When you delay data center construction, the consequences spread far beyond the construction site.
There’s the contractors, who are now in waiting mode. There’s the utility’s expected electricity sales, which move further into the future. What’s more, cooling equipment may sit undelivered.
But the chips intended for that facility can still ship somewhere else.
That difference is where I’m looking for opportunity in the AI selloff.
As I write this, major indices are climbing while AI stocks are getting hammered.
After the midterm “Bluenami,” investors are reassessing how quickly America can build the facilities needed to power AI. My assessment is that construction will slow over the next 12 to 24 months, forcing earnings estimates lower across the AI supply chain.
But the damage will be uneven. Understanding that distinction can help us identify where to reduce exposure, and where the selling may create opportunity.
Construction firms, power providers, and suppliers of cooling and electrical equipment depend heavily on particular projects moving forward. When those projects stall, their expected revenue can move further into the future.
Chip, networking, memory, and optical component suppliers have more flexibility. If a planned facility gets delayed, some equipment can go to another location where construction is proceeding.
That does not spare those suppliers from a slowdown, but it gives them more ways to keep serving demand.
Tower Semiconductor (TSEM) illustrates the opportunity. Its customers committed $290 million in prepayments to reserve manufacturing capacity, supporting contracts covering $1.3 billion in silicon photonics revenue for 2027. Those commitments provide evidence of demand, even as we reassess when shipments and earnings will arrive.
There is also a potential benefit to a more measured expansion: Less pressure to build everything immediately could ease supply shortages and reduce the need for aggressive financing. Near-term growth expectations may come down while the buildout becomes more durable.
That is why I see an AI trade that is bent, rather than broken.
In this report, I’ll examine Tower alongside Nebius Group (NBIS), Cloudflare (NET), Fortinet (FTNT), and Fastly (FSLY). They require different investment cases.
For example, Nebius must finance and deliver physical computing capacity. While Cloudflare, Fortinet, and Fastly offer exposure to application delivery, security, and adoption, with their own valuation and execution risks.
My strongest conviction remains in the core components that can follow AI spending as its destination changes.
Read on for the business evidence, buying conditions, and risks that distinguish these five opportunities.
The Buying Framework
The first question I ask during a selloff is what changed inside the business.
If customers are canceling orders, financing is becoming unavailable, or earnings expectations are collapsing, a lower stock price may be justified. If the underlying opportunity remains intact while investors become less willing to pay for it, the same decline can create a better entry.
I want to understand who is paying, how dependable that demand is, and what remains after the bills are paid.
For a manufacturer, that means studying customer commitments alongside the cost of adding capacity. For a cloud operator, it means matching contracts to financing and delivery obligations. For a security company, it means checking whether customer growth is producing durable revenue and cash flow.
Then there’s valuation… Familiar consumer brand can become expensive when too many investors crowd into it for safety. A technology company can become attractive when its price falls faster than its earnings outlook. The sector label alone does not settle the question.
Nor does every stock below its high qualify as a bargain. Some of the names in this report are growth opportunities whose valuations still require strong execution. They should not all be described as cheap on earnings.
Finally, on price action, a stock approaching an attractive level is not the same as one finding buyers there. I’m especially mindful of that distinction with Fortinet.
Let’s get into it:
Five Different Ways to Participate
Nebius: A High-Upside Bet That Requires Strong Execution
An AI cloud company can have customers waiting for every available GPU and still face a difficult question: How does it pay for the next batch and get them running on time?
That is the challenge facing Nebius Group N.V. (NBIS). It is also what makes the stock so sensitive to expectations for AI spending.
In our Election Day discussion on Being Exponential, I named Nebius as my highest-upside idea for a potential AI rebound in 2027. The reasoning was straightforward: If major technology companies keep investing and fears of widespread spending cuts ease, investors could become much more willing to pay for Nebius’ expected growth.
But I also emphasized that it was not my preferred way to balance potential returns against risk. Nebius could deliver substantial gains if that outlook holds — and suffer substantial losses if it does not.
The midterm results add another consideration. We now expect data-center construction to slow over the next 12 to 24 months. Demand for AI computing can remain strong while the facilities needed to serve it take longer to finish.
For Nebius, that distinction matters enormously. It sells working computing capacity. Facilities, electricity, and equipment must all be ready before customers can use what they have ordered.
Its customer commitments provide reasons for confidence. Microsoft signed a roughly $17 billion multiyear agreement. Meta committed $12 billion for dedicated capacity, with an additional commitment covering up to $15 billion of available capacity. Management has said committed backlog exceeds $40 billion.
Customers also help pay for expansion. Our previous coverage highlighted management’s estimate that prepayments cover roughly 50% to 60% of capital spending associated with new contracts. Access to secured financing, substantial cash, and Nvidia’s backing further distinguish Nebius from weaker operators.
Those advantages support our original thesis: A shakeout among AI cloud providers could leave the strongest competitors with a larger share of the market.
They do not remove construction risk. A chip shipment can change destinations more easily than a contracted facility can relocate. Delayed permits, power connections, or construction could push revenue back while costs continue.
Nebius remains a potentially powerful rebound opportunity, but its financing and delivery schedules must support the growth investors are paying for.
After the midterms, I would give those schedules as much attention as new customer announcements. Strong demand supports the opportunity; turning that demand into operating capacity determines the results.
Tower: The Optical Components That Can Follow AI Spending
An AI data center can contain thousands of powerful processors and still struggle if those processors cannot exchange information quickly enough.
Tower Semiconductor Ltd. (TSEM) helps solve that problem.
Tower is a specialty semiconductor foundry. Customers design chips, and Tower manufactures them using specialized processes. Its growing AI opportunity is silicon photonics — technology that helps move information using light, supporting faster connections with greater energy efficiency.
That makes Tower the clearest fit among these five names with our post-midterm preference for chips, networking, memory, and optical components.
We expect a slower buildout, with some investment shifting to other locations. Components can follow that investment to facilities that are ready for them. Tower’s opportunity therefore depends more on overall demand for advanced connectivity than on any single construction site.
This is relative protection, not immunity. A broad slowdown can delay orders and reduce earnings expectations even for well-positioned suppliers.
Still, Tower has substantial customer commitments behind its growth plans. In May, it announced contracts covering $1.3 billion of silicon photonics revenue for 2027, supported by $290 million in customer prepayments. Management also reported larger contractual commitments for 2028.
Customers are committing money to secure manufacturing capacity. That provides stronger evidence of demand than forecasts alone.
Tower’s specialized expertise adds another attraction. Producing optical components reliably at scale takes years of development and customer qualification. Once customers design products around a foundry’s technology, changing suppliers can be costly and time-consuming.
Previously, we put shares at roughly 15 times forward EBITDA, a measure of operating earnings, against a long-term model of 35% to 40% annual EBITDA growth through 2030. Those estimates predate the election results and need to be assessed against the slower construction outlook.
Tower also has its own factories to expand. Customer commitments must become reliable production, shipments, and profits.
The attraction is a specialized supplier with committed customers and the flexibility to serve AI investment across locations. We want to see that demand translate into earnings, even if the path takes longer than previously expected.
Cloudflare: Helping Businesses Put AI to Work
Once a business launches an AI application, customers expect it to respond quickly, stay available, and protect their information.
Cloudflare Inc. (NET) helps businesses meet those demands through its global network, security services, and developer tools.
Workers AI lets developers run models on Cloudflare’s infrastructure. AI Gateway helps manage requests to model providers, including monitoring usage, controlling traffic, and routing around failures.
For a business running an AI assistant, those capabilities can help keep the service working and avoid unnecessary computing costs.
Cloudflare’s case rests on businesses using more of these services as they deploy applications. That is a different opportunity from supplying chips for the next data-center project.
The distinction is important after the midterms. Slower construction does not eliminate the need to operate and protect applications already in use. But it does not automatically benefit Cloudflare, either. The company runs physical infrastructure, invests in its network, and faces its own capacity and equipment costs.
I want to see developers become paying customers, existing customers expand their spending, and that activity produce durable revenue and cash flow.
We previously identified NET around $250 as attractive. That was a valuation judgment at that time, rather than a permanent floor. A premium share price can fall sharply if growth disappoints, even when the underlying business continues expanding.
Cloudflare offers exposure to AI adoption and application services. Its inclusion here depends on those opportunities translating into stronger financial results, alongside an entry price that leaves room for attractive returns.
Fortinet: A Security Need That Extends Beyond the AI Buildout
A delayed data center does not remove the need to protect offices, factories, cloud applications, and networks already operating.
That is the starting point for Fortinet Inc. (FTNT).
In Being Exponential, I highlighted cybersecurity as an area worth examining because of AI safety and security concerns. I also distinguished that opportunity from the investment case built around continued AI infrastructure spending.
Fortinet fits that separate security argument.
Businesses have more systems to protect, and managing a different tool for every task can become complicated. Fortinet brings many of those functions together, helping customers manage protection in one place.
Its reach extends beyond office computers to systems connected to factory equipment and other industrial operations, where a successful attack can interrupt physical production.
Our previous coverage highlighted third-quarter 2026 revenue of approximately $2.08 billion, up 20% from a year earlier. Product revenue grew 37% to $765 million, while the operating margin was about 34%.
Those results describe an established, profitable business. They give us concrete measures to follow: equipment demand, services growth, and continued profitability.
Customers can still delay upgrades, and competitors can win business. An essential service does not make its provider attractive at every share price.
Fortinet could be worth buying on a turn near $130. Simply falling to that price would not meet the condition. We want evidence that buyers are returning and that the business outlook supports the purchase.
Fortinet’s appeal rests on a continuing need for protection, with potential additional demand as AI makes technology environments more complicated. That case deserves its own assessment alongside our AI component investments.
Fastly: Improving Operations, With More to Prove
Users rarely think about the infrastructure behind an application. They notice when it loads slowly, stops responding, or becomes unavailable.
Fastly Inc. (FSLY) helps businesses deliver and protect those services. Its network serves and processes traffic closer to users, while its security products protect websites, applications, and software connections.
The strongest reason to examine Fastly is the improvement in its own operations.
In the second quarter of 2026, revenue rose 23% year over year to $183.3 million. Security revenue increased 43% to $41.7 million, and GAAP gross margin expanded to 63.3% from 54.5%.
Those results support the view we expressed on Oct. 26 that Fastly looked fundamentally attractive. Security gives it more opportunities to deepen relationships with customers already using its delivery services.
Progress was uneven, however. Fastly reported a $15.6 million GAAP net loss. Free cash flow was positive at $3.6 million, but below the prior-year quarter’s $10.9 million.
We want better growth and margins to produce more consistent cash generation.
AI adoption could create additional demand for Fastly’s services, but we should not assume that continued spending by the largest technology companies flows directly into its revenue. Customers must choose Fastly’s products, use them, and generate enough business to improve profitability.
That keeps Fastly’s investment case distinct from our strongest conviction in AI chips and related components. It is an operating improvement story with potential benefits from application growth and security needs.
Competition, customer concentration, and pricing remain important risks. Large customers can change how they distribute traffic, and higher usage does not automatically produce higher profits.
Our earlier comments did not specify a buy-below price. The decision therefore rests on the operating evidence and valuation available when buying.
Fastly deserves attention if its progress continues — particularly if revenue growth, margins, and cash generation begin improving together.
Adjust the Exposure, But Keep the Conviction
The AI trade may be bent for now, but our core thesis remains intact: Demand for computing power is growing, and more capacity will be needed to serve it.
Now that the midterms are settled, I expect that capacity to arrive at a slower, more measured pace over the next 12 to 24 months. Earnings estimates across the AI supply chain will need to adjust, potentially including projections further into the decade.
Our response is to reduce exposure to businesses most dependent on new data-center construction and increase exposure to the chips, networking, memory, and optical components that can follow investment to other locations.
Those suppliers can still face delays. But they have more flexibility when a customer changes where it plans to build.
Among the five names in this report, Tower most directly fits that preference. Nebius does require separate judgment, but its customer commitments and financing relationships are strengths. Those aside, it must still bring physical capacity online. Cloudflare, Fortinet, and Fastly offer opportunities tied to application delivery, security, and adoption, each with its own operating tests and entry conditions.
Just note that we should not treat them as equally insulated from the slowdown, or buy them simply because their shares have fallen.
There is also a reason to remain constructive beyond the immediate selloff.
A steadier buildout could ease pressure on power supplies, equipment availability, and financing. Lower near-term growth expectations can coexist with a more durable expansion and stronger foundations for earnings in 2028, 2029, and beyond.
That is why buying selected AI stocks can make sense alongside reducing other positions in the same theme.
The opportunity is to concentrate on the businesses best positioned to keep serving AI demand as the pace and location of spending change. We can maintain conviction in the technology while being disciplined about which stocks we own, what we pay, and the evidence needed to keep owning them.
We hope you found this special report useful. Before we go, let us remind you that you’re now also a member of my free Hypergrowth Investing newsletter.
And this is the work I’ll keep doing in Hypergrowth Investing: following the companies behind the buildout, explaining what changes in their businesses, and looking for opportunities when the market’s expectations move too far.
By uncovering early investments in hypergrowth industries, we put you on the ground floor of world-changing megatrends. Keep an eye on your email inbox for my next Hypergrowth Investing article soon. I typically send them every day of the week. In the meantime, you can check out our website by clicking here.
