The most important technology development of 2006 was a press release from a books-and-CDs retailer announcing a service to rent computing capacity by the hour. Almost no one understood what they had read. This letter is about how often the consequential moments arrive disguised as administrative announcements.

On AWS, Read Carefully

Amazon Web Services launched its storage service in March and its compute service in beta in August. The pricing was meaningfully below the cost of running equivalent infrastructure in-house. The pricing was also meaningfully below the cost at which Amazon was running its own infrastructure. Both observations are relevant. Both have implications.

The implication that matters most to us is structural — every venture-backed company founded after 2006 will be built on rented infrastructure rather than owned infrastructure, and the capital efficiency of new companies will, in consequence, increase by an order of magnitude. The next generation of consequential companies will be built for less money than the last one. This is good news for founders and worth understanding for investors. It is also, we should note, ambiguously good news for venture firms whose returns historically depended on the capital-intensity of the businesses they backed; smaller checks compound differently from larger ones, and several venture business models will need to adjust.

We have, in the second half of this year, walked through every active commitment in our portfolio with each founder, asking whether their infrastructure plans should be revised in light of AWS. Two of them have already revised; three are evaluating; the rest concluded that the existing infrastructure plan was right for them. We did not require any specific outcome. We did require that the question be asked. The question matters more than its answer, in our experience.

On the Companies That Will No Longer Need to Buy Servers

The companies in our portfolio founded in the second half of this year have, almost without exception, deferred infrastructure purchases that the same companies, founded eighteen months ago, would have made in their first month. The capital that is not being spent on servers is, in the cases we are watching, being spent on people. We expect that to change the shape of these companies materially over the next five years.

The deeper change, which we expect to be visible only in retrospect, is in what kinds of products are now possible. Several product categories that were considered structurally unprofitable in 2003 — because the cost of provisioning infrastructure for unpredictable demand was too high — are becoming feasible in 2006. The companies that will dominate those categories in 2010 are, in many cases, being founded right now. They do not yet have product-market fit; some of them do not yet have product. But the substrate they will build on has shifted, and we expect the founder cohort that recognizes the shift earliest to be disproportionately rewarded.

On What "Platform" Is Quietly Becoming

The word "platform" was, until recently, applied retrospectively to companies that achieved network effects. It is now applied prospectively to companies that hope to. We do not endorse the prospective usage. A platform is a fact, not an ambition. The most useful test we have found is whether third parties have already built businesses on top of it without being asked.

By that test, very few of the self-described platforms we evaluate this year actually qualify. AWS qualifies; we know of three companies that have built revenue businesses entirely on top of it, with no commercial relationship with Amazon. The PC operating systems qualify; the mobile ones do not yet; the various web frameworks proposed as platforms in 2006 mostly do not. We have begun applying the test as a binary filter. A company that calls itself a platform without satisfying the test is, in our reading, a company that has misunderstood what platforms are.

On the Investments We Made Because of an Internal Memo

An internal memo we circulated this summer, on the implications of pay-as-you-go infrastructure, ended up justifying three of our six commitments in the second half of the year. We are not sure whether this represents good thinking or selection bias. We mention it because, in our experience, the investments that are easiest to defend in retrospect are usually the ones we hesitated longest over at the time.

The memo itself was unremarkable; it described, in eight pages, the operating implications for venture-backed companies of a world in which compute and storage are rented rather than owned. The implications are not novel; they had been discussed by capable people for several years before AWS launched. What was novel was the proximity of the implications. AWS made the abstract argument concrete enough that a founder could be expected to act on it, which made it concrete enough that we could be expected to underwrite the founder's actions. The interval between the abstract case and the concrete one is, in our experience, where the most consequential investing decisions are made.

A Closing Note

The year's most important news arrived as a ninety-word press release. We have been training ourselves to read more press releases, more carefully, ever since. The signal-to-noise ratio of corporate communications is, in our reading, lower than it has ever been; the signals that remain are, in consequence, more rewarding to identify. We will spend more time on this in future years.

The Partners
Winzheng Family Investment Fund · December 2006