The Engine Room
I spent 26 years in channel sales. HP, then Fujitsu, then Cisco. Servers, storage, networking, the unglamorous boxes that make everything else work, sold through thousands of partners in 132 countries.
You develop a feel for these things. Most "next big thing" announcements turn out to be noise. Refresh cycles get rebranded as revolutions, and vendors talk themselves into bubbles every few years. I watched the dot-com build-out from inside HP. I watched the smartphone supply chain rip through the channel a decade later. I have a working bullshit detector for hype cycles.
This one is different. The numbers from the last twelve months make that very hard to argue with.
A note up front. I am not a financial advisor and nothing here is investment advice. This is what humAIne is seeing in the market right now, with the data behind it, written for people who can read a chart without needing their hand held.
The Trade
humAIne's investment thesis keeps returning to the same four layers: chips, data centres, frontier AI labs, and the power feeding all of them. Everything else in the economy is downstream.
That sounds dramatic until you look at what the market has actually done.
Chips, the Brains
From May 2025 to May 2026, the S&P 500 returned roughly 31 percent. A solid year. If you parked money in an index fund you did well.
Now look at what happened inside the chip stack over the same window:
Micron (MU): +770%
Intel (INTC): +483%
AMD: +343%
Taiwan Semiconductor (TSM): +133%
Broadcom (AVGO): +107%
NVIDIA (NVDA): +85%
Six AI chip companies, average return around 320 percent. About ten times the S&P 500 over the same year.
The Intel number is the strangest one on the list. Eighteen months ago Intel was being eulogised, stock below twenty dollars, foundry business written off, activist funds circling. Then the CHIPS Act money landed, leadership got reset, and customer trust started to come back. That turnaround deserves its own piece, and humAIne will write it.
The rest of the chip stack is more straightforward. Compute demand exploded, and the companies on the supply side got paid.
Data Centres and Power, the Body
The chips have to live somewhere. They need cooling, racks, and very large amounts of electricity. That layer of the trade has been even better than the silicon.
Same one-year window:
Vertiv (VRT): +256%. Cooling, power distribution, the racks that hold the whole thing together.
GE Vernova (GEV): +158%. Gas turbines, grid infrastructure, nuclear services.
Bloom Energy (BE): +1,647%. Fuel cells for onsite data-centre power. Yes, that is the right number of zeros.
Oklo (OKLO): +178%. Advanced nuclear designed for data-centre loads.
Fluence Energy (FLNC): +200%. Battery storage and grid-scale energy.
Talen Energy (TLN): +73%. Nuclear power for AWS.
Compare that with the supposedly defensive parts of the market over the same window. Financials returned 5 percent. Healthcare 8 percent. Consumer staples 6 percent. Industrials 29 percent.
Roughly a third of the entire S&P 500 gain came from five chip stocks (NVIDIA, Broadcom, AMD, Micron, Intel). The index is already an AI infrastructure index. Most investors just haven't noticed yet.
The Hyperscalers Are Quietly Eating Everything
The big four hyperscalers are the customers writing the cheques.
Alphabet: roughly $2.1 trillion market cap in May 2024, $4.81 trillion today. Up 129 percent. DeepMind, Gemini, and Google Cloud all rewarded by the market.
Amazon: $1.9 trillion to $2.93 trillion. AWS infrastructure dominance.
Meta: $1.2 trillion to $1.57 trillion. Llama, a massive compute build, an open-source bet that is starting to pay back.
Microsoft: $3.07 trillion, roughly flat from May 2024, but arguably the most AI-leveraged large company on the planet through OpenAI and Copilot. The market is still working out how to price that.
Together those four companies are worth more than $12 trillion, and they plan to spend around $750 billion on AI capex in 2026 alone. That figure is larger than Switzerland's entire GDP. Spent in a single year, flowing straight through to silicon, concrete, copper, and watts.
Frontier Labs, the Brains Behind the Brains
This is where the numbers stop behaving like real numbers.
OpenAI was valued at roughly $80 billion in early 2024. On 31 March 2026 they closed a $122 billion funding round at a post-money valuation of $852 billion, the largest private raise on record. Revenue went from $2 billion in 2023, to $6 billion in 2024, to $20 billion in 2025. They now have 910 million weekly active users. An IPO that could clear $1 trillion is reportedly being prepared.
A ten-times increase in two years. Google took six years to walk that same valuation path from $80 billion to $800 billion. OpenAI did it in twenty-four months.
Anthropic, the lab behind Claude (full disclosure: humAIne runs much of its workflow on Claude): about $18 billion in early 2024. In February 2026 they closed a $30 billion round at $380 billion. Preemptive offers then came in at $850 to $900 billion. By early May 2026, secondary market pricing on Forge Global put them at $1 trillion, ahead of OpenAI. Roughly fifty-five times where they started two years ago.
xAI: a $230 billion valuation in January 2026, total funding raised of $45 billion, revenue growing thirty-eight times year-on-year.
Roll the in-house labs implied inside Alphabet and Microsoft into the picture and the conclusion is the same. Frontier lab valuations have collectively added more than $2 trillion in two years. Funding to foundational AI startups in Q1 2026 alone exceeded the entire 2025 total.
What Actually Drives the Numbers
Three figures worth memorising.
One trillion dollars. Global semiconductor sales projected for 2026, with Q1 alone coming in near $300 billion. The entire year of 2023 was $527 billion. The chip industry is doubling in three years.
$750 billion. Hyperscaler AI capex for 2026, recently revised up from $650 billion by CreditSights. Every dollar lands in chips, cooling, power, or concrete.
Two trillion dollars. The projected size of the AI data centre market by 2032, growing at 27.5 percent compounded from $471 billion in 2026. The smartphone market took fifteen years to reach that scale. AI infrastructure is on pace to do it in six.
Where humAIne Comes In
Two reasons this matters for what we do.
First, humAIne's activist research has been pointed at this stack for two years. Our reports on AI compute infrastructure, semiconductor sovereignty, and European defence all sit on the same thesis: the picks and shovels of machine intelligence are the most durable trade of the decade. The market has now confirmed what was still a thesis when we started writing.
Second, the bottleneck has shifted. Chips are no longer the constraint. Power is. NVIDIA can build more H200s. TSMC can pour more capacity. What you cannot conjure out of thin air is gigawatts of clean baseload electricity, near fibre routes, in politically stable jurisdictions, with cold air and water available for cooling.
That is exactly why humAIne has been spending serious time on Iceland. Idle aluminium smelter capacity sitting on hydro and geothermal power at industrial scale, in a NATO-aligned country positioned on the fibre route between North America and Europe, with a climate that does most of the cooling for free. The board-level strategy work we have been doing on converting smelters into AI compute hubs is a direct play on the bottleneck the rest of the market is only just starting to notice.
What This Looks Like From Where I Sit
Twenty-six years in channel sales teaches you that every infrastructure cycle moves through two phases. In the first, customers ask whether the demand is real, vendors over-promise, and analysts hedge. In the second, customers commit, supply chains break, prices firm up, and the laggards pay premium pricing to catch up.
We are firmly in phase two.
The buyers here are the four largest companies in the world, committing $750 billion of capex in a single year. The supply chain is global and measurable. The bottleneck is copper, concrete, transformers, and nuclear permits, not enthusiasm.
I keep being asked whether the AI infrastructure trade is late. The honest answer, from someone who watched the dot-com build-out from inside HP, is that it depends what you mean by late. Late if you wanted Bloom Energy at last May's price, of course. Late for the entire stack? Not even close. Global chip sales are still growing at 25 percent a year. Hyperscaler capex is accelerating, not slowing. Power supply will be the binding constraint for the rest of the decade.
The market is repricing itself around a new centre of gravity. humAIne is positioned for it. If you read these numbers and feel uncomfortable, that is appropriate. Discomfort is what the start of a real cycle feels like.
Welcome to the engine room.
