As I tuned in to watch Nvidia's (NVDA) presentation to the financial analyst community, I was initially very impressed. The announcements surrounding stacked DRAM on the GPU helps to mitigate the principal bottleneck for extremely high compute throughput machines - memory bandwidth. This will pay dividends across the gamut of Nvidia's high parallel compute performance segments from HPC to PC gaming, and I believe that these advances will help to further widen the gap between integrated solutions and discrete graphics solutions in PCs.
I fundamentally believe Nvidia's GPU business will also see some fairly nice growth vis-a-vis its "GRID" initiative, although it is very difficult to try to model any sort of revenue/earnings there as this is a brand new segment. Management has no idea how well these things will sell, so it would be pure folly for me to attempt to try to make a guess without seeing any real sales data/trends. I do know that the strategy makes sense from a technical standpoint, and the value proposition is certainly there. The question is whether the GRID solutions can offer materially higher ROI for its intended segments than current client-based IT infrastructure. Nobody really knows, but Nvidia has signed on with large, well established OEMs such as Dell (DELL), HP (HPQ), and SuperMicro (SMCI), so if there is a market for these products, I fully expect Nvidia to be able to execute well and capture it.
The GPU business and its growth potential, particularly in HPC and cloud computing/visualization, is profitable enough to sustain the current share price, and believe that should Nvidia choose to spin off its "Tegra" division tomorrow, it would still be a compelling deep-value play.
Tegra Roadmap: Hit The Deadlines, Not The Benchmarks!
That being said, after the most recent presentation, I have to raise a few concerns about the outward presentation Nvidia's "Tegra" strategy. The firm announced that it would be investing $600M into its Tegra initiative during the year. At 50% gross margin (which Mr. Huang implied was what to expect for all of its businesses minus "Shield"), this implies that the business would need to realize a $1.2B/year run rate simply to break even let alone actually have anything to show for at the bottom line. I certainly believe that this is achievable, but I believe that Nvidia's focus on presenting an overly aggressive roadmap in an attempt to "woo" investors may not win them over as it did back in early 2011 when the stock was bid up to unrealistic levels on Project Denver/Windows RT hype:
Here is the roadmap presented at the event:
While Tegra 4 ("Wayne") is shown as hitting in 2013, I don't expect it to show up in devices until Q3/Q4 2013. Unfortunately, while Nvidia was championing that the Tegra 4 is faster than processor X and Y (raw speed in itself for these markets is actually a useless metric; performance at high efficiency per watt is what matters) Jensen admitted during the presentation that being late with the Tegra 4 cost the company some designs. It seems clear that from these statements that the rumors that Nvidia lost the Google (GOOG) Nexus 7 refresh socket to Qualcomm (QCOM) are likely true. Having the fastest/best processor in the world and a highly aggressive future roadmap doesn't matter if you don't meet deadlines, as Qualcomm and the rest of the 28nm gang learned back in 2011/2012 when Taiwan Semiconductor (TSM) had serious supply problems.
If "Logan" is a 28nm part but with a Kepler GPU swapped in, then this is the sort of safe, yearly update that minimizes risk and ensures that the company can actually get to market in a reasonable time frame. My worry is that this will actually be a 20nm part on TSMC's new, unproven process. This could lead to delays, supply problems, yield issues, and other risks in addition to the design headaches that the firm will likely face in swapping in the new GPU architecture. Jensen did not elaborate further on what process this chip would be built on or what foundries would be responsible for sourcing the wafers, so there is a real worry that the next generation chip could run into "unforeseen" difficulties that allow another vendor to steal the design wins. A 3 month delay, as we saw with Tegra 4, was enough to cost Nvidia design wins.
Even more worrisome is the claim that the 2015 part will be built with FinFETs. Given that the 20nm process at both TSMC and GloFo will just be ramping volume production in 2014 and into 2015, it is very difficult to believe these foundries' claims that by 2015 they will be able to ramp the FinFET hybrid 16nm/20nm (16nm FinFET front end, 20nm back end) into volume production. In fact, Dr. Morris Chang, CEO of TSMC, noted on the most recent call that 2014 would be the first year that we see volume 20nm production and that 2015 would be the second...would customers really want to use the 20nm process in 2015 if the 16nm/20nm hybrid FinFET were also available? Expecting 16nm/14nm of any flavor in volume from the foundries during 2015 is setting oneself up for disaster, and the fact that Nvidia is making the claim that its 2015 part will be a FinFET part (when TSMC hasn't even promised availability of this node) seems completely unrealistic and risky. This process risk comes on top of the risk that will be introduced with a custom CPU core as well as a brand new GPU architecture.
The key to winning is to get into the right designs on time, not in PowerPoint "manhood" measurement contests with Qualcomm or Intel.
Nvidia Can Differentiate On Software, Not Hardware Chest Puffing
The key to this business is to win the right designs, and I believe that Nvidia can win on differentiation elsewhere besides touting the fastest CPU benchmarks (hint: Nvidia is a visual computing company). Things such as "Tegra Zone", and even "Project Shield" to help build brand awareness and a connection to gamers on these platforms will ultimately be the differentiators. Nvidia has often touted itself as a "software company" first and foremost, and I believe that despite Windows RT's failure, the firm's ability to out-execute Qualcomm in developing a proper software stack and score the design speaks volumes. These efforts should probably be more targeted towards Windows Phone, Android, and BlackBerry (BBRY) OS going forward, especially as Nvidia pushes into the smartphone arena.
Further, Nvidia can leverage its relationship with game developers in a similar manner as it currently does on the PC and even with Tegra Zone/Shield to position Nvidia-powered chips as the gamer's choice for mobile processors. The brand equity here is much stronger with consumers than the generic Chinese SoC vendors such as Allwinner and MediaTek, and even stronger than that of Qualcomm. Nvidia just needs to push here with all of its resources to win that "gaming" mindshare. This business with the "hardware chest puffing" by sticking in an unnecessary number of power-hungry Cortex A15s will go nowhere fast as this type of CPU performance comes at the opportunity cost of GPU performance which is arguably much more tangible for most "intense" workloads on a phone (games).
Financial Uncertainty Likely To Pressure Shares
It does not seem particularly clear that Nvidia will be able to narrow the losses due to its Tegra investment, let alone breakeven. 1H 2013 will admittedly be weak for Tegra, and while management seems optimistic that 2H 2013 should be better, we still have very little clarity into how this heated 2H momentum will translate into sales. I do not expect Windows RT based designs to sell particularly well, especially as Intel's "Bay Trail" (22nm Atom) will be shipping in volume by then. The Android space will essentially be a 3-way deathmatch between Intel's "Bay Trail", Qualcomm's Snapdragon 800, and the Tegra 4. The quality and popularity of the design wins will be the key metric to determine the "winner" here, and so I do not blame management for keeping quiet on full year predictions here. The company cited "more" design wins than Tegra 3, but this means nothing if the designs do not sell well.
Ultimately, investors need to wait until 2014 for Nvidia's Tegra strategy to get to the phones if they are to see the kinds of volumes necessary to bring the division to breakeven, let alone profitability.
I am on-board with and encouraged by the GPU business, but I am dubious as to whether Nvidia's focus is in the right place on Tegra. While winning benchmarks is nice, the key is to do so in a power efficient manner. It has been demonstrated that even a dual core Cortex A15 in the Samsung Exynos draws far too much power for a phone and barely fits in a fanless tablet. A quad core Cortex A15 design without a material process shrink is unlikely to be able to post the required numbers in a battery constrained environment.
Nvidia's broad strategy for Tegra is to try to realize ARM's (ARMH) dream of a "post PC era" which really translates into the "post Windows" era. Windows RT is a flop as it merely a crippled Windows 8 (no legacy software compatibility), but in the Android space Nvidia can make a real push if Android clamshells and tablets become the future mainstream PC. While Nvidia waits for this to happen, it needs to start winning major smartphone designs to realize the volumes necessary to bring this division to profitability. The firm is well on its way, but I hope that the focus continues to be on getting products done, on time, and in the right high volume devices. Winning benchmarks is nice, but at the end of the day, it doesn't matter if you don't show up to the fight on time.