Semiconductor Business Intelligence

SemiVision : To understand #NVIDIA’s direction in silicon photonics, we must first analyze #TSMC’s progress in this field. The reason is straightforward—NVIDIA is actively collaborating with TSMC. As a semiconductor giant, TSMC’s broad range of process nodes allows it to leverage mature processes for PIC (#Photonic Integrated Circuit) manufacturing.

Competing with what is visible is not a novel approach. It is based on the fallacy that your competitors sit with their feet on the table while you work hard to beat their two-year-old product. This is the standard approach in Semiconductors, but we know by now that Nvidia is far from standard. While all the Hyperscalers are Nvidia customers, they also desperately try to displace Nvidia while the AI leader's profit margins taunt them. Nvidia has recognised this and lists all the Hyperscalers as competitors in its financial reports, but it behaves as unworried as an Englishman in a queue. It is easy to make slides, but we know by now that Nvidia delivers. If you want to know more about the business case of the Hyperscalers Accelerators attempting to dislodge Nvidia, read more here: https://lnkd.in/dNK2Xknz

Being a Cloud Oligarch is not easy when all your hard-earned billions are wasted on making another Oligarch rich. We are talking real money here—the quarterly CapEx of the top 20 Cloud Owners is just over $100B$, while Nvidia bags close to $25B$/Qtr alone. The Cloud owners are not on an escape path at the moment, as Nvidia's profit growth follows the growth rate of the Cloud Capex. You can be sure that they enjoy handing over the cash as much as a Scottish pawnshop owner and would avoid it if they could. While the large cloud companies have all launched their architectural competitors, the designs take time, and the supply chain is busy with Blackwell orders. The recent revelations from Nvidia show that the AI giant has not rested as the Cloud companies launched the challengers. If you want to understand the business case of the competing architectures, you can read more here: https://lnkd.in/dNK2Xknz

One of Nvidia's masterstrokes is creating Nvidia-branded servers. Not only is Nvidia profiting from the silicon but also from the whole hodge-podge of components that make up a server. It is irrelevant to try to assign profits to each part as they are costs—only the complete server is revenue. While Nvidia's gross profit margins are in the high seventies, irrespective of the channel, it is slightly more effective for the hyperscalers to buy servers from the partners, as seen from a materials perspective. Not only is Nvidia in control of its own server business, but the AI giant ultimately decides both the market shares and the profits of the partners, which is why it makes sense for its customers to design alternatives. While the XPU architectures are commercially attractive, they are also difficult to pull off. A design takes a couple of years with the aid of a semiconductor enabler and requires significant investments in the server supply chain, which is also dominated by Nvidia. Read more about the business model of the XPUs here: https://lnkd.in/dNK2Xknz #GTC #Nvidia #AI

We get asked about this a lot, so I thought I'd share the updated vendor ecosystem map for the NVIDIA MGX rack reference design. This summary is high level and of course Shen W., Manoj Sukumaran and I can help you with a much deeper competitive, architectural, or market size understanding. This ecosystem will continue to grow and this snapshot is illustrative rather than exhaustive. #gtc25 Omdia #rackscale #NVIDIA #NVL72

It has not escaped anybody's attention that Nvidia is capturing a nice share of profits from the Server market, especially the customers living in the shadows of the Nvidia AI Death Star. While Cloud companies are free to choose to buy AI servers from Nvidia or one of their partners, Nvidia's control is absolute. Nvidia's revenue model is complex, but the end result is simple. Margins and market shares are decided at internal Nvidia planning meetings. As a result, Cloud companies have started their rebellion and launched their X-Wing fighters pursuing their proprietary architectures. This has made the relationship between the cloud hyperscalers and Nvidia incredibly complex. On the one hand, cloud companies are doling billions of dollars to Nvidia shareholders as customers while being complemented as competitors in Nvidia's annual report. We dive into the business model of XPUs as an alternative to Nvidia GPUs (while looking for an air vent). Read more here: https://lnkd.in/dNK2Xknz

For the last 5 years, the Semiconductor industry has been in the Post-Logic Era. Capacity investment decisions used to be based on analysis of supply and demand while keeping an eye on the investments of competitors. Fab locations were selected based on the cost and availability of skilled talent and a solid supply chain infrastructure. All of that went down the toilet when politicians got involved. Embargo this, embargo that, tariffs up, tariffs down. Subsidies are available everywhere. While the possible end of the Chips Act can help the industry return to normalcy, we are still in the post-Logic era. Withdrawing the Chips Act funding will hurt US companies more than foreign companies. If your business depends on the Semiconductor Industry, you can follow my blog here: https://lnkd.in/d2ZXMHjw

Becoming uncorporated is tough and scary, but so much more fun than the alternative.

TechInsights compiles quarterly financial data for a sampling of public Tier 1 companies important to the automotive industry, including Aptiv, BorgWarner, Continental, Denso, Forvia, Hyundai Mobis, Lear, Magna, Schaeffler Automotive Tech (acquired Vitesco in October 2024), Valeo, and Visteon. In our sampling, the average Tier 1 supplier's total revenues rose 2% in the fourth quarter above the third quarter.  Gross margins at Tier 1 companies averaged a percentage point higher in Q4 2024 over the prior quarter, while the average operating margin fell in the fourth quarter, dropping two percentage points to 5%. Read more in the full report https://lnkd.in/gBzUM_Ef

While TSMC's February revenue was down, it carried most of the Lunar New Year celebrations. Judging TSMC's performance in this period can be challenging unless the average monthly revenue is used. The average monthly revenue showed that TSMC entered the year of the Snake easily. The average monthly growth was 2.6%, which confirms that TSMC's annual growth rate is currently above 30%. https://lnkd.in/d2ZXMHjw