MOST PEOPLE were amazed by Nvidia’s Q409 results. How could it reap that much profit when it doesn’t have any competitive high-end GPU products on the market? The answer is right in Nvidia’s statements and the key is to understand what supply constraint means.
A supply constrained market is an elegant designation that economists give to a simple phenomenon: demand exceeds supply, so there are not enough goods for everyone and you will sell for a high price everything you have in inventory and can manufacture. Also, in a supply constrained market there is no way to quickly increase capacity, so market shares tend to freeze and the fight changes to who will rake in more profits.
What happens is that while the total volume of goods produced remains more or less the same, the amount paid for those goods rises. Keep that in mind.
If you want to buy GPU chips in any meaningful quantity, you have to go to Intel, AMD or Nvidia. According to Mercury Research data from Q409, Intel dominates the low performance and high volume segment of the GPU market with 50 percent of total GPU unit sales.
Source: Mercury Research
Intel’s GPUs are synonymous with mediocre performance and very low costs. But the performance of its mediocre GPUs is steadily increasing with each generation and the chips are incorporating more and more features. And as the cheap integrated graphics processors (IGPs) grow in performance, fewer people need discrete graphic cards. This is where the market grows in volume and here supply capacity is less critical, due to the huge manufacturing capabilities of Intel.
If you want high performance GPUs you have to go to AMD’s ATI Division or NVIDIA. Between them they produce virtually all of the beloved high performance discrete graphic cards for both notebooks and desktops. Their combined market share in those two market segments is 98 percent. However, here the problem with supply is that all of the GPU chips used by AMD and Nvidia come from a single source – Taiwan Semiconductor Manufacturing Company (TSMC).
TSMC was the source of the supply capacity constraint experienced by AMD and NVIDIA last quarter. Simply put, TSMC could not ship enough chips out of its factories to both graphics chip vendors. To make matters worse, a surge in demand for discrete GPU chips in Q409 caught TSMC, AMD and Nvidia off guard.
But demand explains only part of the problem that affected AMD and Nvidia’s results in Q409. The other part of the problem is probably due to overreaction by TSMC to the 2008 world financial crisis. According to TSMC’s financial statements, the company cut its Capital Expenditure (CAPEX) in property, plants and equipment heavily in 2008 and this is likely one of the causes of its supply constraint that surfaced near the end of 2009. To that add the hurdles and growing pains that TSMC experienced in ramping up its 40nm chip fab process and then you have the situation experienced in Q409.
In Q409 Nvidia was fielding its first 40nm chips, the G215, G216, G218 and G220. Those are relatively small chips and so are inherently less vulnerable to low wafer yield prolems than bigger chips such as AMD’s Cypress and Juniper or Nvidia’s G200b and Fermi. Nvidia could have sold a lot more of those chips had the capacity been available but they are aimed at the low and mainstream segments, which have bigger volumes but smaller margins than the performance and high-end segments.
On the other hand, AMD was also fielding its 40nm chips too, but they were not that small. Cypress was far larger than Nvidia’s 40nm chips and was also a high-end chip. It was intended to outperform Nvidia’s G200b and it has been quite successful in doing that. Cypress was soon followed by Juniper, a performance chip.
The problem for AMD was that Cypress and Juniper were heavily affected by TSMC’s 40nm woes, way more than Nvidia’s G21x chips were. Those two chips are more cost-effective than Nvidia’s 55nm high performance chips and with them, AMD could have wiped the floor with Nvidia in the high-end and performance segments had the manufacturing capacity been available. So in a sense, the supply constraint that Nvidia complained about on its quarterly conference call also protected its market share and inventory value.
We can also see in Mercury Research data that Nvidia’s market share grew a bit and its revenues in the desktop segment also grew a bit. AMD lost some market share in some segments and kept market share in others. But let’s look at what supply constraint can do to market prices. The volume numbers are from Mercury Research and the sales numbers are straight from the quarterly reports of Nvidia and AMD.
Regardless of the rise in the quantity of chips sold, Nvidia got a smaller share of the overall revenues split between Nvidia and AMD.
For AMD the supply constraint effectively prevented it from taking more market share from Nvidia than it did, about 2.9 percent in Q4, but it allowed AMD to sell its chips in Q4 for prices higher than Q3 prices, so that is the reason for the leap in AMD’s revenues, a 40 percent jump quarter on quarter. This is because it dominated the high price and high margin market segments with Cypress and Juniper.
Supply constraints or no supply constraints, Nvidia’s product mix worsened when compared to AMD product mix in Q4 and in Q1 the picture will be even worse. If its product mix worsened compared to competitors, in normal market conditions and without supply constraint, it would lose some market share. With supply constraint, the amount paid per part would rise, or at least not decline, as would its margins. But instead, firm prices compensated for part of its lost revenues in the high end market and kept its average selling price and margin per chip more or less stable.
So this is what happened. A combination of strong demand, limited manufacturing capacity and screwups by TSMC granted some relatively good results to Nvidia.
Are those results sustainable? No. In Q1 it will face tougher competition from AMD that will mean depressed margins unless there is another increase in demand. By the end of 2010 and in 2011 some changes will happen on the GPU market that will definitely change the landscape. Nvidia’s first challenge is Fermi, which is not a mere graphics chip, but rather is a bet in a new business model.S|A