Currently, the AI market is hugely concentrated around a few winners, which move the whole market, and these mostly sit on the foundational and hardware layers.
You’ll find this data clearly shown on my dashboard.
However, as I’ll show, another layer of the market is finally opening up substantially: that of vertical AI players.
Those specialized generalist companies are trying to build a valuable AI to tackle a specific vertical (accounting, finance, healthcare, and so forth).
In December 2022, I explained exactly that; when I referred to the middle layer of the AI stack, I was referring to vertical AI companies.
I also raised a warning on Salesforce.
Fast forward over two years, it seems SF has gotten the message and re-organized the whole company to the AI, betting it all on AI Agents.
It is meant as co-pilots or autonomous orchestration workflow to automate tasks like customer support, which is the first use case SF is aggressively tackling in a bloody fight with Microsoft.
That’s unsurprising, as the market is still under development, and the capital dynamics are not that different from how they played out in the previous technological cycle.
Yet, there is a more skewed effect in the current AI paradigm.
The “Wow-Effect” – The Hedonistic Treadmill And The Hedonic Decline
The hedonistic treadmill or hedonic adaptation is a mental phenomenon where humans relatively quickly tend to tend toward the average level of happiness after they have experienced an outsized positive outcome.
It seems we have already taken for granted something that only three years back seemed impossible.
Starting in the 2010s, up to the “ChatGPT Moment” (November 30, 2022), neural nets were promising for a few researchers who still believed in the field.
On the other hand, it’s worth reminding you that neural nets could not achieve such a level of “cognitive abilities” (in terms of output, as we have little clue of what goes in the black box).
That caught everyone off guard.
Also, people like me have been seeing the field’s evolution for the last decade.
That change of pace, that surprising feat, and the still-evolving AI landscape, which went through three phases, that we’re still seeing playing out.
Since the current AI paradigm, which I’m simplifying as “deep neural nets built on a transformer architecture, pre-trained on top of thousands of GPUs, by feeding the whole web,” we went through three main phases:
- Pre-training phase.
- Post-training phase.
- Reasoning phase.
To be clear, while all these three phases will continue to develop in parallel, each might become the key driver for the next adoption phase.
Each of them comes with its own technical paradigms and business models.
The Web²: The AI Supercycle
·
Jan 4
It’s just part of human nature to take something as impressive as these AI models for granted.
It’s our nature to experience a “wow effect” and integrate it into our daily lives, taking it for granted.
However, the actual psychological test of whether these things are so valuable is when you lose the chance to use them (hedonic decline).
I’d assume that if OpenAI’s ChatGPT disappeared, it would madden many people, and many professionals would not be able to operate at the same level.
In short, while the “wow effect” easily passes, turning into a hedonic treadmill, you go into hedonic decline as soon as you lose that thing you took for granted.
You can do it with any extremely valuable technology you take for granted.
And that’s not the end of the story; as new reasoning models come to the market, making these capabilities cheaper and more scalable, we’ll see a growing set of commercial use cases emerge in many vertical niches.
For the same token, we’ll also see the same pattern of the “wow effect” followed by the hedonic treadmill.
That’s also why, non-trivially, releasing these AI tools to the public is critical to make sure we take them all for granted rather than treating them as a terminator.
When things appear in our imagination, they always seem scary, vs. when we experience them, in a sort of never-ending collective psychological cycle, which looks like that.
I’ll leave the psychological effect of the AI cycle to another issue, for now let’s get back to business…
This is part of an Enterprise AI series of (possibly) daily short pieces to tackle many of the day-to-day challenges you might face as a professional, executive, founder, or investor in the current AI landscape.
All these pieces are freely available to you. If you find the piece isn’t enough to help, you can contact me once you join in as a Founding Member.
A Consumer-First Paradigm, But What Now?
In addition, contrary to many other paradigms, which started from enterprise and then slowly cascaded to consumers, these large enterprise businesses financed the balance sheets of startups, who slowly then quickly moved downstream.
This current paradigm is moved all the way around.
ChatGPT has been the fastest-adopting consumer app, faster than TikTok. Indeed, for how bizarre a name might be (a nerdy one), you can do “the mother test” right now by asking her whether she knows what that is.
Thus, we have had an AI paradigm, catching most of the professionals in the field off-guard while giving these AI tools to hundreds of millions of people across the world almost overnight.
That has created a FOMO effect for enterprise businesses that are still playing catch-up!
The Enterprise Catch-up Game
Connected to the above, enterprise businesses are now trying to figure out how to make sense of it by implementing these technologies as soon as possible while reconciling their entire organizations.
That’s not an easy one to pull off.
I’ve explained how the incumbent paradox plays out in the buy vs. build dilemma.
And how some AI projects within these enterprise businesses will require simple technical implementations while others will turn in entire re-org!
Full piece here:
The Build vs. Buy Dilemma in Enterprise AI
·
Jan 20
Not Just a Tech But An Economic Paradigm
The thing with this current AI paradigm is that it might be way more similar to the microchip, as a computational layer, than the Internet.
![Navigating The AI Supercycle [Video Version]](https://i0.wp.com/substackcdn.com/image/fetch/w_140%2Ch_140%2Cc_fill%2Cf_auto%2Cq_auto%3Agood%2Cfl_progressive%3Asteep%2Cg_auto/https%3A%2F%2Fsubstack-video.s3.amazonaws.com%2Fvideo_upload%2Fpost%2F154910360%2F53e6608e-b2f4-4f0d-8007-33172839dbb9%2Ftranscoded-360419.png?w=1920&ssl=1)
Navigating The AI Supercycle [Video Version]
·
Jan 18
This difference might seem trivial, but it has many practical consequences.
Let me explain why.
Capital Concentration, And Winner-Take-Everything
What’s different today, compared to the Internet era, in terms of capital concentration, development, and deployment?
The web acted primarily as an informational layer, starting with outer-layer transformations like distribution channels and gradually moving inward to transform core business models.
Amazon exemplifies the web integration journey: beginning as an online bookstore (distribution layer), adding digital features (value enhancement), and eventually becoming a comprehensive platform (core transformation).
This outside-in pattern defined the internet era.
In contrast, the AI paradigm works from the inside out.
Rather than starting with distribution advantages, AI begins by transforming the core value proposition of products and services.
This fundamental difference means that while distribution was the key battleground of the internet era, in the AI era, superior core capabilities drive success, with distribution advantages emerging as an organic consequence of a 100x value prop.
This means that the capital is running right after them!
While this dynamic has also been confirmed during the web era, it’ll get even more skewed in the AI era.
This is how it might look visually regarding winner-take-all effects and skewness of the private capital markets.
Remember that the numbers shown are only a reference for understanding this market’s mental model dynamic.
This is part of an Enterprise AI series of (possibly) daily short pieces to tackle many of the day-to-day challenges you might face as a professional, executive, founder, or investor in the current AI landscape.
All these pieces are freely available to you. If you find the piece isn’t enough to help, you can contact me once you join in as a Founding Member.
Not All Capital Is Winner-Take-All
Of course, capital concentration will be different based on the different parts of the AI tech stack.
The Capital Deployment framework shows how the “Winner Take All” dynamic varies across the AI stack:
- The Foundation Layer shows the highest concentration (2-3 dominant players) due to massive scale advantages and capital requirements.
- Hardware Layer and Application Layer also show high concentration but for different reasons – hardware due to infrastructure barriers, applications due to strong network effects, and category dominance.
- The Vertical Layer uniquely shows a more moderate distribution, allowing multiple winners per vertical due to industry-specific needs and specialization opportunities.
This understanding is crucial for PE/Family Offices in capital allocation, suggesting different investment strategies based on layer dynamics – concentrated bets in foundation/hardware/applications versus more distributed investments in verticals.
Capital Deployment Across The AI Layered Stack
Key implication: While AI drives concentration in most layers, verticals offer more balanced opportunities.
As a mental model, think of the foundational, hardware, and application layer as quite skewed (I’m representing only the shape of the foundational layer:
Of course, the same sort of winner-take-all effects are in the application layer, where you can write much smaller cheques for a much larger outcome driven by the category leaders.
While the vertical layer might look way less skewed:
Before we get to the vertical layer, there is a premise: investing in it will give more chances of success and less outsized opportunities.
That’s precisely the point of venture capital. You place many bets, some less skewed, some more.
The more skewed ones will also potentially be the more outsized ones.
Risk vs. Outsized Opportunity
Of course, the premise when it comes to capital deployment is way more than just opportunity, as you also want to look at the potential size of it!
As we move across the AI layers, vertical players with a less outsized opportunity in terms of return might also be where most of a VC portfolio will be invested.
That calls up for another key concept.
Asymmetric Betting
In capital deployment, the inherent risk on one end is balanced by the potential for outsized, asymmetric opportunities on the other.
Even in domains with massive winner-take-all dynamics, the possibility of extraordinary outcomes makes strategic small bets worthwhile.
These environments, while high-risk, offer the potential for exponential returns, where even a single success can outweigh multiple losses.
Thus, in that respect, we might see many VC funds opting for a barbeled approach.
Barbeling Capital Out
Barbelling out is a strategy that balances high-risk, high-reward bets with a stable core portfolio of less outsized but more reliable opportunities.
The approach allows investors to pursue asymmetric opportunities on one end, where even a small success can yield exponential returns while maintaining stability with the core, ensuring steady, predictable outcomes.
By combining these extremes, investors can capture the upside of risky ventures without jeopardizing overall portfolio resilience.
This dual approach mitigates risk while optimizing potential gains, making it an effective strategy for navigating uncertain or volatile markets.
Recap: In This Issue!
- AI Market Concentration: The AI market is dominated by a few key players, particularly in the foundational and hardware layers, shaping the industry dynamics.
- Emergence of Vertical AI: Vertical AI players are gaining traction, targeting specific industries like healthcare, accounting, and finance, offering more specialized and distributed opportunities.
- Salesforce’s AI Transformation: Salesforce reorganized its strategy around AI agents, focusing on automation (e.g., customer support) to compete with Microsoft.
- The Psychological Cycle: The “Wow Effect” of AI advancements (e.g., ChatGPT) quickly gives way to the “Hedonic Treadmill,” where these tools are taken for granted, but their loss (Hedonic Decline) reveals their critical value.
- Consumer-First Paradigm: AI adoption began with consumers before transitioning to enterprises, reversing traditional cycles and creating urgency for businesses to catch up.
- Capital Concentration in AI Layers: The foundational and hardware layers are highly skewed with concentrated capital, while the vertical layer offers more balanced investment opportunities.
- AI’s Inside-Out Paradigm: Unlike the internet’s outside-in model, AI starts by transforming the core value proposition of businesses, with distribution advantages emerging as a byproduct.
- Risk vs. Opportunity: While foundational and application layers carry high risks, they also promise outsized returns. Vertical AI layers provide steadier, more predictable opportunities.
- Asymmetric Betting: Strategic small bets in high-risk, high-reward domains balance potential exponential returns with manageable risks.
- Barbelling Capital Strategy: A portfolio approach combining risky, outsized bets with stable, reliable investments ensures resilience while capturing high growth potential.
With massive ♥️ Gennaro Cuofano, The Business Engineer
This is part of an Enterprise AI series of (possibly) daily short pieces to tackle many of the day-to-day challenges you might face as a professional, executive, founder, or investor in the current AI landscape.
All these pieces are freely available to you. If you find the piece isn’t enough to help, you can contact me once you join in as a Founding Member.
