I think about AI value in three layers.
The first layer is picks and shovels.
During the original gold rush, the miners took on all the risk. The people who sold them equipment got paid regardless of who struck gold.
Nvidia is the obvious example. They didn't win because ChatGPT became popular. They won because every AI company on earth needed their chips to compete. Nvidia's revenue doesn't depend on which AI company wins. It depends on the race continuing. That's a fundamentally different risk profile.
But chips are just the start. The pick-and-shovel layer extends into power infrastructure, cooling systems, fiber optics, data center construction, raw materials like copper and aluminum. Every model that gets trained or served needs physical infrastructure that someone has to build and maintain.
These companies win during the buildout. They get paid whether the end products work or not. That's exactly why Nvidia is projecting $1 trillion in chip sales through 2027 and nobody blinked.
The second layer is what I'd call second-order beneficiaries.
Companies that don't sell AI directly, but whose existing businesses get meaningfully amplified by widespread adoption. When electricity became standard, the obvious picks and shovels were power companies and wire manufacturers. The real long-term winners were the factories, retailers, and logistics companies that could suddenly operate at entirely new scales.
Same pattern now. Cybersecurity, healthcare diagnostics, industrial automation, financial services. They're not selling AI. They're using it to compound advantages they already had.
Key timing difference: picks and shovels win as AI is built. Second-order beneficiaries win as AI gets used. We're still firmly in the building phase, which means this second layer of returns hasn't fully materialized. That's either a risk or an opportunity depending on how patient you are.
The third layer is the one most people skip, and it's the one I'm most confident about.
It's not a stock or an ETF. It's a skill set.
I've watched people in my own network go from standard agency roles to six-figure independent practices by getting genuinely good at building with AI tools. Not prompting. Building. Architecting systems, automating workflows, creating things that would have required a team of five two years ago.
$100K in an AI ETF might return 15-20% annually if the thesis plays out. That's $10,500 to $20,000 a year. The same $100K invested in actually learning to build with these tools can return multiples of that in year one. And unlike every other position on this list, you can't lose the skills you acquire. There's no drawdown on competence.
This is where my bias is strongest, and I'll own it. I run an agency that builds with AI tools daily. Not as a marketing angle. As core infrastructure. The leverage is real and it compounds.
In my last piece I said I'd rather be holding skills and cash when the accounting corrects. This is the skills part. The people who capture the most value from this shift won't necessarily be the ones who picked the right ticker. They'll be the ones who understood the technology well enough to build real things on top of it.
Every gold rush ends the same way. The gold gets harder to find. The equipment stays essential. And the people who learned the terrain end up owning the next thing.
Further reading
- The Citation Economy — the operating manual for one of the second-order layers: being the source AI engines build answers from.
- The Wrong People Are Worried About AI — which parts of the agency and operator model survive automation.
