Investing in AI: 3 Trends to Watch

There’s no doubt that AI is eating the world. But with so much hype surrounding this technology, it's easy to get lost in the noise.

From chatbots that (sort of) understand me to self-driving cars that (mostly) don't freak me out, AI is increasingly woven into the fabric of our daily lives - but as an investor and entrepreneur, the key question is where do I place bets on the future of AI?

Here are my thoughts on where we are headed for AI, investible categories in AI, and 3 trends to watch (at the bottom):

Where are we in the AI maturity curve?

AI is evolving at a pace where some companies are getting disrupted within months of funds. There’s no doubt that the pace of innovation is accelerating given the combination of: (1) Improvements in GPU performance (2) Exponential reduction in compute costs (3) Growth in data volumes (4) Growth in innovative AI research (5) Growth in AI capital

If we listen to AI model builders (e.g., Altman, Zuck, etc.) - they’re taking these resources to improve on:

• Model speed

• Model accuracy

• Depth of knowledge base

• Reasoning capabilities & “EQ” capabilities

From a practical perspective, I predict improvements in these areas will unlock abilities in:

• AI-to-AI communication, where many tasks can now be done on behalf of humans

• Scaled AI with deeper integration into vertical apps. This means integrated solutions for AI into every app. Not just adding a “GPT interface” into products (e.g., Samsung’s AI refrigerator lol).

Because model builders sit towards the bottom of the AI stack, these capabilities will impact every company upstream.

What are the investible categories in AI?

I would broadly categorize the investible landscape for AI in 3 buckets:

1. Hardware Layer:

   • Chips: Specialized AI chips include GPUs (AI training and inference), TPUs (machine learning tasks), FPGA (custom AI hardware use cases), and ASICs (specific AI use cases)

   • Other Hardware: Other hardware requirements include data centers, networking, and power management

2. Model Builders / Dev Layer:

   • LLM Providers: LLM builders compete to provide base models for developers building domain specific AI use cases

   • Developer Tools: Developer apps include data bases, data pipelines, orchestration, security, observability, and more

3. App Layer:

   • Consumer Apps: Consumer apps include front ends to consumer use cases (e.g., video and image generation, etc.)

   • B2B Apps: AI tools for businesses for increased efficiency & new use cases

Examples of companies in each layer as follows:

3 Trends to Watch:

(1) Commoditization of LLMs & Hardware:

We are seeing billions of dollars of capital being poured into competing on both the LLM side and AI hardware side. OpenAI and NVIDIA’s early dominance / monopoly as a result will likely wane as the capability gap tightens.

LLM solutions so far have low switching costs on the back end for app developers - and the clear business strategy would be to build a moat here that fences in apps to LLM ecosystems.

This is the strategy that Nvidia is going after and I wouldn’t be surprised if the LLM builders do it too. One of the primary ways NVIDIA creates stickiness is through its CUDA (Compute Unified Device Architecture) ecosystem. CUDA is a parallel computing platform and API model that allows developers to use NVIDIA GPUs for general-purpose processing. By integrating NVIDIA code into training processes, the switching costs become significantly elevated.

In other words, I expect the developer layer on the AI stack to become hotter as demand for AI devs increase & the big players look to leverage dev tooling to lock in apps to their LLM workflows.

OpenAI is showing their hand in their recent M&A strategy:

• M&A of remote collab platform: Link

• M&A of data retrieval platform: Link

(2) Penetration of Vertical AI Apps:

The value proposition of vertical AI apps is incredible for businesses - improve productivity per employee, reduce headcount, and automate manual processes.

The economics of leveraging these solutions are a no brainer, and in my personal opinion a matter of when not if that the B2B app layer gets integrated in all workflows.

We know from traditional SaaS that education cycles and sales cycles take forever for these client profiles.

But as this market matures, there are distinct advantages that vertical AI apps will have: (1) Ability to charge higher ticket given the amount of cost savings they provide (2) Stickiness from integration of workflows and “career risk” for insiders to switch solutions

The key question here is to ask if company XYZ will get disrupted by the LLM builders. In other words, assuming LLM capabilities 100x in quality from here, will my startup or investment get disrupted?

Qualities I would look for are differentiated user experience vs generalized LLM interface and deep integration into corporate workflows. See: Bloomberg Terminal / Salesforce strategies.

(3) Crypto x AI

Some last words on AI x Crypto. I think the AI x Crypto ecosystem is developing in parallel to web2 AI - i.e., not direct competitors but fundamentally solving a different set of problems.

There are a few tailwinds I see happening for crypto x AI and ways these decentralized solutions can differentiate:

• AI x Crypto as a hedge against centralized mechanisms. AI is increasingly becoming a great power competition which may cause more users to flock to open sourcing and decentralized ecosystems. JD Vance in fact has spoken about the importance of this.

• AI x Crypto as a mechanism for scalable AI. As AI-to-AI autonomy becomes possible, there needs a way for payments to scale alongside this for cost and incentives solutions. Can you name a better programmable money mechanism than crypto?

• AI x Crypto as a way for liquidity. We are definitively in a market where it is difficult to exit in traditional markets. If these windows drag out, I think many devs will consider using AI x Crypto as an alternative venue for liquidity on their companies. Especially if the programmable money thesis plays out and integrates with these products. Higher relative valuations, faster liquidity cycles, less OpEx for compliance, etc make crypto venues attractive.

PS: If you enjoyed this post, I would appreciate if you could forward this email to 3 of your friends!

Subscribe to our newsletter here: Link