Venture funding for Deep Tech, Data Center Silicon and more.
Hardware will drive 60% of Tech revenue growth, but has only received 10% of venture funding.
Highlights from our blog
Data center silicon is a massive market, but prying entry is hard. Part of the problem many companies face entering is that sales cycles take a long time. Not as long as automotive, but still measured in years. Fortunately for aspiring entrants, the market is shifting away from CPU monoliths to a mix of CPUs, GPUs, networking, AI accelerators and FPGAs. This may prize the door open a bit. We look at the math behind this.
Global Foundries is essentially a mixed signal analog chip maker. Their future rests in their ability to convince a few of the analog IDMs to forego their next round of fab building and outsource to GloFo isntead. The transition to 300mm wafers could be a catalyst for this shift.
The world seems obsessed with AI, GPT and LLMs. As a friend put it, using Chat GPT today is like looking at the first Nokia phone in the 1990’s - it was the first mobile that the average person could appreciate, but looking at it then there was no way to predict where smartphones would end up. A lot of impressive technical achievements, but still a lot of work needs to be done to understand how to use it and what it really means. Before we get there, we have to endure a very frothy hype cycle.
Industry analysts predict that 60% of “Technology” revenue growth in the next five years will come from hardware, but over the past five years hardware only received 10% of venture funding. That bright AI future will need to be built on the back of some significant hardware investments. Now is the time to redress that imbalance.
Other News We Found Interesting
Semis and Hardware
Qualcomm launched a new series of chips for “IoT”. In a big break from past practices, these chips are essentially stand alone processors, something the company has typically not sold in the past. As much as “IoT’ is a worn-out term, Qualcomm has some pretty clear end markets here, and they seem to be trying to position themselves above the fray of low-end, cheap IoT devices, for more serious use cases in industrial and higher-end consumer devices. Probably most significant is they announced this launch with an army of partners. A few months back, we pointed out the lack of such partnerships in many of their past announcements. They are slowly building up that sales muscle.
The team at EdqeQ raised another $75 million to build their “base station on a chip” solution. They have a promising offering, but it will take a lot of funding to get there.
Intel held an analyst event for their data center products. The main news was that they are on track to deliver on their key manufacturing process improvements next year. They also announced that they are exiting their server design business. From a dollar perspective, this is tiny, but the unit they sold provides motherboard designs for Intel chip customers and as such was a useful sales tool for the company in the data center. It does not change things strategically, but shows Intel is cutting so deep it risks creating other problems for itself.
Intel also announced that they would provide foundry services for Arm chips. Once upon a time this would have been major news, Coke buys Pepsi, Cats and Dogs living together. Now the market is not sure what to make of it. Intel Foundry Services (IFS) is still years away. And Intel made similar announcements about supporting RISC V, a year ago, only to abandon those plans. They should support both, so either this is a deliberate snub of RISC V (which seems unlikely) or Intel is just trying out everything to see what sticks.
Mediatek made a fortune as the fast follower in mobile. Now they seem to be attempting something similar in automotive. The difference this time is that many of their prospective clients are busy designing their own chips. That being said, if US sanctions cut off China’s auto companies from accessing TSMC for their own chips, Mediatek will be the best positioned to capture the opportunity.
Google’s VCU invented a whole new category of chip. AMD has just launched a competitor, of sorts.
Silicon Carbide (SiC) is seeing surging demand from EV makers, but the market is still in early days. It reminds us of Gallium Nitride (GaN), another new material whose promise for semis has been touted for a decade. Building a supply chain for new materials takes time.
RISC V ‘leader’ SiFive is sometimes described as the “Red Hat of the RISC V world.” Unfortunately, Red Hat may not be the shining example people think it is.
Networking and the Cloud
A very detailed look at how AWS is trying to tackle the telco market. The conclusion is that a lot depends on how telcos approach the building out of software services into their networks. Microsoft’s Azure seems to be looking to provide a comprehensive “telco in a box” suite, while AWS is betting on just offering everything and letting telcos choose what they want.
A look at how Microsoft Azure built its AI infrastructure. Interesting perspective, but frustratingly light on numbers.
As we noted above, we are a bit cautious about GPT and the latest AI tools. Our take is that Chat GPT can write well, but no one is really comfortable letting it run totally loose, serious work still needs a human to re-read and edit it. The same holds true for this software developer using GPT-powered CoPilot to write code.
Semis stocks have been punished this year on fears of too much inventory and weakening demand. It looks like the pattern may be repeating in telco and networking equipment.
Load balancing is an important topic for networks. Web sites and their underlying infrastructure need to manage a lot of traffic. This space is a weird intersection of networks and software, which do not usually go well together. It is both a multi-billion dollar hardware market, but increasingly developers are writing software to accomplish the same task. The two approaches are not mutually exclusive, but it does highlight some deep engineering trade-offs.
Another primer on Large Language Models (LLMs). Fairly approachable but noticeably light on practical usage.
Other Items of Note
Lazard dropped the latest version of their very detailed analysis of the cost of energy. Fascinating reading, but a single summary slide at the start would have done wonders.
Why does an awful, plastic-wrapped sandwich cost $15 at New York airports.