"Open Source Is The Ultimate Leveller"
Photo Credit :
AMCC has revenues of only $248 million. Still you take on Intel in the server space. The $53-billion giant could have you for breakfast without even realising it.
Why just breakfast, they could do it for a paan or even as the clove on the beeda (laughs). Look, I have taken on challenges and done well in the past. Before taking over as CEO here in 2009, I was at Marvell (Technology Group, another fabled chip design firm), where I started and grew a business — building the iPhone's Wi-Fi capabilities, etc. Today, every laser printer from HP carries my chip. I built the business from zero to three-quarters of a billion dollars in six years.
Our key legacy businesses at AMCC were dropping off. In the past, AMCC was a niche telecom and an embedded processor player. So when we drew the strategy roadmap in 2009, we were clear that we wanted to address a larger market. We foresaw the explosion of data (and that) its management from mega data centres would pose challenges. The US Department of Energy estimates that five years ago, data centres consumed about 1 per cent of the power produced, three years ago, it was 2 per cent and this year, it will be about 4 per cent. This accelerated power consumption by mega server farms is clearly not sustainable. (A server farm is a group of networked servers). We wanted to become the silicon providers for the next generation of data centres in both connectivity and computing. I agree, it is an ambitious target.
So, how big is the pie here?
On the infrastructure market for data centres, these are our estimates: server systems alone are $50 billion and server silicon is about $12 billion. Connectivity — which include cables and fibre optic interface chips — is $5-7 billion and an additional $20 billion is in the form of equipment. Even if we get a sliver of this, we would have done a good job. Friends tell me I must be insane to attempt this, but there were also calls from the Googles and the Amazons of the world. Facebook has open-sourced its server specifications. Gone are the days of high-priced, proprietary, closed architectures that guzzle power.
Semiconductor technology, energy efficient computing, communications solutions for data centres, telecom, enterprise and consumer applications
$248 million in 2011
Canada, China, Denmark, India, Japan, Taiwan, UK, US and Vietnam
Is the churn in the data centre an opportunity for players like you?
It starts at a higher level. Even now, 70-80 per cent of the server market has the Windows ecosystem and the rest is captive ecosystem, which includes vertically integrated large players such as Cisco, Google, Amazon, Facebook, Zynga, Twitter, etc. They run distributed file systems for cheap, which have caught on in lieu of Windows tools. This is a fast-growing segment. In the open source ecosystem, such white label servers supplied by Taiwanese firms rule. These large enterprises are worried about their energy costs. Open source is the ultimate leveller and provides alternative architecture such as ours a better chance at success than in the past.
But for somebody of your size, isn't this a risky bet? While the ARM architecture wins on power intake, aren't there issues on scalability, performance and security?
Instead of incrementalism, I believe in taking a quantum leap in thinking. These kinds of market shifts happen only once in decades. Five years ago, did people envisage a single data centre consuming 20MW of power? We want to capitalise on this shift and have made investments to take advantage of it. When we did the deal with ARM, we licensed the micro architecture. We did not get ARM's cores... we got 300 pieces of paper, out of which half were blank, which we helped fill up and thus got a discount (laughs).
Look, Intel for long has run this as a performance and a margin business and till recently had no incentive to change that. Why did Google start building its own servers? It realised, ‘hey, I don't need half the stuff that HP and Dell sell me in their servers; I don't need four network interfaces; I don't need five chips, five memory copies'. Hence, changing all that could change the effective power requirement.
The arguments on performance and scalability are not true. Today, the largest video-on-demand network service is run on the cloud. It is not run on Windows. It is Netflix on Amazon. Netflix's profitability is dependent on Amazon's ability to offer both network and server bandwidth to host their movies. So, scalability and performance are not issues anymore. What we have done is outside of main memory and external storage; we are integrating network and storage interfaces. By the way, all this has been done in mobile. Agreed, there we are talking in gigs (gigabytes) and not megs (megabytes). But at a process level, we have done it.
I don't agree it is a risky bet. Outside of the X-Gene (processor), we are launching 12 new products this year. Small, embedded processors, signal processors, etc. Even outside the ARM business, we will do well this year. We will turn around the company. We have made a bold technology bet, but not a financial bet. We will be the first with the 64-bit ARM architecture and have a headstart even if others license some stuff from ARM. HP has already announced that it is no more beholden to the x86 architecture. In 18-24 months, we will see a meaningful inflection point in terms of market share and value share being taken away from the x86 architecture in the server market.
What would be India's role in your scheme of things?
India plays a crucial role in what AMCC does globally, with 200 of our 600-plus workforce being located here. They are creating critical pieces of communication and computer scale-out parts of X- Gene out of India, and even more on the software and benchmarking side, too. The work being done out of India will determine our global success.
(This story was published in Businessworld Issue Dated 23-07-2012)