When Goliath Wins: Who Benefits From LLMs?

As a venture capitalist, I’m in the business of investing in nimble startups that run circles around slow-moving legacy incumbents: betting on David in a battle vs. Goliath. Add a major technological shift like mobile or the cloud and David’s chances are even better to end the reign of the incumbent. 

But the technological shift du jour, Large Language Models (LLMs), isn’t working the same way as it so often has in the past. In a number of instances we’re seeing large companies leveraging LLMs to crush their smaller competitors. To understand why Goliath may win this time, we first need to understand why it’s usually a good idea to bet on David.

Why David Wins

Large legacy companies are typically late to adopt technological innovation. These companies move slowly because they can’t recruit top tier tech and product talent. PE-backed companies are even more appealing targets because PE firms usually focus on optimizing operations and reducing R&D. To a small team with strong product chops and amazing engineering, they’re sitting ducks.

Large tech companies (as opposed to large legacy companies) often have engineering talent but may still move slowly due to conflicting incentives. One of my favorite professors, the late Clay Christensen, coined the concept of the ‘Innovator’s Dilemma’: the idea that a market leader may not want to adopt an innovation because it would cannibalize their existing profitable business model. We’ve seen that dynamic occur repeatedly in tech over the years. 

Perhaps no company provides a better example than EastmanKodak, which had excellent digital camera technology but would not roll it out for fear of cannibalizing their existing film sales. Similarly, Tumblr was late to mobile, creating an opening for Instagram. Oracle was late to the cloud, which led to Salesforce stealing significant market share against its CRM product. As Clay pointed out, it’s not only weak engineering that can lead to David defeating Goliath.

Enter LLMs

As NFX’s crowdsourced market map shows, LLM startups are cropping up in every industry.  But we’re seeing many instances where the incumbents are better able to derive advantage than startups are. 

A great illustration of this is the host of startups that are applying LLMs to spreadsheets. It’s an excellent idea. But why will they win when Microsoft’s Excel will be powered by OpenAI and Alphabet’s Google Sheets will be powered by Bard? The short answer is that they probably won’t. Ditto for many other applications: Co-pilot is already integrated into Microsoft 365; ChatGPT is integrated into Notion, etc. In industry after industry, technologically capable incumbents have already found a way to leverage LLMs and bolster their market dominance. 

Which raises an interesting question: If tech-enabled incumbents weren’t very quick to adopt the cloud or mobile…Why is this time different? Why were tech-enabled incumbents able to incorporate LLMs into their products in a matter of months? 

The New World Order

First, unlike the start of cloud computing nearly twenty years ago, some of the biggest companies in the world today are themselves tech companies with agility and engineering talent. I love to back engineers out of Amazon because Amazon’s bar is high, they ship quickly and they’re well trained. But these were great engineers when they were at Amazon, not only after they left. 

Second is the ease of integrating LLMs: adopting cloud and mobile technology required extensive infrastructure setup, a daunting task for a mature company. LLMs, on the other hand, can be accessed through APIs with relative ease. As my partner Ramy Adeeb succinctly put it recently, “what used to be a massive science project is now few lines of code into existing workflows.”  

Third, the Microsofts, Amazons, Alphabets and Apples of the world don’t just have good engineering talent. They also have massive distribution via extensive user bases and robust marketing capabilities. They possess the means to swiftly reach millions if not billions of consumers. Killer distribution plus bleeding edge tech is a formidable combination. The future looks pretty bright for Goliath/Microsoft CEO Satya Nadella (pictured below courtesy of Midjourney).

​​

Too Good To Be Disruptive

For the final reason, we return to another of Clay Christensen’s insights. Clay defined innovations as either sustaining or disrupting, which meant it would bolster the incumbent or upend it. He differentiated between the two with a somewhat surprising test: sustaining technologies are technologies that improve product performance. Most large companies are adept at turning sustaining technology challenges into achievements. 

Disruptive technologies, meanwhile, are innovations that worsen product performance for customers–at least in the near term– discouraging the incumbents from exposing their existing customers to it. Mobile was too slow. The cloud was not secure. And electric cars suffered from a short range. Paradoxically, LLMs tend to be a sustaining technology because they work so well. (Maybe not for Google search - but we’ll save analysis that for another post.)  

On top of that, cloud and mobile often necessitated new business models, pricing, marketing and sales. It was an enormous lift and business model shift for a mature company to incorporate them - hence the Innovator’s Dilemma. LLMs, on the other hand, can enable new business models but are also easily incorporated into existing products to bolster the market leadership of incumbents. See: Microsoft Excel and Google Sheets. Unlike past technological shifts, LLMs were simply too good and too easy to adopt, enabling tech company Goliaths to succeed instead of startup Davids. 

Should Startups Pursue LLM Opportunities? Absolutely Yes!

The takeaways for startups here are twofold. First: choose your target carefully. While companies like Microsoft and Amazon can integrate LLMs with relative ease, large legacy companies in antiquated industries do not have the capability to do so. They lack both  engineering and fast decision-making, making them the same attractive targets they’ve always been.

Second, startups need to be thoughtful about their product. Try to find less obvious but equally powerful uses of the tech. This can mean focusing on less sexy markets rather than building the 50th competitor to Jasper. Or by connecting LLMs to workflows to make them more difficult to copy via an OpenAI plugin.

When it comes to LLMs, David won’t always beat Goliath. And startup founders have to be very strategic about which fights they pick. A slingshot only goes so far.

(Special thanks to my partner Ramy Adeeb for his invaluable feedback in writing this post.)