NVIDIA CEO Jen-Hsun Huang posing in front of a GPU-powered supercomputer
As Intel is looking for a new CEO, something surprising did happen recently: the company has hinted that it may bring in someone from the outside, which would be a first for Intel. So far I’ve heard names like Sanjay Jah, the former Motorola CEO (and ex-Qualcomm COO) as potential replacement for Intel’s Paul Otellini who is retiring in May 2013. Tech site BSN has started a new Internet buzz by coming up with a totally different rumor: according to them, Intel is looking at buying NVIDIA, and would name Jen-Hsun Huang (NVIDIA’s CEO) as Intel’s new CEO. Now, *that* would be something that we would like to observe. This rumors is backed by pretty much nothing, but it is fun to discuss: so, how likely is it to happen? Let’s look at the pro and cons for Intel.
Precedent: AMD buys ATI
There is a precedent: the ATI acquisition by AMD which is working so far, maybe not financially, but ATI has been well integrated into AMD. The only thing is: the AMD/ATI “fusion” failed to produce anything that puts either Intel or NVIDIA at risk, for now.
Most people don’t realize this, but the risk you run when acquiring a high-tech company like NVIDIA is that the company’s assets and possibly even patents are nearly not as relevant as a number of key employees. If Intel was to buy NVIDIA, it would need to make sure that they are not going to walk away because they don’t like the new corporate culture. This is always a very risky investment. Naming Jen-Hsun Huang as a the new Intel CEO may cement that loyalty, but it also risks alienating key Intel employees – you never know.
With a market valuation of nearly $8B and a profitable business at NVIDIA, the financials would work and NVIDIA is by no means “too expensive” for Intel to acquire (assuming that the NVIDIA board wants to sell), even with a good 20% premium on the company’s value.
Past and current NVIDIA stances may jeopardize an acquisition
However, Jen-Hsun Huang has expressed many times the idea that Intel’s X86 architecture isn’t “optimum”, especially for mobile devices, saying “if you had to start from scratch today, would you use X86? Probably not”.
To be fair, Jen-Hsun Huang has a point as X86 brings a tremendous business value-proposition by providing impeccable binary backwards compatibility and high-performance, but at the expense of size and power efficiency (when compared to ARM designs). In a mobile world, power efficiency should always come first, and that’s why Intel is still experimenting (and getting better at) in that market. But Huang’s view on Intel’s architecture may spook the Intel board as they may not want any radical changes to an otherwise very profitable business. Intel has $11B of net income a year versus $0.5B for NVIDIA.
Could NVIDIA provide a “hyperspace jump” to Intel into the Mobile space?
A lot of people would hope that an NVIDIA acquisition would propel Intel into the mobile market in a big way. It would certainly be interesting to see what would come out of such a union, in technological terms, but in the mobile space, NVIDIA has to overcome a challenge that is even harder than building great chips: Apple and Samsung alone have 50%+ of market share in the smartphone business, and they want to use their own chips. Who builds the handset decides what chip it will use.
On top of that, Qualcomm has a huge market share, thanks to a combination of processors, GPUs and modem that nobody else can bring to the table right now. To get a significant share of this market, NVIDIA/Intel would need to build a chip that is so compelling that competitive pressure would force Apple or Samsung to relinquish using their own chips, which are very good by the way.
LTE: Keep in mind that NVIDIA has acquired Icera (more Icera stories), a modem company that designs programmable modems. The idea is simple: instead of building many modems designs, each tailored to a few bands, a programmable modem could be updated and adapted to handle more situations and protocols. While it looks great on paper, there are analog radio problems that are very difficult to work around, even with a programmable approach. That strategy completely makes sense for NVIDIA as it will allow the company to handle as much worldwide LTE coverage as possible with less engineering manpower. Qualcomm, which has plenty of manpower, will try providing more optimized multi-modem solutions which are tailor-made. This is a David vs. Goliath fight, although it’s not clear where Goliath’s weak point is…
High-Performance Computing has gone “GPU”
Another point that could substantiate this rumor is that graphic processors (GPUs) have become extremely important in super-computing (NVIDIA now powers Titan, the fastest supercomputer in the world), and are already making strides into the on-demand cloud computing market, with Amazon selling on-demand GPU-based computing power. At the moment, GPUs are the only type of ultra-dense compute chips that are commercially available. Why? Because they already have established commercial applications: graphics, CAD and gaming.
Graphics is what pays for all the R&D that go into the chip. Graphics is what makes it possible to build tens of millions of these chips, and therefore benefit from economies of scale. From there, GPUs can expand to general computing and other areas where massive parallelism makes a huge difference. Without building a good graphics processor, there is no way to have a financially viable high-volume, cost-effective ultra-dense compute chip. Graphics is indeed the best reason to acquire NVIDIA if such rumors ever came true — and the idea that Intel may one day replace its HD4000 internal GPU with a GeForce should make users around the world fantasize.
NVIDIA chips manufactured with Intel’s leading process
I bet that NVIDIA hardware architects would salivate at the idea of using Intel’s latest manufacturing processes, but in reality those edgy “fabs” would most likely be allocated to high-end CPUs which generate amazing gross margins for Intel. In hypothetical terms, if NVIDIA was provided the opportunity to design chips using that technology, it would be equivalent to jumping a few years in the future in terms of transistor density, etc… Nice, but dream on.
Conclusion
In the end, such an acquisition is in the realm of the “possible” and on the surface, there are compelling reasons for either scenarios (buy or not). But without knowing what the personal relationship between everyone involved is, it’s impossible to come up with an accurate prediction. There are a fair number of people who need to sign up on this, and my own take is that this seems a bit out of the comfort zone for Intel’s board, so I would be surprised if this would happen – but if it did, that would indeed be very exciting. I sure can’t say that buying NVIDIA would hurt Intel, on the contrary, but an acquisition would primarily be about Graphics and Computing and not really about Mobile in the near future.
What do you think?