The AI code smoky

You’ve seen what’s happening…everywhere.

But, how the heck do AI companies make money?

They’re investing hundreds of billions into data centres full of cutting edge chips.

Are they making returns on this investment?

If not now, when?

You might remember from my previous articles that web technologies solved the problem of data exchange.

Cloud computing solved the data-streaming problem.

And for technology’s next trick…

AI does the same thing for data manipulation, turning unstructured information into decisions, predictions and increasingly, actions.

Where’s the money?

AI has a revenue problem.

Let’s look at recent history.

In the 2000s, US retail giant Target hired a statistician named Andrew Pole. He built a machine learning model that assigned female customers a pregnancy prediction score based on 25 products they bought — unscented lotion, large bags of cotton balls, and supplements.

Target could time the coupon mail-out to the trimester.

Famously, a Minneapolis father complained that his teenage daughter was receiving maternity coupons.

A week later, he rang back to apologise. The model had known before the family did.

Creepy, and just as true as the fact Target’s revenue grew from US$44 billion in 2002 to US$67 billion by 2010.

The program became a foundational case study for predictive marketing in retail.

People forget AI has been around for ages!

The use cases were boring and quiet — logistics, data and process work.

In part 1 of this series, I covered DART, the Dynamic Analysis and Replanning Tool, which was deployed all the way back at the start of Operation Desert Storm in ‘91.

Its logistical efficiencies during the conflict generated close to US$1 billion in operational savings.

It wasn’t a one-off.

Fast forward to the 2000s and the familiar names that built off the dotcom bubble turned into trillion dollar companies.

Fast forward further to today, and capital expenditure by the hyperscalers Amazon, Google, Microsoft and Meta) doubled between 2025 and 2026.

That’s just two years.

They’ve committed to spend a combined US$630 billion for 2026. 2027 capex is forecast to exceed US$1 trillion.

It sounds absurd until you see what it’s paying for.

The AI build-out demands power — nuclear, solar, gas, battery storage. It demands water for cooling. It demands raw materials in volumes that pull mining, processing and manufacturing industries along with it.

And the numbers I’m giving you could be outdated by the end of the week as investment grows.

$1 trillion on black

No one’s throwing US$1 trillion at a side project; this is about unlocking a potentially huge market.

This is where the dotcom ghosts appear.

In every example so far, AI lived in purpose-built tools; it couldn’t scale because it just wasn’t accessible.

You couldn’t talk to it. You couldn’t scale it.

LLMs (large language models) created an accessible natural language interface that everyday people could use as a general intelligence assistant.

What changed wasn’t the AI. What changed is who could use it.

And it became wildly popular.

Commonly, the AI services most people pay for are ChatGPT, Claude, Gemini, and Grok.

They started as chatbots. They’re now drafting legal briefs, generating ad campaigns and diagnosing medical scans. This is what the techies call “the product surface” expanding.

The appetite for development outsizes the current revenue because the AI brand is being built alongside the infrastructure.

No one’s born wanting chocolate.

You try it. You love it. Some of us can’t live without it.

Open AI was the first sweet treat.

OpenAI is projected to spend US$125 billion per year on training it’s AI models by 2030.

Then there’s Anthropic, who is projected to spend US$30 billion per year on training over the same period.

Right now, OpenAI is spending at a roughly US$25 billion annualised run-rate. Anthropic is at US$30 billion. Growing fast… but spending faster.

These companies are cooking AI-powered ovens fueled by cash; OpenAI projects a US$14 billion bottom-line loss for 2026 alone. Both companies are trying to maximise market share, believing the profits will come. But philosophically, they are very different.

OpenAI was founded in 2015 as a non-profit. The company’s mission was AI for the benefit of humanity.

Sound familiar?

The same idealism that gave the world TCP/IP — common web protocols — and the web.

OpenAI is going public at a valuation of near US$852 billion.

Meanwhile, Anthropic signed government contracts in July 2025 for US$200 million, with restrictions baked in — no mass surveillance, no fully autonomous weapons.

When the Pentagon demanded the restrictions be removed, Anthropic said no. They lost the contracts.

Federal AI procurement is roughly US$52 billion a year. OpenAI took those contracts without the restraints. Anthropic walked.

Anthropic’s move resonates with the philosophy of the early internet, that openness and trust pay off long after the build is finished.

Berners-Lee had the same crossroad in 1993. He chose to keep the Web in the public domain. OpenAI went the other way.

Amazon’s Web Service made its real money outside Amazon’s retail business. It enabled iconic builds, such as Netflix, Airbnb and Stripe. Those enterprises brought the early income to Amazon, which allowed it to scale is web services business safely.

Death of a Software Engineer

This isn’t an Arthur Miller tragedy, it’s today’s labour market.

Anyone can code; is that going to put software engineers out of work?

Anyone can garden; that fact also didn’t put landscapers out of business.

Anyone can paint a wall: a painter still gets paid to do it properly.

My experience in this field tells me that AI’s coding-as-a-service (like Claude code) is an impressive proof of concept. But it’s not where AI’s longevity stands.

History tells us that the biggest returns go to second-wave companies.

Consider improved logistics, direct marketing, and efficient drug discovery. AI, combined with human creativity, creates opportunities through data manipulation.

Efficiencies are where AI’s payoff lies. Some companies have cracked that code.

Take Rebar — a US company with an AI operating system for the construction industry. Computer vision analyses construction blueprints to reduce quote generation time by 60–70%, boosting customer win rates by 2–3 times. This is a current real-world example.

It’s where we will see the next Netflix, Airbnb and Uber built upon. AI Businesses that will embed themselves in very boring industries and create outsized opportunities.

The internet, the cloud, AI. Three booms fit into the same pattern. Government plants the seeds. Commercial capital builds the infrastructure and monetisation follows…eventually.

I’ll be watching three things over the next 24 months:

• Who is generating the revenue, not just spending on compute?
• Who is building what nobody can rebuild in six weeks?
• Who’s still in business when the cash burn catches up?

The fortunes from the internet boom didn’t go to the builders. They went to what was built on top. Same pattern for the cloud. The same pattern is coming for AI.

Watch boring industries. That’s where the giants are setting up shop.

Next week, I’ll have more on what specific industries are about to get a major reboot, and how I think investors can benefit from spotting the change quickly.