AI Funders vs AI Suppliers: Who Is Really Winning the AI Boom?

Something unusual happened in June 2026. The Magnificent Seven, the group of mega-cap stocks that most people associate with the AI boom, lost roughly $2.3 trillion in combined market value over a single month. Microsoft had its worst month since December 2000, with the stock falling around 17-18% in June 2026. Meta fell after raising its 2026 AI spending plans to as much as $145 billion. Amazon dropped. Tesla dropped. Even NVIDIA, which had just reported $81.6 billion in quarterly revenue, slid about 13%. 

At the same time, the VanEck Semiconductor ETF was up roughly 64% for the year. The Roundhill Memory ETF, which holds SK Hynix, Micron, and Samsung, had more than doubled since its April launch. Same AI story, same macro environment, completely opposite market outcomes. The reason is that the market has started separating the companies spending money on AI from the companies collecting it.

Let's break down who sits in each camp, what the numbers actually say, and whether the side the market is currently rewarding is the right one to be on.

AI Funders vs AI Suppliers: The Two Sides of the AI Boom

Think of the global AI buildout as one enormous construction project. On one side, you have the financiers: Amazon, Microsoft, Alphabet, and Meta. These four hyperscalers plan to spend a combined $725 billion on capital expenditure in 2026, up 77% from roughly $410 billion in 2025, according to Financial Times analysis. They are building data centers, buying land, securing power infrastructure, and acquiring the hardware that AI needs to run.

On the other side are the suppliers: NVIDIA for GPUs, TSMC for chip manufacturing, SK Hynix and Micron for memory, Broadcom for custom AI chips, and ASML for the lithography machines that make advanced chip production possible. Their revenue rises in direct proportion to what the financiers spend.

The market has started calling one group Funders and rewarding the other, the Collectors. The reason is a question it kept ignoring in 2024 and 2025, and is now refusing to put aside: when does all this spending show up as profits?

Why Big Tech Stocks Fell While AI Supplier Stocks Rose

June 2026 was not a broad tech selloff. It was a specific reassessment of what hyperscaler spending actually means for shareholders when the returns are still theoretical.

Source: CNBC (July 1, 2026), TradingView

The entire Magnificent Seven shed roughly $2.3 trillion in market value across the month, as per CNBC.

What actually triggered this? Meta raised its 2026 capex guidance to as much as $145 billion. The stock dropped roughly 6% after hours. Not because Meta's advertising business was struggling. It was not. The market flinched at the size of the spending and the uncertainty around its payoff.

Alphabet was the only Funder that escaped with a positive reaction at April earnings, gaining roughly 7-10% after disclosing a cloud backlog of approximately $460 billion, nearly double the prior quarter. That single contrast tells the whole story: the market now rewards the Funders who show receipts, and punishes the ones who only show plans.

One nuance worth noting is that even NVIDIA, despite being a direct AI supplier, was not spared because it is also part of the crowded Mag 7 trade. But outside the mega-cap group, several semiconductor and memory suppliers continued to outperform. 

How Big Tech’s $725 Billion AI Spending Benefits Chip Suppliers

These capex numbers are not projections or analyst estimates. The companies said them on earnings calls.

Source: Company Q1 2026 earnings calls; Financial Times analysis

Every dollar of this $725 billion lands in someone's income statement. NVIDIA captures the GPU spend. TSMC manufactures those chips under foundry contracts. SK Hynix and Micron supply the high-bandwidth memory that connects the chip to everything it needs to process. ASML builds the machines that allow TSMC to operate at cutting-edge scales. Broadcom handles custom silicon increasingly designed in-house by the hyperscalers themselves.

The Collectors do not need AI to generate profits for their customers. They only need the customers to keep ordering. And Goldman Sachs projects that total hyperscaler capex could reach $1.1 to $1.4 trillion by 2027.

The Pickaxe Principle: Why AI Chip Suppliers Are the Picks and Shovels of the AI Boom

There is a historical template to refer for this. During the California Gold Rush of 1848 to 1855, hundreds of thousands of prospectors flooded into California chasing gold. Most went broke. The people who built real fortunes were the merchants selling supplies to the miners. Levi Strauss arrived in San Francisco in 1853 selling dry goods and later riveted workwear. He never mined an ounce of gold. Sam Brannan became California's first millionaire by buying up mining equipment and reselling it at steep markups before the rush even properly started. He did not need to know which mountain held the most gold. He just needed miners to keep showing up.

The hyperscalers are today's miners, placing enormous bets on an AI payoff that the market is starting to question. NVIDIA, TSMC, SK Hynix, and ASML are the merchants, collecting regardless of which AI model ultimately wins.

Here is how few of the Collectors are performing right now.

Source: NVIDIA 8-K (SEC EDGAR, April 2026), TSMC IR, Broadcom Q2 FY2026 earnings release, ASML Q1 2026 results

The Memory Chokepoint: Why HBM Memory Stocks Are Gaining Pricing Power in AI

Within the Collector camp, memory is where pricing power sits in its most concentrated form.

Three companies control over 90% of global DRAM supply: Samsung, SK Hynix, and Micron. Their combined market share as of Q4 2025 was roughly 90.5% (Samsung approximately 36%, SK Hynix 32.1%, Micron 22.4%), as per TrendForce. And all three have been shifting production capacity toward high-bandwidth memory, known as HBM, which is what AI chips need to operate at the speeds the workloads demand.

HBM is not a commodity product. One HBM chip uses roughly three times the wafer area of a standard memory chip. So when a supplier converts a production line to HBM, it does not just tighten HBM supply. It shrinks conventional DRAM supply at the same time, because the same factory cannot produce both. Both sides of the market tighten simultaneously.

The result: DRAM contract prices rose approximately 93-98% quarter-on-quarter in early 2026, one of the steepest quarterly increases on record, as per TrendForce data. HBM capacity for 2026 is essentially sold out. SK Hynix alone has reportedly locked in roughly two-thirds of NVIDIA's HBM4 orders for its next platform.

Think of HBM like the UPI switch that NPCI runs in India. When PhonePe sends money to a GPay user, it does not matter which app is growing faster or which bank is more popular. Every transaction still has to go through NPCI's rails. The switch collects from all players regardless of who wins the app-layer battle. Whoever controls the chokepoint earns a premium on every transaction, every time.

Are AI Chip Stocks Still Cheaper Than Big Tech Stocks?

The general assumption is that fast-growing technology suppliers trade at premium valuations. That logic has partially inverted right now, and it is worth paying attention to why.

Source: GuruFocus, StockAnalysis.com, as of late June/early July 2026. 

NVIDIA and TSMC are growing revenue significantly faster than any of the Funders, yet they trade at similar or lower forward multiples in several cases. Broadcom at roughly 41x is the expensive outlier, pricing in an aggressive AI ramp. Meta at roughly 17x is the cheapest Funder, but it is cheap for a reason: the market has the least confidence in Meta's AI revenue story relative to the scale of what it is spending.

The pattern across the table suggests something specific. The market is currently placing more certainty on booked supplier revenue than on the returns hyperscalers promise to generate from their spending. That is unusual. Normally, faster growth commands a higher price. When it does not, there is a mispricing somewhere, and figuring out which side is mispriced is the entire question.

What Analysts Expect From AI Capex and Semiconductor Stocks

Source: Bloomberg, Yahoo Finance, Fundstrat research, deVere Group statements (Q2 2026)

The analyst community broadly agrees on the direction: hyperscalers will eventually see returns from AI spending, but the timeline is uncertain, and the market is no longer willing to wait without proof. The July and August earnings season will be the first serious test of whether that proof materializes.

What Could Hurt AI Supplier Stocks From Here?

The Collector thesis rests entirely on one assumption: the Funders keep spending.

If any of the four major hyperscalers cuts capex guidance at Q2 earnings, the supplier order books start to shift within weeks. Memory is the most exposed link. DRAM stocks have historically lost 40 to 60% within six months of a pricing peak. This current upcycle is already roughly 30 months old, which matches the length of the longest prior DRAM upcycle on record.

There is also concentration risk within the supplier indices. The top three holdings in SMH make up roughly 36 to 44% of the fund. TSMC carries Taiwan-China geopolitical risk. NVIDIA faces ongoing China export-control uncertainty. Neither of these is hypothetical.

On the Funder side, the ROI question has teeth. Microsoft's capex is running at levels that visibly compress its near-term free cash flow. Meta is partly debt-funding its AI build. If AI revenue does not scale fast enough to absorb those costs, the market will not extend patience indefinitely.

Watch for three numbers in the coming Q2 earnings season: cloud revenue growth rate, any new disclosure of AI-specific revenue, and whether full-year capex guidance holds, rises, or gets cut. Those three data points carry more signal than any analyst note right now.

Our Take: Are AI Suppliers Better Placed Than Big Tech?

The Funder vs Collector split is not a permanent new order in markets. It is a specific moment of skepticism toward spending without visible returns, running alongside genuine enthusiasm for the suppliers already turning that spending into real revenue.

On the Collector side, the thesis is grounded in reported results. NVIDIA, TSMC, Broadcom, and the memory names are not pricing in hope right now. They are reporting numbers that match the story. NVIDIA and TSMC stand out for combining rapid revenue growth with forward multiples that are not stretched given that growth. Memory names carry more cyclical risk given where we are in the pricing cycle and deserve to be sized accordingly.

On the Funder side, Alphabet has the strongest immediate case. It was the only hyperscaler that convinced the market its spending was working at April earnings, backed by a cloud backlog that nearly doubled in a single quarter. Meta at roughly 17x forward earnings is interesting if its advertising cash flows continue to fund the AI bet without compressing margins. Microsoft and Amazon are harder to argue for purely on near-term valuation at current capex levels, though both remain strong long-term businesses.

The clearest framing: if you believe AI infrastructure spending continues at current or higher levels through 2027, the Collectors are better-positioned for the next one to two years. If you think a capex slowdown is coming, memory and custom chip names absorb the shock first and fastest.

There is no single answer for everyone here. But the market is saying something specific right now: selling picks and shovels carries more certainty than finding gold. The last time people stopped listening to that lesson, they ended up washing dirt.