Mid-June trading looked like a riddle. The S&P 500 was green, but a chunk of mega-cap tech sat red. Meanwhile, lesser-known chip suppliers and builders of the AI plumbing ripped higher. Same theme, totally different P&L stories.
That gap isn’t a mood swing. It’s the market finally pricing who’s getting paid right now for AI, and who’s footing the bill. The split is turning into a defining feature of 2026 equity performance.
Put simply: the folks selling shovels are cashing in before the prospectors strike gold. And the index is starting to tilt in their favor.
The Big Picture
Editor’s note: Hyperscalers kept talking capex intensity while CFOs hinted at slower monetization curves. Micron’s print cut through the noise: scarcity still rules. I’ve been watching ASP commentary and lead times more than headlines; that’s where the next twist usually shows up. — Andrei Popescu
Two parallel truths can both be right. AI demand is still surging. But the near-term profits are landing with suppliers of chips, memory, networking gear, and contract manufacturing capacity, not with the platforms spending billions to deploy it.
AI’s land rush pays suppliers first and platform owners later. Markets are finally treating those timelines as different investments.
We’ve got hard evidence. Micron posted a blowout quarter, reporting record fiscal Q3 revenue of $41.46 billion and GAAP net income of $28.24 billion, underscoring how AI memory demand is translating straight into cash flow for suppliers (Micron Technology (press release / GlobeNewswire)).
At the same time, index mechanics are pouring fuel on the trade. S&P Dow Jones Indices is adding Marvell Technology and Flex to the S&P 500, a move that mechanically pushes passive capital into AI infrastructure names when the rebalancing hits (TechTimes). CoreWeave is joining the Nasdaq-100 as well, another signpost for the flow of index money toward the suppliers building data centers and racks (CoreWeave (press release / BusinessWire)).
Where the AI profit pool actually sits
In this cycle, it’s not mysterious. The best near-term economics sit with the companies that control scarce components and throughput, not with the buyers trying to deploy them at massive scale.
Gross margins and pricing power
Suppliers with must-have parts can raise price, allocate supply to the highest-margin buyers, and collect prepayments. Memory is tight. High-speed networking is tight. Advanced packaging is tight. As long as that’s true, their P&Ls keep smiling.
Platform owners, by contrast, are still in “build first, monetize later” mode. Training bills hit now. Monetization lags. If you’re a cloud provider, you’re often subsidizing access to keep developers and enterprises inside your ecosystem while you figure out the right pricing model. That compresses margins.
Inventory cycles and capacity
When supply is constrained, purchase commitments, capacity reservations, and long-term agreements flow to the suppliers. That supports visibility. When capacity finally loosens, those same suppliers feel the pain first. We’re not there yet in most critical nodes of the AI stack.
| Group | Primary revenue driver | Capex burden | Near-term margin trend | Key 2026 catalysts |
|---|---|---|---|---|
| Chip suppliers (memory, accelerators, networking) | Component ASPs, volume allocation | High but leveraged to utilization | Expanding where supply tight | Capacity ramps, pricing discipline, record prints (e.g., Micron’s FQ3) |
| Contract manufacturers/integrators | Build-out volume, integration services | Moderate; scale-driven | Stable to improving | S&P 500 inclusion flows (e.g., Flex) |
| Cloud hyperscalers/platforms | AI services, subscriptions, ads uplift | Very high; multi-year | Under pressure near term | Pricing changes, product adoption, inference cost declines |
| AI startups/integrators | API usage, enterprise contracts | Variable; opex-heavy | Mixed; burn vs. growth | Procurement wins, model breakthroughs |
Passive flows and the index effect
We underestimate how much index plumbing moves money. When a name joins a major index, passive funds have to buy. Actives often front-run that. The effect can be sizable when the company sits in a narrow theme like AI infrastructure.
What just happened
- AI compute scarcity drove a two-year scramble to lock in supply. Suppliers gained leverage and pricing power.
- Capex budgets at cloud providers swelled, with the market tolerating the spend while rates drifted and growth stayed intact.
- S&P Dow Jones Indices said it would add Marvell Technology and Flex to the S&P 500 effective June 22, pulling mechanical demand into these tickers (TechTimes).
- CoreWeave was selected for the Nasdaq-100 with the same effective date, channeling passive flows to another infrastructure-heavy name (CoreWeave (press release / BusinessWire)).
- The spread between suppliers and spenders widened as flows chased what’s working and earnings confirmed the skew.
Is this all flows? No. The flows just amplify what earnings already said. Micron’s record quarter didn’t happen because of index adds. It happened because AI memory demand is real and price-sensitive buyers paid up when supply was tight (Micron Technology (press release / GlobeNewswire)).
Volatility check: the Broadcom wobble
Even within the winners, the road is bumpy. In early June, Broadcom shares dropped roughly 13% to 16% on guidance read-through worries, and it pulled the wider semi complex down for a few sessions. The message wasn’t “AI is dead.” It was “expectations got ahead of the cadence” (TechPulseGlobe).
What that tells us
First, the market is hypersensitive to second-derivative changes. If orders are still strong but growth is slowing a touch, stocks priced for perfection can slip fast. Second, these businesses are exposed to a handful of procurement desks. A big buyer pushing deliveries by a quarter can look like a cliff in near-term numbers, even if the multi-year thesis is intact.
So yes, suppliers are winning. But they are not bulletproof. Guidance and backlog quality still drive the tape day-to-day.
Why Big Tech is bleeding near-term
For the platform giants, AI is a cost center before it’s a profit center. That’s not a value judgment. It’s just how deployment curves work.
AI opex the market ignores
Everyone sees the capex line. Less visible are the opex and opportunity costs: talent, data curation, safety reviews, go-to-market, compute for experiments that never ship. Those don’t show up as “AI revenue,” but they do show up as margin drag.
Unit economics still forming
Monetization levers exist, but they’re not mature. Price per token, per seat, per workload? Bundled into enterprise suites or charged as usage? The answers matter a lot. Until customers internalize clear ROI on AI features, pricing pushes face friction. Meanwhile, inference bills pile up.
Ad, cloud, and software mix shifts
Even if AI lifts engagement or upsells, it can move mix in unhelpful ways in the short run. If the revenue arrives as lower-margin compute rather than high-margin software, blended margins fall. That’s especially visible when the spend happens in one big gulp and the revenue drip-feeds in later.
From chips to cables: who’s actually minting cash
Follow the bottlenecks. Memory. High-bandwidth interconnects. Advanced packaging. Liquid cooling. Rack integration. The players positioned here are scooping up dollars while cloud providers and labs battle each other for delivery slots.
Memory is the tell
Micron’s record FQ3 print wasn’t subtle: AI workloads are memory monsters, and the suppliers are meeting the moment with price and volume leverage (Micron Technology (press release / GlobeNewswire)).
Networking and custom silicon
Marvell’s S&P 500 inclusion doesn’t make it a better business by itself, but it confirms something the market already knew: the plumbing layer is strategic, and buyers will pay for throughput (TechTimes).
Systems, racks, and the factory floor
Flex landing in the S&P 500 plugs the contract manufacturing angle right into the benchmark. If you’re building whole racks and integrated systems for hyperscalers, 2026 has treated you far kinder than if you’re paying those invoices (TechTimes).
What could close the gap
This spread isn’t carved in stone. It’s a function of timing, supply, and unit economics. A few developments could shrink it.
Better AI unit economics at the platform layer
If platforms prove clear ROI on AI features and shift customers to durable, premium-priced tiers, they can flip AI from cost sink to margin tailwind. That requires not just features, but usage that sticks.
Supply normalization in the bottlenecks
As memory and advanced packaging add capacity, pricing power eases. That can take supplier margins down a notch and move more value to the buyers who can deploy at scale.
Cheaper inference and smarter orchestration
Software improvements, sparsity, and workload routing can cut inference costs per unit of utility. If platforms deliver the same outcomes with less compute, they keep more of the revenue.
Macro and rates
If rates fall or stabilize and growth re-accelerates, investors may be more forgiving of front-loaded AI spend. Conversely, tighter financial conditions would punish long-dated payoffs and keep pressure on spenders.
Risks & What Could Go Wrong
- Overbuild risk: if buyers double-order and capacity floods in 2027, suppliers could face inventory corrections and price resets.
- Export controls and geopolitics: license changes can whipsaw order books and reroute supply chains quickly.
- Yield or packaging hiccups: a technical bottleneck can stall shipments and hit both revenue and credibility.
- Energy and grid constraints: delayed data center projects shift revenue timing and dent platform ROI.
- Index crowding: forced passive buying can reverse just as fast if the name drops out or flows slow.
- Valuation air pockets: perfection pricing makes even good prints look “not good enough,” as the Broadcom move reminded us (TechPulseGlobe).
Being on the right side of the AI trade didn’t cancel risk; it just changed which risks matter most.
If you want a steady, sober read on these moving parts, Crypto Daily tracks the macro-to-micro of AI, semis, and the broader digital economy without the hype. Their briefings are handy on rebalance weeks and earnings days: cryptodaily.co.uk.
Frequently Asked Questions
Are chip suppliers guaranteed to keep outperforming Big Tech?
No. They’re winning now because of scarcity and pricing power, but that can fade as capacity ramps. If platform monetization improves while supply loosens, the edge could shrink or reverse. This is not financial advice.
How do index inclusions like Marvell and Flex affect prices?
When a company joins a major index, passive funds tracking that index must buy shares. That mechanical demand can lift prices into and around the effective date. But it doesn’t change the company’s fundamentals by itself (TechTimes).
Why did Broadcom drop if AI demand is still strong?
Stocks priced for perfection react to small shifts in guidance or growth cadence. Broadcom’s early-June slide showed how sensitive the group is to read-throughs and order timing, even if the multi-year demand picture stays intact (TechPulseGlobe).
What numbers matter most in the supplier reports?
Watch average selling prices (ASPs), backlog quality, lead times, and capacity additions in tight nodes. Memory and networking suppliers with firm pricing and disciplined allocation tend to show it clearly in gross margin and free cash flow.
What should I track in Big Tech earnings to gauge AI payback?
Look for explicit AI revenue disclosures, pricing changes on AI features, inference cost per unit, and commentary on customer ROI. If AI feels less subsidized and more like a paid, sticky product, margins should tell the story.
Does this AI split affect crypto or miners?
Indirectly. Data center buildouts and power markets compete with mining for energy and rack space in some regions. If AI capex soaks capacity, miners can face higher costs or slower expansions. The macro read-through matters more than a direct revenue link.
Is Micron’s record quarter a one-off?
It reflects current scarcity in AI memory. Whether it’s repeatable depends on how fast capacity ramps and whether demand holds through 2027. For now, it’s a clean datapoint that the supplier side is monetizing AI demand (Micron Technology (press release / GlobeNewswire)).
Disclaimer: This article is provided for informational purposes only. It is not offered or intended to be used as legal, tax, investment, financial, or other advice.
Be the first to comment