Three Hypotheses About Economic Development with AI
From Merriam-Webster - Develop: to cause to evolve or unfold gradually, to lead or conduct (something) through a succession of states or changes each of which is preparatory for the next
The following was not written or embellished with any form of AI - straight from my head to your eyes.
For the past couple months, a few hypotheses about the development of AI in our economy have been floating around in my head. While I can’t prove or disprove them just yet, I can make them public and open to scrutiny.
Beyond just putting them out there, a mentor of mine recently told me that it would be foolish not to test these hypotheses, whether it be with my money or more importantly, my time. If you’re in my life and find that I’m doing neither, please call me out! And as always, these are hypotheses. I invite any and all dissent.
1. The gap between Middle Market Firms and Large Enterprises will narrow.
AI adoption happens quickest in smallest teams. The smaller the team, the greater the likelihood of a quicker ability to adopt new tools, practices, and ways of working. Assuming smaller, leaner teams are able to move more quickly, its likely that Middle Market Companies (PE-owned, family-owned, <$1B revenue) can move quicker than Large Enterprises (>$1B revenue) within the same industries by adopting AI and actually seeing realized value from AI investment.
As a result of this AI adoption, these Middle Market Companies will be able to price their goods and services more competitively to Large Enterprises, thus narrowing their revenue gap and market share. The firms that can adopt AI the most effectively and efficiently will more greatly narrow the gap between themselves and their competitors.
Lastly, due to the current AI boom, Middle Market companies now also have access to more tools, services, and products to improve their businesses, which were previously reserved for Large Enterprises with deep pockets. Think consulting services, marketing agencies, access to expertise, access to analysis, and more.
The sole caveat that comes to mind for this (thanks Pranav), is that it’s unlikely that the gap will narrow much for titans of industry like Google, Nvidia, JP Morgan, Exxon, BP, Eli Lilly, and many others. If I had to pick a bucket of which Large Enterprises would be threatened most by Middle Market incumbents, it would (arbitrarily) be companies that have <0.50% weight in the S&P 500 index.
2. The future of Large Enterprises in the age of AI is on a single, unified data platform.
Having used unified data platforms, such as Palantir’s Foundry, I can truly say that I see the AI Adoption light. For large companies that want to “use AI” and become more efficient from AI investment, the only right answer is a single platform on which data can be loaded, transformed, and built on top of. Entire businesses can and will run on one platform where both AI Agents and Employees work together to run the business. Even today, competing platforms exist from companies such as C3.ai, Palantir, Databricks, Snowflake, and many more.
Currently, people are deploying AI in companies in one of three ways: on unified data platforms, rolling out ChatGPT/Claude, or shipping specific “point” solutions which serve one purpose (think a website, an automation, or a workflow). I’m due to make another post detailing ‘the right way to deploy AI’ for companies, but the teaser for that is as follows; building and deploying point solutions is not scalable, and deploying LLMs without user adoption is a waste of money, but a unified data platform allows for future scalability of impact while allowing for scalability of user adoption too.
On a single platform, users and agents can go to one place to gather context and execute their work, instead of solving problems by gathering data from disparate sources again and again. The complication that a lot of smart companies have diagnosed when solving problems and building solutions with AI, is that it’s the back-end data infrastructure that is troublesome, not the actual act of building a solution. A unified data platform, solves that complication.
3. Lightweight, bespoke, and smaller AI models will triumph in the next decade.
Today, we are pampered with Claude, ChatGPT, Gemini and many more models. The money you put up to use AI is far less than the actual objective usage value you get out of it. You pay $20 for a subscription to get way more than $20 of computational power and infrastructure expenses like electricity. This is public knowledge but feel free to fact check me (05/2026).
We can use hundreds or even thousands of dollars worth of tangible tokens for less than a fraction of the cost. The reason for this is because model development and the race to AGI is insanely subsidized, both by private capital / VC as well as the public markets (Google’s insane CapEx of ~$180B on ~$480B revenue, for example).
Instead of agents using models with hundreds of billions of parameters, it is far more practical and scalable to use smaller, fine-tuned models that are use-case, industry, task, or even role specific. A day will come where finance analyst models, chemical engineer models, and supply chain models are used in companies instead of Opus or GPT models.
In the short term, though, it doesn’t make sense for companies to spend more on fine tuned models instead of using the cheaper (but heavily subsidized) ChatGPT or Claude today. But in the long term, you don’t need a 400 Billion+ parameter model to give you a summary of a contract. You need bespoke models that are perfect for specific tasks. That said, how these models are made and who will make them, is an entirely different story.
The rampant, and often ineffective, use of AI will continue to grow in the next few years. The previous hypotheses are practical perspectives on how the dominoes will fall from someone who has taken the lens of AI adoption within mature and developed companies.
While my personal belief is that meaner, leaner teams (…aka startups…) will be best positioned to take market share in any industry, assuming AI-first business processes, it will take time for smaller companies to build the trust of customers at scale. The long-term business relationships of Middle Market Firms and Large Enterprises will delay their disruption from incumbents, but leaner AI-first competitors will triumph.
While these mean, lean companies scale, AI usage and impact for established companies will continue to develop via smaller, more-bespoke models running on unified data platforms. This mode of deployment will ultimately cause revenue gaps to narrow between Medium and Large Companies in the same industries.
I do think all 3 of your hypotheses will come true. I can definitely see specialized models will be the way forward - actuarial models, stock fundamental analysis models, engineering models... it can spawn to a lot of models. Really thought provoking article!!!
another thing is that technical moats will be a lot harder to build, and companies that invest in marketing, partnerships and distribution will be set up to stay relevant!