Last week, private equity firm THL announced it plans to acquire AMI, a leader in firmware for compute devices, especially in the data center. When we first heard about this deal, we had to double check which AMI this was. There have been a few companies with similar names over the years, and who even thinks about firmware anymore? Not many people apparently.
Firmware is one of those obscure areas of compute that is simultaneously critical and also largely ignored. Not coincidentally, we have been doing a lot of work lately digging around such dark corners of the industry. And despite the ubiquitous nature of firmware, almost no one talks about it much.
Firmware is software that sits in between a processor and the operating system of a computer. It includes distinct categories like BIOS – connecting the processor to all the other elements of the system; Boot – feeding instructions to the processor to get started after powering on; and Security. AMI provides firmware for PCs, smartphones and most importantly for servers.
As we have discussed a few times recently (and here) servers come in all a whole range of variations. Every server design ends up being a little different – mixing and matching CPUs/GPUs/ASICs, memory, networking, storage and power components to an infinite degree. When the CPU powers up, it needs to know what hardware components it has to work with, and the firmware makes all this happen.
This is messy work. Firmware engineers need to understand all the low level primitives of all this hardware. Typically, they work with the processor designers years ahead of chip launches. This way, when chips start coming back from the fab, they can be configured into servers quickly. This is not the fancy design that the chip companies like to do. Nor is it cutting edge software which the hyperscalers like to do. It is critical, but no one gets very excited about it.
Enter AMI. AMI got its start designing firmware for PCs, back in the 1980’s and 1990’s this was a huge business because PCs were so new and the Windows ecosystem was evolving so rapidly. They made the transition to smartphones in the 2000’s. And in the past ten years, as cloud servers blossomed, they added that capability as well. However, they are essentially the only company that was able to stick all three of those landings. Their competitors fell away over the years, never making the move to new product lines. In general, many of the ODMs and OEMs can design basic firmware, but those solutions are often not robust enough for putting into production at scale. Today, almost any company that wants to design a server but does not want to write their own firmware needs to work with AMI.
And of course, this is happening just as data centers are getting ever more heterogeneous and complex. The hyperscalers are all designing their own CPUs and AI ASICs. Data centers are transitioning from CPU to GPU to accelerator. All this means that the work of building firmware is becoming harder and also more important. We have written about the difficulties in standing up scale GPU clusters, and part of that difficulty stems from the need for companies to deliver better firmware.
Dig a bit into server designs and the difficulties start to emerge and thus the opportunity for AMI.
Leave a Reply