Web Infrastructure And a Startup Funding Manifesto


Two-thousand eight will be remembered as a watershed year for many reasons, but two are of special interest to the startup community: the meltdown in the financial markets and the emergence of cloud computing.  Tighter capital means investors will be more cautious, and startups can expect lower valuations. However, the growth of cloud computing provides a possible opportunity for both investors and startups: cheap and easy experimentation.

To many in the IT industry, especially those considering starting companies, the motives of venture capitalists can seem confusing at best and devious at worst.. But like any other industry, VC has its own business model. While startup founders often think about investments simply in terms of percentage returns, VCs tend to think in terms of absolute return. Getting back $10 million on a $1 million investment is a huge percentage ROI, but getting $50 million from investing $10 million is a larger absolute return. The rational behavior of VCs thus seems irrational to others.

VCs, then, are doing two things: betting and experimenting. The bets are on the entrepreneurs creating the companies and, to a lesser extent, the ideas they have.  The experiments are used to identify and explore new markets. Bets are pure risk and risk management in which the goal is to maximize upside while minimizing downside. Experiments, are about discovery, and discovery means exploring many blind alleys before finding success. Scientists run tens or hundreds of experiments that fail for every one that succeeds in revealing something new.  The same is true for startups: if you are not failing, you are not innovating.

The U.S. Bureau of Labor Statistics reports that the IT industry had the lowest 2- and 4-year survival rate of any industry during the ’90s (63 percent and 38 percent).  Why? The companies, like investors, must balance risk and experimentation. Many companies start with one idea, only to discard it and try several others before arriving at the idea on which they succeed. Google (s GOOG) is the most visible example of this, having launched its advertising platform years after starting their (non-monetized) search business. Many companies run out of money before discovering their successful offering.

The need for experimentation and scarcity of resources makes it essential that both VCs and startups be able to run their experiments as cheaply and efficiently as possible.

In the past, building a web-based startup required purchasing servers, colocation and bandwidth, and hiring the operations people to take care of load balancers, routers, disk storage, etc. This is a lot of expense and often companies either under provisioned or, more likely, bought too much equipment in anticipation of demand that never materialized. Most of this equipment later could be found on eBay (s EBAY), which, while good for people purchasing expensive equipment for pennies on the dollar, wasn’t the revenue building plan the investors had in mind.  Such capital expenditure is clearly not conducive to rapid, frequent experimentation.

With the launch of services like Amazon Web Services, Google App Engine and Force.com, however, capital costs of infrastructure are zero, having been converted to operational costs that directly reflect actual demand.  For those with applications that can work within the constraints, AppEngine and Force.com will automatically scale them on demand.  The more sophisticated applications, designed from the start for horizontal scale, are better suited to the set of infrastructure primitives that make up AWS.

All of these platforms are infrastructure, but they are more than that for startups and investors: they are tools for risk reduction and experimentation.  They are engines of innovation.  In the new world these services create, startups using them have an enormous advantage over their competitors in their ability to experiment and adjust, and investors can get much less risk and much more exploration done with the same capital. Anyone building their own infrastructure will be seen as already failing.  What many see as simply a mechanism to reduce costs is, in reality, something much more powerful.  Fast, frequent failure is the most reliable path to success.

If your business plan relies on building infrastructure, it’s time you revisited it.  If you don’t, your investors will.

Benjamin Black is an infrastructure troublemaker, currently at Joyent, who has blazed a trail of game-changing technology, leaving behind angry execs at Microsoft, Amazon, and Internap. Vijay Gill builds networks and infrastructure.


Paul Wall

Interesting read. However, as neither author has actually worked at a startup, I worry the prospective might be a tad inaccurate.

Ranjit Nayak

Very interesting observations for all entreprenuers and investors. It seems like the common sense approach – using leased infrastructure. I would be surprised if startups decide to build their own data centers.

Benjamin Black

Thanks for the thoughtful response, Adam. I’ll take your points in order:

1) You are absolutely right that, for sophisticated apps/sites, you need a capable operations staff. For apps that can run on the more abstracted stacks, like AppEngine, though, there can be little need for additional operations folk. Fitting within the length target for the article meant focusing on one aspect of the total problem. Perhaps we can do another about operations.

2) While you can certainly get started on a DSL line and a machine under your desk, the cloud is far more attractive, in my opinion. The costs are usually quite small to start and the availability and scalability advantages are enormous. Bootstrapping doesn’t have to mean skimping on infrastructure, anymore.

3) Force.com could be seen as expensive. It could also be seen as incredibly cheap because there is no capex and no large, up front software investment. If something is not part of your core business, outsourcing it can be a big win. Hence, force.com. Please remember that there is no free lunch: complexity doesn’t disappear by moving to cloud offerings, it just moves up the stack where it can be more easily and cheaply managed.

4) AppEngine had to start somewhere and, since Google has incredible depth in Python that’s what they chose. I don’t expect they’ll stay Python-only and I absolutely agree with their logic. Go with what you know, then work out from there.

5) There is, indeed, no kicking the cloud. I see that as a good thing.



Hi Benjamin. Good article.

However, you fail to point out that…

1) OF COURSE you still need great operations engineers, even if you’re running completely on the cloud. Anyone who says otherwise is selling something.

2) Plentyoffish.com ran on Markus Frind’s desktop machine and an Internet connection for lord-knows-how-long. Probably way longer than anyone reasonable would do so. But still, it goes to show you that even cloud computing is overkill for most projects. Get yourself a DSL line and a static IP address, and you can get started.


3) The pricing of Force.com is unbelievably expensive. Plus, it just pushes the hassles to a different layer of the stack.

4) Google App Engine only lets me write apps in Python 2.5. Exsqueeze me? Baking power? How about PHP, Ruby, Java, Perl, or (gulp!) C#? Or will they gladly kick me over to Amazon and/or Microsoft for those applications which represent, oh I don’t know, like 99.999% of all web traffic?

5) I like hardware because I can kick it. There’s no kicking the cloud, man.


John Doe

“infrastructure troublemaker” is an understatement, as is the scope of the management chain he leaves behind angry

John Doe

Classic = “Benjamin Black is an infrastructure troublemaker who has blazed a trail of game-changing technology, leaving behind angry execs at Joyent, Microsoft, Amazon, and Internap. Vijay Gill builds networks and infrastructure”

Mark Evans

Very well said. This article puts everything I’ve experienced over the past ten years in perspective; spot on. Big opportunities are in front of us.

Comments are closed.