The big news in semis yesterday was Nvidia announcing it was going to buy Mellanox for $6.9 billion. This deal has been percolating for a while. (We posted about it in early February). Which leads us to believe that this must have been a wild ride of an auction process. Somewhere a banker earned their bonus. The proxy should come out in a few days and we imagine it will make for an interesting read.
We think there are three key things to learn from this deal.
- The semi industry is continuing its consolidation unabated
- Nvidia is partly motivated by cost synergies which we think is the major unsung ‘hero’ of consolidation
- The market for networking semis is now pretty much sewn up
Let’s walk through these.
All the way back in 2015 we wrote about the shrinking of the semis markets. In that post, we posited that in ten years there will only be five chip companies left in the US and Europe. We owe you a post updating the list of who we think the five will be, its changed since then, and Nvidia now looks likely to be one of the survivors. This has had big implications for the industry. In October, we wrote about how consolidation in semis vendors has allowed the industry to raise prices, but that has in turn opened the door for big customers to once again start building their own chips in-house, something that had disappeared in the 00’s. Put simply, there is no room left for single market chip companies. Mellanox builds great products, but they were simply to small to survive independently. They are not the last company to face this.
This leads to the reasons Nvidia had for buying Mellanox. Typically, when chip companies look for M&A targets, a good place to start compiling a list is to look at the boards where their products appear in electronics. Then look at the chips next to it, and buy those companies. The buyer has a sales team and they can sell two chips almost as easily as just selling one. (Yes, we are oversimplifying, but not that much.) This is a revenue synergy story, growing revenues through acquisition. The more chips the buyer provides to that board the greater their ability to grow revenue and possibly also to increase prices. For Nvidia, Mellanox provides a broader footprint in selling into data center customers who are buying a lot of Nvidia GPUs for Machine Learning, but also need advanced Mellanox networking products to connect all those GPUs. And this synergy is the focus of Nvidia’s public justification of the deal.
However, there is another force driving deals – cost synergies. To tape out a chip on the leading process technology (7nm today) can cost $50 million, just to get to the point that a chip can get manufactured. After that there follows a multi-month process of improving the performance of the chip and increasing manufacturing yields. This process can cost tens of millions of dollars more. Add to this the cost of design tools and IP licenses, and the upfront costs of chip production quickly add up to big numbers. The more products a company has the more they can reduce those costs. They can amortize some of the upfront costs over more chips. That multi-month improvement process gets easier for follow-on chips as the operations teams can re-use their learnings. Companies also get better leverage in negotiating with IP licensors and software tools companies. This is not as simple a cost synergy as being able to fire 1,000 engineers (which is also sometimes possible), so it tends to get overlooked, but is a major factor in getting these deals to work.
It is important to note that this reality creates a cycle. (Whether virtuous or vicious is in the eye of the beholder.) Companies that have lower chip production costs get better margins which give them the cash to do more deals. Our point being that in semis consolidation drives more consolidation.
Which leads to our final point, there is not much left to buy in the networking space. Any company that is not already a major supplier of networking chips now has very few options left if they want to buy their way into the market. For those following the market, this is not news, once Avago bought Broadcom it was pretty clear the game was in its end stages. Mellanox was just the last sizable asset left in the space.
This is important because the pressure on the network continues to increase. Every time you read about data center growth or the explosion in cloud computing and machine learning, keep in mind that all of that heavy compute math requires extensive networking. That side of the data center is now essentially controlled by three companies (Broadcom/Avago, Intel and now Nvidia). This is a $5 billion ‘niche’ inside semis, and the only way to buy into that market is to buy one of those three companies, and no one can really afford to do that. If history is any guide, this means prices for these chips will go up, especially for the fastest, most advanced categories. It should not be lost on anyone that Nvidia’s shares were up heavily on the news.
Set against that tightening pool of suppliers (oligopoly is such a dirt word), is a shrinking pool of dominant customers – largely the ten or so leading data center builders, most of whom are cloud service vendors of some sort. As we pointed out in our previous notes on semis, these rising prices are going to push more of those customers into designing their own chips. Many are already doing so, and yesterday’s news is just going to expand the in-house chip teams’ mandate to look into networking parts.