A Supply Shock Priced As A Demand Shock: Why We Believe AI ETF AGIX Is Well Positioned
By Max Chen &
Derek Yan, CFA
AI stocks have had a rough few weeks.
Semiconductors sold off, hyperscaler earnings sparked worries about spending, and headlines about data center permit freezes added to the anxiety. To many investors, it looks like the AI trade is running out of demand.
We believe the market is reading the signals backward.
What we are seeing is not a demand shock. It is a supply shock: there is more demand for AI computing power than the industry can currently build, power, and deliver. In our view, the market is pricing that scarcity as if demand were fading, and that disconnect may be creating an opportunity for long-term investors in the KraneShares Public-Private AI & Technology ETF (AGIX).
A Demand Collapse May Look Very Different
Start with what we believe a real demand collapse would look like. Prices for compute, the specialized chips and data center capacity used to run AI models, would fall. Capacity would sit idle. Order backlogs would shrink.
We are seeing the opposite.
Rental prices for Nvidia's H100 chips rose about 32% last month, and prices for its newer B200 chips rose 25%.1 IREN, an Australia-based cloud computing and AI infrastructure company, said this week that prices for its compute are still rising.2

Microsoft told a similar story. The company described capacity, not demand, as the main constraint on its growth, and its commercial remaining performance obligations (RPO) reached $677 billion, up 73%.3 RPO is revenue Microsoft has already been promised under signed contracts but has not yet earned, making it one of the clearest measures of demand still in the pipeline.
In our view, rising prices, full capacity, and a growing backlog are the signature of scarcity, not of a slowdown.
That does not mean the selloff in chip stocks was irrational. A slower U.S. buildout does reduce how many units of equipment and memory get shipped over time, so the drop in the Philadelphia Semiconductor Index (SOX) partly reflects investors repricing the slope of growth.
But scarcity has another effect: it raises the earning power of the capacity already installed. We believe that points to stronger revenue at lower valuations, not just in AI hardware but across the AGIX portfolio.
The numbers support this. Over the past 30 days, analysts have raised forward earnings estimates for AGIX's holdings, while the portfolio's forward price-to-earnings ratio, what investors pay for each dollar of expected earnings, has come down. Earnings expectations going up while valuations come down is, in our view, an attractive setup for long-term investors.

The performance data quoted represents past performance. Past performance does not guarantee future results. The investment return and principal value of an investment will fluctuate so that an investor's shares, when redeemed or sold, may be worth more or less than the original cost. Current performance may be lower or higher than the performance quoted. For AGIX performance data current to the most recent month-end, please visit kraneshares.com/etf/agix.
Permit Freezes Sort Winners From the Rest; They Don't Stop the Buildout
The second worry is power and permits. If new policies delay powered data center shells, then chip demand at the margin depends on how many megawatts can actually be deployed, and spending becomes a function of permits. That has led some investors to assume 2028 will be the peak for AI capital spending, with a downward slope after that.
We see it differently. In our view, a permit slowdown can act as a healthy brake, especially for a portfolio like AGIX.
Start with what a freeze can and cannot reach. A year ago, roughly 23 gigawatts (GW) of data center capacity was already under construction worldwide, about three-quarters of it in the U.S.4 So, a moratorium only governs what comes next: UBS estimates about 30% of expected 2028 additions are at risk.5
Now look at who is on the funded side of that line: Meta, Microsoft, Google, SpaceX, and Nebius.* These companies are building multi-gigawatt data centers with power plants they pay for themselves. They did not need Texas to say yes, because they bought their own power.
What may suffer is the speculative tail, and that tail is large. ERCOT, the operator of the Texas power grid, has 474 GW of large-load connection requests in its queue, about five times the grid's peak demand.6 Behind many of those requests are companies that raised significant capital but are behind schedule. Some projects in the connection queue may face challenges related to financing, customer demand, or deployment.
Those projects compete with the companies AGIX holds for the same turbines, transformers, and electricians. If projects are delayed or canceled, equipment could potentially become available to established operators. The 2028 supply glut that could have crushed AI computing prices may never arrive.
In other words, a brake on new permits is supply discipline imposed from the outside. It also helps answer two other bear cases: spending that cannot be deployed is spending that is not made, so hyperscaler free cash flow can recover sooner, and fleets that cannot be expanded tend to run longer at full utilization.
Look at the Tenants, Not Just the Landlords
The number that did the most damage this earnings season was free cash flow, the cash a company has left after paying for its operations and investments. In the September quarter, capital spending exceeded operating cash flow at three of the largest hyperscalers:
- Alphabet: free cash flow of –$9.1 billion7
- Amazon: –$33.5 billion8
- Meta: –$2.1 billion9
Only Microsoft stayed positive. What unsettled the market is that companies that funded a decade of growth from their own cash are now borrowing to build.
We think that is the wrong place to look.
The hyperscalers are the landlords. They build the space and the compute that hosts their tenants' demand, and a landlord's cash flow during construction is not where the health of the business shows up. The tenants' profits are.
As AI spreads, Fortune 500 companies are becoming those tenants, buying compute as part of what it costs to make their products. And the tenants are currently doing well: earnings for the 493 S&P 500 companies outside the Magnificent 7 appear to be growing 32.6%, compared with 22.6% for the Magnificent 7 themselves.10
For two years, the market has gauged AI demand from a single number: how fast one lab, Anthropic or OpenAI, is growing its annualized revenue. That measures one intermediary. The real story is the diffusion underneath it.
Pharmaceuticals is a good example. It is one of the slowest-moving, most regulated, highest-stakes industries in the economy, and it has started buying compute the way it buys manufacturing plants:
- Roche added 2,176 Nvidia Blackwell GPUs this spring, bringing its total above 3,500.11
- Lilly launched LillyPod with 1,016 Blackwell Ultra GPUs.12
- The Novo Nordisk Foundation’s Gefion supercomputer runs 1,528 GPUs.13
As more Fortune 500 companies buy their own compute and see real returns, we believe the sustainability of AI spending stops being a question only for the Magnificent 7.
What It Means for AI ETF AGIX
We agree the market needed a reset. The AI trade has run hard for two years, and leveraged liquidity has left the market structure vulnerable.
But a reset is not the end of the story. Rising compute prices, record backlogs, disciplined supply, and broadening demand from companies far beyond the Magnificent 7 all point, in our view, to a long runway ahead.
AI ETF AGIX is designed for a ten-year horizon, aiming to capture the long-term growth of what we believe is one of the greatest innovations in human history, and one that is still in its infancy.
After the turbulence, we expect growth to continue on a different slope: steadier, healthier, and, we believe, more durable.
Holdings are subject to change. *Meta, Microsoft, Google, SpaceX, and Nebius are AGIX holdings as of 10/31/2026.
For AGIX standard performance, top 10 holdings, risks, and other Fund information, please click here.
Citations:
- Data from Ornn, "GPU Market Prices," as of 11/13/2026.
- CNBC, "Neocloud IREN says $4 billion in annualized revenue being delayed due to customer request," as of 11/5/2026.
- LongYield, "The State of the AI Buildout: What Q3 2026 Earnings Season Actually Showed," as of 10/31/2026.
- BloombergNEF, "AI Data Center Build Advances at Full Speed: Five Things to Know," as of 3/24/2026.
- Yahoo Finance, "Wall Street May Have Lost $2.5 Trillion, but This 30% Warning Could Be an Even Bigger Problem for AI Stocks," as of 11/7/2026.
- Dave Manuel, "What a Democratic Texas Governor Means for the Data Center Boom," as of 11/4/2026.
- SEC Archives, "Alphabet Announces Third Quarter 2026 Results," as of 10/28/2026.
- Fidelity, "Amazon Sees AWS Growth Above 40% in 2027, 2028, Wedbush Says," as of 10/30/2026.
- Meta, "Meta Reports Third Quarter 2026 Results," as of 10/21/2026.
- FactSet, "EARNINGS INSIGHT," as of 10/30/2026.
- Roche, "Roche launches NVIDIA AI factory to accelerate the development of new therapeutics and diagnostics solutions," as of 3/15/2026.
- Nvidia, "Now Live: The World's Most Powerful AI Factory for Pharmaceutical Discovery and Development," as of 2/26/2026.
- Nvidia, "Denmark Launches Leading Sovereign AI Supercomputer to Solve Scientific Challenges With Social Impact," as of 10/23/2024.




