The market around AI is hopping.
OpenAI just announced a fundraising at $150 Billion, and new consumer applications are popping up everyday.
Whether it is consumer-facing video apps like Luma, Runway, or eventually Sora; or enterprise-grade LLMs such as Llama (Meta) or Strawberry (OpenAI), the market is ripe with activity.
But is the overall market overhyped, under hyped, or at just the right amount of hype?
Let’s dive in.
Unbundling the AI Business Model Across the Value Chain
The Value Chain of AI is complex and ever-evolving, with multiple major Big Tech players funding the hardware infrastructure (chips) that power their own software infrastructure (cloud), from which an array of apps are beginning to emerge, for both business and consumer.
From a big picture perspective, this flow matters a lot because to understand the business model of each layer requires a careful analysis of the capital flows from end to end. At this point in time:
- the Big Tech companies largely fund the costs of the hardware infrastructure layer
- they use these GPUs (Graphic Processing Units) to power various LLMs (Large Language Models) across different sectors and categories of their software ecosystems
- through a mix of intrapreneurship (companies developing their own apps) and entrepreneurship (free-market development), various consumer and enterprise apps are coming into the ecosystem
The Infrastructure layer is the most synonymous with a lot of the hype we have seen in public markets with the likes of Nvidia rising more than 10X over the last 18 – 24 months. Nvidia produces the GPUs that the major technology players buy, which are made from semiconductors, in what is a very capital and engineering-intensive process to produce from silicon.
Demand for these units has been exponential, thus Big Tech players from Microsoft through to Meta have been investing Billions into GPU capacity to fund their LLM model development.
Of the various “Models” listed above – Open AI, Anthropic, Runway, etc. – many are separate entities that are funded principally by Big Tech companies. Amazon invested $4 Billion in Anthropic for example.
By the time an application reaches the market – whether ChatGPT for consumer search or some kind of Copilot for enterprise functions – the end product is heavily subsidized based on the amount of capital required to power the models.
Thus, questions remain about the sustainability of many of these business models and whether or not we are in the midst of another major ‘Dot Com Bubble’ type of scenario.
The Generative AI Apps Layer
A few months ago Goldman Sachs released a somewhat controversial report titled Gen AI – Too Much Spend, Too Little Benefit?
.
The gist of the report was that despite the promise of the technology, we are too early in the cycle given the expectations of capital return for investors.
More broadly, people generally substantially overestimate what the technology is capable of today. In our experience, even basic summarization tasks often yield illegible and nonsensical results. This is not a matter of just some tweaks being required here and there; despite its expensive price tag, the technology is nowhere near where it needs to be in order to be useful for even such basic tasks
A lot of the tasks that have been improved to date would be considered relatively ‘simple’ in a macro sense, whether it is making coders more efficient using apps like Cursor, or creating better search experiences for consumers via products like Perplexity.
The continued levels of large investment into GPU capacity isn’t any guarantee that we will see quantum leaps in the potential application layer either.
What does it mean to double AI’s capabilities? For open-ended tasks like customer service or understanding and summarizing text, no clear metric exists to demonstrate that the output is twice as good. Similarly, what does a doubling of data really mean, and what can it achieve? Including twice as much data from Reddit into the next version of GPT may improve its ability to predict the next word when engaging in an informal conversation, but it won’t necessarily improve a customer service representative’s ability to help a customer troubleshoot problems with their video service.
A more granular, case-by-case approach is necessary, and we will likely see certain industries lead the charge in terms of adoption. Right now it makes sense that tech-forward industries and financial services are leading the adoption cycle.
.
The overall question is not so much about the potential or upside in Generative AI – because it is enormous – but the expectations and timeline to reach that target.
Too much optimism and hype may lead to the premature use of technologies that are not yet ready for prime time. This risk seems particularly high today for using AI to advance automation. Too much automation too soon could create bottlenecks and other problems for firms that no longer have the flexibility and trouble-shooting capabilities that human capital provides.
Yet since the release of that report, and especially in the last couple of weeks, big moves have happened in the market, and new business models are already emerging for the top players in Generative AI. Are we on the verge of a major inflection point?
The $600 Billion AI Business Model
In that same skeptical light, Sequoia VC published a similarly-themed piece recently in an effort to quantify the level of return necessary to justify the hype.
The way that this number was calculated was based on a top-down analysis of GPU capital investment relative to expected margins on software.
These calculations, while crude, give us some insights into the AI Business Model and how much scale it needs to achieve at the consumer and enterprise application end.
The conclusion the author reaches is similar to Goldman’s, which is that there will be big winners and big losers, and it will be a long journey down the path to determining who those winners are. This is very similar to many major investment cycles in new technologies throughout history.
The overarching point is that by any objective standard, we are in a speculative frenzy.
Speculative frenzies are part of technology, and so they are not something to be afraid of. Those who remain level-headed through this moment have the chance to build extremely important companies.
Now that we have established that and know what to look for (short-term application, revenue, etc.), let’s cut into some of the emerging applications in Generative AI.
Top Generative AI Applications
OpenAI & The ‘Strawberry Revenue Model’
In September 2024 it was announced that OpenAI had raised $6.5 Billion in venture capital along with $5 Billion in debt at a valuation of $150 Billion through participation by Apple and others.
They released their o1 LLMs to “push AI to PhD-level intelligence.” Also referred to as Strawberry, these models will be pushed towards Enterprise clients with a slated price tag of $2,000 USD per month. Currently, those with GPT Plus or Teams can access a preview of the models.
Currently, Chat GPT dominates the consumer market with their first product, priced at $20 USD per month.
There are various leaps in the models relative to competitors, which is why the technology has generated so much hype since its release, following the historic fundraising announcement.
The slated $2,000 price tag is 100X the currently base price of Chat GPT Plus, so it certainly will be interesting to see how much they sell.
Nevertheless, the hype may be warranted if it can generate enough productivity gains for those who subscribe.
Cursor (Anysphere)
Cursor is a product targeted specifically at Developers. It has a similar business model to Chat GPT with Cursor Pro at a $20 per month price point.
The value proposition of Cursor is that Developers can effectively scale-up their productivity levels and avoid having to hire more Developers in some situations.
It is typically paired with other AI applications like Claude (Anthropic) and Replit to create new apps or improve existing capabilities.
Within Cursor, you can switch between models (Claude Sonnet, OpenAI o1, etc), giving it functionality that could potentially scale on the back of a beefed-up version of OpenAI’s Strawberry.
The company – Anysphere – is a young team of software engineers who recently raised $60M USD from high-level venture investors at a $400M valuation.
Anysphere raised $60 million in an investment round led by Andreessen Horowitz, and included funding from OpenAI’s Jeff Dean, John Schulman, Nat Friedman, and Noam Brown, with the overall valuation at $400 million.
The overall effect on writing code for Developers has been shown to markedly improve with Generative AI, and Cursor appears to be the product leading the pack among Developers.
It showed that efficiency among developers grew by 26% while the number of code completions increased by 38%.
While not expected to be selling for $2,000 per month nor reaching outside of the Developer niche, a $400 Million valuation is a lot less than a $1.5 Billion valuation.
The Anthropic Revenue Model
Recently it was revealed that Anthropic was tracking towards 1 $Billion in revenue in 2024 and was planning to raise money at a $30 – $40 Billion valuation.
As mentioned above, Anthropic had previously received investment from Amazon (and Google).
Similar to OpenAI, they are expected to release new Claude models in the very short-term, and will then be competing head-to-head with OpenAI for market share in consumer and enterprise, not to mention whatever integrations Amazon and/or Google are working on.