Where to Build in the Semiconductor Space: Charting Paths in the Era of AI
Insights for aspiring founders in the semiconductor industry
With the current AI boom, there is newfound attention towards semiconductors. I’ve never seen my worlds collide in such a way as it is now. Having spent years researching global chipmakers and their supply chains—Nvidia, Intel, AMD, TSMC, Dell, and Apple among them—I’ve witnessed a shift from then to now as a generalist early-stage startup investor. Very few people I knew outside of my job back then knew what Nvidia was or could even pronounce it. Fast forward seven years later in what feels like an opposite world to me, and everyone is tuning into Nvidia’s earnings calls and conferences (even nicknaming their annual GTC conference, which has been around since 2009, to “AI Woodstock”). It’s been remarkable to see how the new and surging demand for semiconductors has inspired ideas among new founders venturing into the realm of AI.
With heightened focus on any sector comes a wave of startups aiming to reinvent it. There is a lot of debate regarding new AI startups and if they can carve out their own niches as major players or if they will merely become features for existing industry incumbents. I’ve seen this debate across many different verticals but I’ve also started to think about it more in the semiconductor space itself, with some ambitious founders even aspiring to rival the established chipmakers and manufacturers directly.
Hence, I am writing to share the three different avenues I’ve seen aspiring founders focus on so far, and which of these areas I believe deserves more attention.
Creating a new chipmaker altogether (competing with Nvidia, AMD, Intel directly).
Creating a work productivity tool to help the incumbent chipmakers save time and focus on their goals.
Creating a way to increase GPU capacity while using less energy.
—
1. Becoming the next big chipmaker.
Directly competing with established chipmakers demands a substantial investment of both time and capital. If you sacrifice either of these things, then you sacrifice quality and consequently, your customers. While Sam Altman's ambitious goal to raise trillions of dollars for a new chip venture seems extreme, it underscores the monumental scale of commitment required. Even with prominent figures like Altman involved, the path remains challenging. I’ve seen what a massive incumbent like Intel faces in this incredibly intricate and unforgiving industry, repeatedly striving to compete against rivals while maintaining customer relationships. Moreover, unless you're prepared to build your own manufacturing facility (which could take over half a decade and requires billions of dollars), you'll fight for production capacity at TSMC, where competition, especially from established players like Nvidia, intensifies. If you try to seek alternatives to TSMC, then you likely won’t get access to the leading-edge equipment essential for maintaining competitiveness.
Globalization is also mandatory if you wish to be a leading player. Some of the finest chipmakers, though not U.S.-based, cater to American clientele. Becoming a leading chipmaker means you are immediately competing with the best companies worldwide.
Truthfully, I think it will take a lot to compete directly in this space. Usually when you hear the word “incumbent” you think of a slow-moving company that is unable to innovate quickly which is why startups are motivated to jump in and steal some market share. However, that’s the opposite case for an incumbent semiconductor company. These companies have been beating Moore’s Law for years, consumed with making their products smaller and cheaper. I don’t know if a startup with less credibility, less capital, and less time will be able to easily jump in and catch up. (Maybe when Moore’s Law truly ends I’ll come back to this.) The incumbents are all well aware of the current supply < demand situation and are strategizing now on how to increase capacity.
2. Shooting for “Light-Tech.”
Enter "Light-Tech"—a term I'm using to describe the software solutions that aim to redefine traditional workflows for chipmakers. While established chipmakers focus on creating cutting-edge technology, their processes remain laborious and legacy-bound, creating an opportunity to sell AI productivity tools.
However, even for Light-Tech companies, there is a big challenge ahead. Your early adopters would predominantly be the largest chipmakers in the world. There are no smaller chipmakers worth targeting as customers, and even if you decide to go after them first, it would not be indicative to the sales process or needs of the larger incumbents. Therefore, you will need to begin your company by putting your MVP in the hands of your dream customers.
To achieve this, several crucial factors come into play:
The founders must possess the right background, experience, and network to sell to these major ruthless customers. You will need the reputation and credibility to earn their trust and deliver flawless software solutions.
The value prop for these initial pilots can’t be small. While your software may not directly contribute to generating new revenue, your sales pitch should align with their mission to build and innovate quickly and flawlessly.
You must also think ahead in terms of onboarding and integration processes. You don’t want to break anything at these companies because they can be reluctant to change their well-oiled machines. Your product has to be worth it and has to be seamless, at least initially.
Lastly your business model, more specifically your ACV. If you are going after the largest players in the beginning, then consequently there’s no where to go but down. There’s a small number of players but they are huge. You will need to think of ways to expand your ACV within these customers and upsell on other products. If you want to build a venture-backed startup, then your eye needs to remain on becoming a $100M ARR business in 5-7 years. It’s counterintuitive but that goal can become more difficult if you are starting off working with your largest customers.
However, I don’t think Light-Tech is solving the most urgent problems faced by chipmakers or their customers. There’s no “hair-on-fire” problem here.
3. Solving the problems that AI has exacerbated.
This last avenue is what I feel is most important.
The most dire problem we’re currently facing in the semiconductor space is AI itself. Its popularity has caused the need for exponentially more GPU capacity and that means more infrastructure and more energy requirements. This is an issue from a supply/demand perspective but also a climate perspective. Additionally, startups currently have to pay the high costs of GCP, Azure, or AWS for their cloud needs which can be debilitating. There are companies trying to solve for capacity needs by creating new server or GPU farms, but not only is that a tough operating model for a startup, it doesn’t solve the issue of energy usage. I’ve seen other founders making progress in this space. For example, creating a network of remote GPUs has been a thoughtful thesis, taking into account capacity needs, costs, and energy usage by increasing utilization of existing GPUs.
Overall, I think this is the hair-on-fire problem emerging founders can focus on in the semiconductor space. It would be solving the problems for smaller AI companies that can’t afford their computing and data costs, as well as semiconductor incumbents that are trying to increase their capacity output and conserve energy. This area still has a huge uphill battle but it is a battle worth fighting.