NVIDIA and TSMC announced this week that the Taiwan Semiconductor Manufacturing Corporation will remain NVIDIA’s primary source of graphics processors for the foreseeable future, however, analysts still predicted that TSMC’s revenue will still be affected strongly by strategic alliance between IBM Microelectronics and NVIDIA.

It is obvious that any foundry agreements between TSMC’s clients and other contract makers of semiconductors impact the Taiwanese chipmaker sales, nevertheless, we should not forget that despite of loads of difficulties and obstacles semiconductor market has been growing rapidly for the last ten years. Maybe the pace now is not as rapid as on the dawn in the seventies and the eighties, but it is still quite fast. Remember that NVIDIA always says that this year will be a great year for graphics chips, hence, based on this, we should expect growth of revenues for all three companies: TSMC, IBM and NVIDIA itself.

Let us try to analyse TSMC’s prospects of loosing substantial part of NVIDIA’s orders.

Manufacturing costs of every single GPU depend on the cost of a wafer and the yield. Given that IBM should charge more per wafer than TSMC, it becomes clear that NVIDIA only can get economical advantage from moving to IBM in case it produces very complex processors which yield is either considerably lower at TSMC or NVIDIA simply cannot ramp up the production quickly at the Taiwanese foundry.

There is also a compromise between GPU complexity and acceptable yields in order to keep the costs at low level. Since all makers of semiconductors constantly improve their manufacturing technologies they will always be able to manufacture very intricate semiconductors, but some manufacturers will be able to do that earlier than the others. Revenues are driven by mainstream products, not the high-end ones, so, rapidly expanding computer market will help to boost revenues of the firms that produce mainstream chips. As for the companies that are able to produce higher-end solutions, they will enjoy higher margins, but are not likely to expand the share and revenues really quickly (especially keeping in mind that Asia, where only the price matters, is one of the most influential driving force of the industry).

TSMC makes 90 thousands of wafers per quarter for NVIDIA. It is estimated that TSMC will produce 95% of NVIDIA’s wafers this year, so, 5% that will be made at East Fishkill, New York, are likely to be the most high-end offerings from the Santa Clara, California based graphics company. These expensive GPUs will be the upcoming NV35 and probably  the NV40 (if it really comes out this year) as well as professional Quadro-series based on the mentioned GPUs. The report over here states that TSMC could lose $7.5-to-$11 million per quarter to IBM in 2003. That figure could go up to $35-to-$45 million per quarter in 2004 as a result of the IBM-NVIDIA foundry deal.

Unfortunately, it is not said why the figure is to go up from 3 to 6 times next year. Will IBM manufacture more than 5% of NVIDIA’s wafers? Or the market of discrete graphics cards will grow 3 to 6 times in 12 months? Or maybe the overall market share of discrete GPUs will shrink very considerably and there will be more space for high-end 3D graphics solutions than now because more customers will go with expensive devices since inexpensive products will have to compete with advanced integrated sore-logic sets (for instance, now the GeForce4 MX competes with the nForce2 in some cases, while ATI’s RADEON 9000/9200 will rivall ATI’s own chipsets for mainboards)? No idea, to tell you the truth, but let us consider the chances for IBM to gain its 5% of NVIDIA’s products towards 15% or even more next year.

15% to 20% of production is quite possible in some cases. For instance, in the fourth quarter of 2002 more than 50 million of graphics devices were shipped into the marketplace. Roughly 9 million of them were high-end graphics cards, such as ATI’s RADEON 9500 and 9700-series, Matrox Parhelia-series, NVIDIA’s GeForce4 Ti-series and also some professional products from companies like 3Dlabs, HP and SGI, according to Jon Peddie Research. The GeForce4 Ti appeared to be a very great seller and I believe that the biggest share of the higher-end market belonged to NVIDIA last year in general and in the Q4 in particular. In case the trend continues, TSMC can really lose significant part of its revenue unless the market recovers and especially the market of mainstream products recovers and starts to rapidly gain momentum. Until now everything looks bright for IBM and NVIDIA, but we should keep in mind a very important detail: the GeForce4 Ti family consists of numerous products that are all based on the same chip. For instance, the GeForce4 Ti 4200 for $150 utilises the same GPU as the GeForce4 Ti 4600 for $399. It is clear that the GPU which powers so wide range of devices is manufactured in vast quantities and brings its maker and developer huge revenues. But what do we see in the marketplace now? We see that NVIDIA has a number of GPUs for different market segments: the GeForce FX 5200, 5600 and 5800. The latter is the most powerful and the hardest to produce. In case the trend to have different chips for high-end, mainstream and low-end market segments continues, we will hardly see IBM manufacturing 15 to 20% of NVIDIA’s wafers, unless NVIDIA sacrifices its desired 35 to 40% gross margins.