Amazon.com and the fabless semiconductor industry

[VentureBeat criticized Amazon for lack of vision. Baris Karadogan, of ComVentures, begs to differ and commends the company.]

A key theme at the Web 2.0 conference was infrastructure, centered around Amazon‘s big announcement to offer start-ups all their infrastructure needs.

Amazon wants to give start-ups bandwidth, storage, server software and even fulfillment of physical goods — all at a reasonable price. This is all the boring stuff you need to go from your great idea to a product. Google and Salesforce.com call this infrastructure “the cloud,” Microsoft calls it “a million pounds of batteries,” and Amazon.com calls it “muck.”

The idea goes like this. Getting your web application to scale globally and serve hundreds of millions of users is expensive. You need to build big data centers and buy software and fat pipes. Google spends more than one billion dollars per year on this (and their ability to do this is a key competitive advantage over others ; see here).

Startups can’t afford that kind of capex. We VC’s don’t see Series A opportunities that say “we are thinking of raising $500K to launch our web site and about $700M to build a global data center.” These startups go to hosting providers and buy computing and storage by the drink. This is what Amazon wants to provide startups, everything; the software, the hardware, the bandwidth and even physical fulfillment.

Sound familiar? This is very much like what happened in the semiconductor industry as it went to the fabless model. The transistor was invented in 1947 and the first IC (integrated circuit) was built by TI in 1958. By 1978, the semiconductor industry reached $10B in sales. As chips got smaller, it became harder and more expensive to build the foundries to build the chips. While Intel and TI continued to own their own fabs (where you build chips), smaller companies could no longer afford the billions of dollars required to build them. As a result, in 1987, TSMC, the world’s first independent semiconductor foundry was established. It is still the largest. Now a startup no longer needed invest the capex to manufacture their chips, they could design it, and the foundry would build it for them. Since the foundries amortized the capex over a number of companies, the model worked. The foundries in the semiconductor world are the data centers of the Internet world.

In the Internet world, a few big players have emerged that have the wherewithal to build their own big data centers. Microsoft, Yahoo, Google and Amazon are what Intel and TI are in the semiconductor business. Now we are seeing that some of them, like Amazon are selling their capacity to startups, and some pure play data centers businesses are emerging. I suspect we’ll see more of them, establishing themselves and rolling up.

Now back to our story. After the large foundries established themselves, fabless semiconductor startups started showing up and growing their business to respectable sizes. They design the chips, TSMC built them. The two big fabless semi companies are Broadcom and Nvidia, which are both around $2.5B in annual revenues. So a healthy ecosystem formed with independent foundries building chips for fabless semiconductor companies which focus on applications and software.

So what can we learn from this analogy? Despite the fact that the fabless semiconductor revenues reached $41B in 2005, it is still merely 18% of the overall semiconductor industry. Intel, who owns its own fabs, has about $40B in sales, nearly the size of the entire fabless semiconductor industry. Those who were early and kept their infrastructure, got really really big. They delivered products faster and cheaper than their fabless competitors. Same will be true for Google and Amazon (and Yahoo if they can get their act together). Amazon could offer to take care of the “muck” for a small startup, but as soon as that startup gets big, Amazon will able to do the same faster and better because they control the critical piece, just like Intel.

So I applaud Amazon’s strategy. It’ll get an early look at the apps, and maintain control of a critical piece of know how. As data center complexity grows, it will be harder and harder for a newcomer to build their own datacenter, and will be more and more reliant on Amazon. What would be a nightmare for them is for an upstart to out-innovate the “muck.” So Amazon’s strategy to dominate early on prevents startups from going there and independent data centers from surviving, since few companies have Amazon’s scale. This has got to be Amazon’s biggest move so far to ensure its sustainability.