Search interesting materials

Sunday, March 15, 2015

The economics of cloud computing: an Indian perspective

When you buy computer hardware, and build a glass house in India, the costs incurred are those of:

  1. Skilled and specialised labour
  2. Electricity for computers
  3. Electricity for cooling
  4. Real estate
  5. Depreciation of hardware
  6. Cost of capital on all the equipment
  7. The problems of achieving high availability on the above.

Firms all over the world are finding that the make vs. rent choice shifts in favour of cloud computing: You shut down your glass house and rent the services of computation in the cloud, run by the likes of Amazon or Google.

Some of this flows from the universal logic of specialisation and economies of scale. Google hires Ph.D.s in aerodynamics to optimise airflow in data centres. Almost all end-user firms are unable to justify this scale of outlay. Google is able to put a data centre right next to a hydel plant and enter into a long-term contract for 100 years of electricity. Almost all end-user firms are unable to justify this kind of thinking for their data centres.

In addition to this, I think there are some interesting and uniquely Indian perspectives.

Cooling


If the ambient temperature is 5 C instead of 30 C, this reduces the cost of cooling.

Electricity


Electricity is very costly in India as the commercial sector is paying for subsidies and inefficiency. This is particularly after you take into account the complexities of ensuring high availability.

Real estate


Land is very costly in India.

Capital


Most important, the cost of capital is very high in India.

For Amazon or Google, the cost of debt capital is 4% and the cost of equity capital is 7 to 8% (in USD).

For a big listed company in India, the cost of equity capital is 15% and the cost of debt capital is 11% (in INR). For smaller companies it's much worse.

It makes a lot of sense to reduce the balance sheet size to the extent of the cost of the data centre, and replace it by a stream of rental payments to cloud computing providers outside the country.

This idea has many interesting implications. The propensity for Indian firms to rent from overseas cloud computing providers goes up when capital controls are introduced, goes down when inflation targeting works, etc.

Cloud vendors in India?


I don't see how cloud computing vendors in India can be competitive, as they face these same uphill problems of operating in India: expensive real estate, expensive electricity particularly after taking into account the HA issues, and expensive capital.

They might argue: The labour cost is lower; it's cheaper to get a Ph.D. in aerodynamics in India. My fear is that the labour cost component in the operations of a big data centre is quite small.

So far I have assumed that the core hardware (electronics + cooling) is at high seas prices. There are many mistakes in the Indian tax system, and this could easily not be the case.

Gains for end-users


Before cloud computing, if you were an end-user organisation in India, you had no choice but to deal with the problems: expensive hardware, low economies of scale in systems administration, the high cost of HA electricity and cooling, real estate and capital. Cloud computing is a big gain for India in terms of the reduction of costs that are given to end-users. Other countries will do what's their comparative advantage (producing cloud computing services); firms in India will import these services.

4 comments:

  1. I used to work at a company that tried to do this in the high-performance computing space (or at least that's what it evolved into). Disaster doesn't come close to describing the experience, and I'm amazed that the Tata's didn't think things through...

    ReplyDelete
  2. Agreed. And yes, the core hardware comes with an import duty of around 30% for most of the products against around 4 to 5% in countries like the US. That differential alone is huge barrier to start, given that almost none of these core components are currently made in India.

    ReplyDelete
  3. There are two aspects which are not considered in the analysis above:

    1. Geolocation - A lot of services require that the consumer of the service is physically near to the data center because of the limitation of speed of light which creates a perceptible time delay such as voice calling or watching high resolution videos. Also, the international bandwidth can be extremely costly to purchase for some of these heavy workloads. This is a technical and quality of service issue.

    2. Data governance and security - The government is a large purchaser of IT services and it is unlikely to host it data outside where it can secure the servers/storage or even the communication links. Snowden documents have shown that the spying agencies are tapping even submarine cables.

    The government would also prefer that the data and communications of citizens and institutions are easily obtained for investigation purposes. I believe (though I need to cross check) that the Companies Act 2013, while allowing only electronic copies of the accounts, requires a copy to be stored within Indian borders.

    We have already seen the big issue around this with Blackberry messenger services a few years ago. I am surprised that we haven't seen more of this with WhatsApp or Gmail et al. Maybe the current legal mechanism is proving adequate (which I doubt) or our investigative agencies are very adept at tapping (which is also questionable).

    In either case, I am listing reasons for the datacenter to be physically present in India. It would probably a provider such as Amazon/Microsoft or Google doing the set up given their competitive advantage at winning this market.

    ReplyDelete
  4. There is an important angle to the choice of location for cloud hosting which your article may have missed.

    Cloud hosting services are already reporting that almost each major economy is coming up with local data residency laws. This means that practically every APac country will ask for a local data centre within their jurisdiction. This is already taking away advantages of scale from cloud service operations -- most countries do not have local markets large enough to justify large investments in in-country data centres by these services. And if they do not set up these data centres, they miss out on business from some of the largest users of servers: government, telecom, banking and securities/commodities trading.

    This is not a case for on-premise hosting of servers. This is a case to set up data centres even where larger macro factors logically indicate that separate data centres should not be set up.

    ReplyDelete

Please note: Comments are moderated. Only civilised conversation is permitted on this blog. Criticising me is perfectly okay; uncivilised language is not. I delete any comment which is spam, has personal attacks against anyone, or uses foul language. I delete any comment which does not contribute to the intellectual discussion about the blog article in question.

Please note: LaTeX mathematics works. This means that if you want to say $10 you have to say \$10.