NVIDIA is entering a key period for its performance in 2012 – the company is launching a large number of quad-core smartphones and tablets, first integrated baseband products and the rumors of the upcoming massive Kepler launch are gaining in strength.

Recently, NVIDIA reported their quarterly numbers for the fourth quarter of their fiscal 2012 which ended on January 29th. Thus, ahead of very important couple of weeks, we decided to take a look at NVIDIA financials and how are they going to cope with the string of product launches and how well the company is positioned to take a full advantage of the opportunities such as Qualcomm’s delay of its quad-cores by a quarter (to first quarter 2013 i.e. CES 2013) and recent picture showing that Apple iPad 3 and iPhone 5 are powered by a dual-core A5X chip.
First and foremost, in the fourth quarter they were able to earn $953.2 million revenue, which is a decline of 10.6% from the third quarter and up 7.5% from the same quarter in the last year.

In fiscal 2012 (February 1, 2011 until January 31, 2012) NVIDIA managed to earn $4 billion in revenues, which is a 12.8% sequential increase from their fiscal 2011. Their GAAP net income (profit) came in at $581.1 million ($734.4 million Non-GAAP), which translates to GAAP earnings per share of $0.94 ($1.19 Non-GAAP), which is an increase of 118.6% (46.9% Non-GAAP) over last year.

We took the time to take an in-depth look at the comments made at the conference call following the earnings release and present the highlights in the following paragraphs.

Game console revenue
Interestingly NVIDIA reported a seasonal decline in game console revenues. A few weeks ago AMD reported strong seasonality in that regard. This leads us to believe that it’s not actually driven by season, but a shift of customer interest. Sony PlayStation 3 console hasn’t performed well during the holiday season 2011, while AMD enjoyed good sales results thanks to Nintendo Wii and Microsoft Xbox 360.

Sony PlayStation 3 didn't had as much successful Holiday season as the company expected.
PlayStation 3 didn’t had as much successful Holiday season as both NVIDIA and Sony expected.

There is also a rumor that AMD scored deals of the next generation versions of Wii and Xbox, while it is not certain yet whether Sony will license NVIDIA technology again. Rumors on the PS4 side talk about the new PowerPC processor as the core and PowerVR graphics, a beefed up version of the graphics chip that powers PlayStation Vita, iPad 2, iPad 3 and many other smartphones and tablets.

28nm Process and Its Woes
The 28nm manufacturing issues effects on NVIDIA were the cornerstone topic of this conference call. Financial analysts were concerned about the adverse effects of it on the margin of the company. Upcoming line of GPUs codenamed Kepler is going to be manufactured entirely on 28nm. Thus their costs go up, since each new process comes with higher cost compared to the previous one in a more mature phase. However, to stay competitive companies like NVIDIA need to leverage leading edge nodes for better chip performance.

The problem with the low 28nm yields at TSMC is that NVIDIA typically pays on a per-wafer basis, which means their margins go down when the yield is bad. This should alleviate over the course of the year as the yields improve. According to NVIDIA the yield on 28nm is lower than expected. This is not to say the yield is bad overall. The CEO said that yields are actually better than they were at the same point compared to 40nm – which we all know was a disaster for all companies on that node. NVIDIA just projected them to be higher in the last year. Jen-Hsun also pointed out that they can’t solve the problem themselves. The company can only continue to work with TSMC to learn from experience as the process gets used. With a growing number of learning cycles, yield will improve gradually over the year? at least NVIDIA says so.

According to industry sources and re-iterated by Jen-Hsun, competitors using the same 28nm technology are in the same boat. NVIDIA just lined up with the comments made by AMD and Qualcomm on the yield situation. During a press briefing for the Radeon HD 7900 series, AMD explained that yields are "ok considering the process is in an early stage". AMD is a bit ahead of the curve when it comes to launching new node products (first to 55nm, 40nm and now 28nm), but their availability is not the best in all cases. The chips themselves work well and overclock well, so at least from a quality perspective the 28nm process seems to be ok.

28nm wafer with test print (ARM Cortex A-9 CPU IP) by GlobalFoundries. Taken in March 2010
28nm wafer with test print (ARM Cortex A-9 CPU IP) by GlobalFoundries. Taken in March 2010

Since there are a lot of companies moving to 28nm technology at the moment, NVIDIA can’t get all the wafers they would like to have. As the 28nm process is ramping this should ease over the year, but all the 28nm GPU launches from both AMD and NVIDIA in the first two quarters aren’t going to help the situation. According to our sources, Qualcomm is manufacturing certain SoCs at the 28nm node at both GlobalFoundries and TMSC.

However, as the GlobalFoundries capacity is constrained even with TSMC aggressively ramping 28nm technology, capacities are lower than the demand for them. Jen-Hsun Huang said that he expects 28nm capacity to be constrained throughout the year. This could lead to a bit higher prices for consumers in the end. At the very least prices are not going down in this situation.

It’s becoming evident that the market could use another leading edge foundry competing with TSMC. GlobalFoundries is poised to do so, but is not quite there yet. If it were, companies like NVIDIA could choose between different offerings and competition would ensure progress on both sides. Sure there is cost associated with going to a different foundry, i.e. the chip design would need to be adapted for different design libraries. But with some smart management these could easily be recouped by a much larger sales volume. Naturally, we’re not talking about 2012 since you need anywhere between 18 months and three years to execute a dual-foundry strategy. In case of NVIDIA, what is not helping is that the company is no longer considered a top notch job destination as there is a lot of restriction on the stock (we were told executive stock compensation no longer exists) and you effectively "work for salary." This lead to several highly rank executives leaving the company, but NVIDIA demonstrated the company wants to hire fresh talent, and not just in Silicon Valley. In his recent talk with NPR, Jen-Hsun discussed about the availability of Silicon Valley-type jobs worldwide.
Tegra business: Growing Strong
NVIDIAs Tegra business got a lot of attention from financial analysts and it will continue to do so. The company said that in this quarter we should expect the launch of Tegra 3-based superphones and tablets, which should be publicized on the Mobile World Congress in Barcelona starting next week.

NVIDIA Tegra 3 Die clearly shows the small amount of silicon occupied by the five ARM Cortex-A9 cores.
NVIDIA Tegra 3 Die clearly shows the small amount of silicon occupied by the five ARM Cortex-A9 cores.

The CEO almost slipped regarding an upcoming product from HTC featuring Tegra 3 but quickly said "a partner like HTC would be fabulous for NVIDIA to partner with". Naturally, that all came before companies such as Huawei, LG, ZTE and Fujitsu started pre-announcing their dual- and quad-core powered smartphones ahead of the show. On top of that Windows 8 should help them increase sales of their SoC aimed at tablets and superphones starting in the third quarter. However, WoA (Windows 8 on ARM) will not be the main growth driver for Tegra in 2012.

NVIDIA disclosed that Tegra 2 managed to earn $360 million, which is an impressive number for a business which just three years ago yielded with less than $50 million. For 2012 they expect at least 50% growth in this segment, which would increase revenues between "$400 and 600 million". Most of it will be Tegra 3 sales in this year. Tegra 3 also sports a bit higher ASPs than Tegra 2 and NVIDIA is confident that this situation will not change throughout 2012. One analyst noted that this is a lower projection than the company made in the past. The CEO explained that this is due to some wrong assumptions regarding Samsung, who are mainly using their own SoCs and only pick ones from NVIDIA if their own can’t serve their needs. The original projection assumed a higher share at Samsung.

Another topic was 1000 RMB (Chinese Yuan) phones. At current exchange rates this would equate to about $160. Jen-Hsun moved to explain that basically there are dual-core phones under development which should provide similar functionality to an iPhone 4 at this rather aggressive price point. As the moniker reveals, it is aimed at the Chinese market. For NVIDIA it is possible to cater to manufacturers of such phones with their Tegra 2 SoC in the near future. NVIDIA believes this is the fastest growing segment in the near future. Just announced ZTE Mimosa X definitely falls into this category. The only problem is that if ZTE reaches $160 price point covering a 4.3" qHD LCD, a dual-core processor and dual 5MP cameras enabling 1080p video conferencing – what’s the point in paying $600 for a smartphone?

NVIDIA continues to invest heavily in this segment to fuel continued growth. They are already more successful than initially believed. Jen-Hsun Huang explains to analysts that aggressive investments are necessary to leverage such opportunities. One area of investment is the development of a SoC integrating the Icera modem for the mainstream LTE market. It has been hinted at occasionally already, but it seems NVIDIA is getting serious on this one. For fiscal 2013 (calendar 2012), NVIDIA plans to launch a Tegra 3 Plus, i.e. T35. From the information leaked so far, the rumors go from a 28nm die-shrink with Icera LTE baseband, to a specialized, higher clocked version of Tegra 3 targeting "netbook" form factor with Windows 8. That won’t be known until the end of summer.

Regarding the tablet market, NVIDIA believes there is a space for tablets above a $199 Kindle Fire that are more general purpose. For example the current high-end ASUS Transformers Prime based on Tegra 3 shows, that such a product can be successful. For more cost-conscious customers the $249 tablet from ASUS sporting the same SoC is poised for success or "it’s going to sell like hot cakes" like the CEO puts it.

PC / HDD market
NVIDIA was one of the companies who had to take a considerable hit in sales due to the HDD shortage after the floods in Thailand. NVIDIA expects the shortage to continue throughout the first and possibly also the second quarter of 2012. In the fourth quarter of last year the shortage wasn’t as severe as projected by some since the supply chain was well-stocked with inventory that subsequently dried up. Consequently HDD prices surged which is bad for a company like NVIDIA, since a large part of their GPU business happens in the channel. The CEO specifically mentioned China as a country where a lot of PCs are built in the channel.

Regarding inventory buildup of GPUs, Jen-Hsun Huang noted, that companies are very careful and generally don’t load up more inventory as necessary. In general he believes the channel inventory is very lean across the board. The CEO also cleaned up a misconception among the analyst community, that more expensive HDDs lead to lower ASPs on the GPU side of things, to have a stable total BOM (bill of materials). Healthy attach rates show that this is not the case – people having demand for a GPU will continue to buy one regardless of the HDD pricing.

NVIDIA doesn’t expect the PC market to grow in 2012, but expects to grow their business by gaining share in the GPU business and growing their professional business. Especially in the notebook segment, NVIDIA is poised to take some share back from AMD. The CEO explained that they have many design wins at basically every single OEM with their upcoming Kepler-generation 28nm GPU, which will be paired up with new Ivy Bridge notebooks in the second quarter of 2012.

Professional sales: All eyes on Maximus and Tesla Halo Effect
In the professional segment, NVIDIA aims to expand their sales with their new Maximus technology. It boils down to selling an additional Tesla board along with a Quadro board for professional workstations, which accelerates certain computations in professional software (e.g. CAD, Rendering, etc.). The idea is that one GPU will be used for visualization, the other for simulation. NVIDIA claims that this provides more tangible performance increases than adding additional CPUs to the system. So the hidden rationale here is to shift money spent from CPU to GPU which is good for NVIDIA.

The other major driver for the professional segment is the launch of products based on the new Kepler GPU later this year. Not only will Kepler bring straight performance increases, NVIDIA also readies technologies to make it easier to program GPUs. OpenACC <http://www.openacc-standard.org> specifically gets mentioned, which surprisingly for NVIDIA is an open standard technology simplifying the definition of loops and other code regions to be offloaded on a GPU.

NVIDIA Tesla S2070 1U GPU server
NVIDIA Tesla S2070 1U GPU server

Even though the company does not want to say it, the 2012 will probably see sales of Tesla GPGPU boards exceeding the overall number of Teslas shipped from 2008 until the end of 2012 – as our internal sources are saying. The number of supercomputer design wins is helping the number of Teslas being sold, but the halo effect is yielding with greater results.

We heard about over two dozen universities ordering $15-25K system equipped with a single Quadro and two to three Tesla cards for 6-12 month evaluation before making a decision to build the next generation supercomputer. For example, we know a certain university that was so delighted with the initial results that after just four months, they shifted their next purchase from a 2000 CPU system into a 1000 GPU system, increasing the computational power. Moreover, what was zero revenue for NVIDIA and a neat revenue for Intel turned into a win for AMD, with their 16-core Opteron (not too fast, but serves well as a GPU feeder). The biggest winner is NVIDIA, since the company went from zero to $4.52 million dollar revenue at list price.

While $4.5 million is chunk a change in terms of overall numbers, the number of universities using the method above is moving at a rapid pace. Still, we feel that NVIDIA Tesla is missing a critical application that would drive it into Intel territory, which is raking billions of dollars of profits on powering cloud computing data centers.

The Bottom Line
According to public information, NVIDIA has no debt and $3.13 billion in "cash, cash equivalents and marketable securities." This is the amount of money after majority of costs of Icera acquisition have been accounted for. Taking a look at annual increase, one year ago NVIDIA had 2.49 billion in the bank – acquired Icera and increased the cash balance by $639 million. Furthermore, Intel paid its annual royalty fee – the part of the $1.5 billion settlement.

From the looks of it, NVIDIA has enough cash to develop Project Denver for 2013 launch as the desktop/notebook processor, as well as continue the investment in Maxwell – which slipped into 2013 at the same pace Kepler slipped from second half of 2011 into first half of 2012. Then again, Intel, AMD, Qualcomm – all companies are slipping on execution when it comes to advanced process nodes.

The $3.13 billion – compared to AMD’s $550 million cash and $2.0 billion debt – leaves the company in quite a comfortable position. We might add that the company has quite the open check book for further acquisitions, should they deem necessary.