337,980, or around $38,000 more than the purchase price. With leasing of course, you aren’t able to
leverage any residual value of the equipment, but in many HPC environments, a lot of users are just doing fork-lift upgrades, so there is no residual
If you want to keep the equipment at the end of the lease term, there’s an IGF plan for that, known as the Full Payout option. Using that arrangement,
the same $300,00 purchase would cost $321,660 over three years, or $342,840 over five years. It’s worth noting that when you get out to the four-
and five-year timeframe on this Full Payout option, monthly payments are not only significantly lower than that of a three-year term, but aren’t
that much more than those of a Fair Market Value option. In other words, the advantages accrued to your cash flow are similar when you lease or
borrow at these longer terms.
Speaking of cash flow, these same IT financial service providers sometimes will offer a buyback program, where you can sell equipment you own to them
in order to generate some cash. That money can be used to lease back the equipment or for anything else. Since these financial service providers
are well-integrated into the IT sales channels, they can resell equipment they’ve purchased on the secondary market if you choose not to lease
it back yourself.
Of course, you can always do traditional loans with these institutions as well. IGF, for example, even offers zero percent loans on a 12-month payback
time-frame – basically free money for a year. As with traditional consumer loans, there are a lot of permutations to choose from, and with interest
rates at historically low levels, there are many good deals to be had.
The bottom line is that leasing or financing would enable a lot of HPC sites to better utilize the funding they allocate for infrastructure, software,
and services. By turning a CAPEX expense into an OPEX one, these organizations can achieve more financial flexibility. And when it comes to high
performance computing, that’s money in the bank.