In his famous book, the 7 Powers, Hamilton Helmer divulged the 7 Powers as a means to achieve persistent returns in business by laying out a strategy. The S-curve states that for a given market that deems itself sustainable it must undergo 3 main Phases:
1.Phase I: Innovation — for a company to create Power it must bring an innovative product to market.
At this phase, price is indicative of the time/money expense/contribution the product allows. Put simply, if you previously used an abacus to calculate, and I just invented the calculator, it is justifiable of me to set a price equal to the time/money I saved you. The intuition being you’re prepared to pay up to the opportunity cost I’m saving you.
Powers: You innovate by counter-positioning or providing a cornered resource(like IP).
Info: There’s just you, the inventor. You set the price.
Fun fact: everybody underestimates market size here. We thought the car industry would be a $100M industry. It’s really $2.86 Trillion!
Graph for cars made since 1900. Image by Darrin Qualmann
2.Phase II: The Bottleneck is Squeezing. You have increasing competition. There’s money to be made.
In the 2nd phase, competition comes in and says: “Hold on, I can do this cheaper!”. Incumbent products bring operational expenses down. As they increase volume to fulfill the growing customer base, their scale economics improve. Phase II is also where differentiation and niche creation happens.
Powers: Scale Economies, Network Economies, and Switching Costs
Info: We say that: “competition drives prices down”.
Fun fact: Apple and IBM set a price war in the 1980s, making PCs ever more affordable, from IBM’s PC 5150’s price of $8.532(inflation-adjusted) to a little over $1k in 2023 for an Apple Air.
3. Phase III: From Lower to Lowest to None. Eventually, increasing competition drives prices to equal marginal cost.
Powers: Process Power and Branding
Info: In a perfectly competitive environment, the price equals the marginal cost of making the product, i.e. zero profits. In reality, there are naturally induced barriers that prevent perfect competition from happening, like switching costs, branding, initial capital costs, etc.
Fun fact: General Motors tried to replicate Toyota’s operations in the early 2000s but made no advances, while Toyota seized market share from 4.1% to 20.7% in gross sales. That’s process power — operational excellence!
LLMs are a technology that has been boiling in the frog water for over 50 years but took the world by storm in 5 days. The S-Curve can help us better understand the natural development of Large Language Models(LLMs).
Following the 3 Phases of the S-Curve, I believe we’re entering Phase 2 of LLM development — The Bottleneck.
Phase I: LLM Innovation
OpenAI is the definitive disruptor of the LLM technology. Its ChatGPT has set the scene for what would become “language automation”.
Price: you see, their strategy value lies in spanning verticals such as bio, fintech, SaaS, and EdTech, with powerful generative capabilities, i.e. content creation, customer support, writing code, etc. They could set any price in the world if people would accept it. Why the $20 GPT-4 Pricing Tier? So that it would be available to mass-market consumers. If you previously wrote a school essay in 3 hours, and ChatGPT does it in 15 minutes, you’re willing to pay.
Powers: Cornered resource(the proprietary technology) and the Counter-Positioning — nobody made LLMs accessible to mass consumers easily before.
We’re here now! ChatGPT set the ground for what will become persistent differential margins for the innovator(OpenAI). The S-Curve bottleneck is rather wide because of:
Network Economics — viral product effects + increasing verticals positioned OpenAI as the clear leader in LLM development. Additionally, they are following in the footsteps of Apple in creating a product ecosystem with its GPT-4 Stores, plugins, and applications. CaC/LTV is
Scale Economies — OpenAI attracted and incentivized ($1M) the best AI engineers in the LLM world to build the future there. As a result of the category-defining product, OpenAI can do more iteration, customer feedback looping, and product improvements than any other incumbents out there. Additionally, they have the best unit economics per value-add per customer, and scale economies work in their favor. Everyone else is just burning $10Ms of cash on models, teams, and marketing to meet the same kind of demand. Talent, investors, and users all want to follow the new “sheriff in town”.
Image by Author
Phase III: This is Where We’re Headed
Normal S-curve transition times, i.e time from Phase I to Phase III, are significantly longer than we will encounter for LLM development. Here’s one for the car industry. It took 23 years for 100M people to own cars, and it took LLMs 18 months to reach 300M users.
Process Power & Branding— as OpenAI seizes and expands to more customer volume, they can expect to seize market share indicative of monopoly. Their upscale into enterprise with Microsoft as the strategic partners means delving into a $374B market ripe for disruption of processes.
I would argue that LLMs are going to have to go through their own political, social, and commercial evolutions. The way in which LLM teams acquire and train data will not ultimately yield better results for queries. Rather, fine-tuning organisational-level models will replace generalised models.
Phase III of LLMs will be significant by the following events:
OpenAI & Microsoft are looking to sell enterprise. BIG time!
A new infrastructure layer will start to emerge to support the business’ commercial efforts. I am referring to LLM prompt injection, inference call stack, chatbots, fine-tuning, dataset operations, etc.
OpenAI will continue to expand its product ecosystem with GPT-4 Stores, and product integrations. The company will continue to hedge its edge in talent, investors, and customer volume as another move to monopoly.
Continued pursuit for improved context window length(256k+)
Open-source LLM models will allow every company to fine-tune and deploy LLMs far more easily.
New startups will incorporate LLM capabilities into their designated verticals. Task automation, human confirmation.
Multimodality, like Google’s Gemini will take
Enterprise companies will get educated on LLM Prompt Injection. See the full report here:
To Summarize …
The S-Curve tells us about how new markets are established.
The 7 Powers tell us what creates persistent differential margins for businesses.
LLMs have the fastest time to Phase II ever(18 months).
We can expect 2024 to be the year LLMs get to Phase III, a place of monopoly and product ecosystem creation as well as the onboarding of enterprise players, not to mention the buildup of the infrastructure layer.
