AI Infrastructure Opportunities at Disrupt 2026
Explore how AI’s growth is driving demand for power, data centers, and new infrastructure businesses. Meet Ambrosia Energy and Bloom Energy at Disrupt 2026.
AI's Next Big Challenge Is Physical - and That Creates Opportunity
AI tools may run in software, but the systems behind them don't. Training and operating AI models takes data centers, reliable power, grid connections, cooling, electrical equipment, and the people and processes needed to keep it all running. As AI adoption grows, those physical requirements are harder to treat like background noise.
This shift is the focus of a session announced for the Smart Systems Stage at TechCrunch Disrupt 2026. Ben Longmier, CEO of Ambrosia Energy, and Bill Thayer, SVP and Head of Datacenter Solutions at Bloom Energy, will discuss where the AI infrastructure boom is creating real opportunities. The discussion is expected to look at constraints across the stack, the new business categories those constraints could unlock, and what founders should watch as AI moves deeper into the physical world.
For founders, investors, and technology leaders, the useful question isn't just whether demand for AI infrastructure will rise. It's where that demand turns into something durable - and where a temporary shortage gets mistaken for a long-term market. Below is the context for the session and a practical way to think about where to look.
Why AI growth is becoming an infrastructure problem
Most of the public talk around AI focuses on models, chips, and applications. Those parts matter - but none of them runs by itself. A model needs compute. Compute requires a data center. A data center needs power, cooling, network connections, and equipment that can be delivered, installed, and maintained. If any step in that chain falls behind, it limits what the rest of the system can do.
That's the difference between an AI product roadmap and an AI infrastructure plan. A software team can often ship updates without building anything new. Expanding the physical capacity for AI workloads takes longer, requires capital, depends on local infrastructure, equipment availability, and a set of operational realities you can't fix by releasing code.
Compute demand depends on more than chips
It's tempting to treat a compute shortage as "a chip problem." But the question is broader: can compute actually get put to work if the supporting infrastructure isn't ready? A new accelerator doesn't help if the power connection isn't there. More data-center capacity doesn't matter if cooling, electrical equipment, or other systems become the bottleneck.
Constraints also tend to move. A team may solve one limiting factor, only to find another part of the stack slows deployments. For people building or investing in infrastructure, that's more actionable than asking "What's the hottest AI tech right now?": Which dependency is slowing real capacity, and what would it take to remove it?
A bottleneck isn't automatically a business. It becomes one when customers have an urgent operational problem, current options don't fit, and a new solution can be deployed and supported at a cost customers will actually pay.
Power and data centers make the physical layer visible
Power generation and grid connections are central because data centers need a dependable electricity supply. Data centers also need cooling and electrical systems designed for how they run. These aren't optional "extras." They're part of the conditions that let compute capacity function.
So more AI capacity planning is now also energy and physical infrastructure planning. That doesn't mean every AI company should become an energy company. But founders and technology leaders may need to understand the dependencies their products create - and investors may need to look beyond software adoption and model performance.
The session featuring Longmier and Thayer is built around that wider view. Instead of treating AI like software alone, the discussion will focus on the infrastructure beneath it and the pressures that could shape business opportunities.
Where constraints can turn into business opportunities
Infrastructure booms pull in capital and attention, but they don't spread returns evenly. Some areas get steady demand; others see a quick surge that fades once capacity catches up. The challenge is telling the difference between a real, recurring need and a headline-driven opportunity.
The AI infrastructure stack touches several connected areas, including:
- Energy and power: supplying electricity for expanding computing workloads.
- Grid access and connections: bringing power to the sites where facilities are built.
- Data centers: providing the physical environment for compute, networks, and operations.
- Cooling: managing heat as equipment runs.
- Electrical equipment: supporting power distribution and control inside facilities.
- Infrastructure software and services: coordinating, monitoring, or operating complex physical systems.
This isn't a prediction that every category will produce a successful startup. Think of it as a map of dependencies. Each part of the stack tends to have different buyers, timelines, technical requirements, and barriers to entry. Something that's easy to sell in one area may be much harder to sell in another, even if the overall "AI infrastructure boom" story sounds similar.
A shortage is a signal, not a business model
When something is scarce, it's easy to assume that anyone who supplies more will build a lasting company. Often, that's skipping key questions. The shortage might be local, not universal. It might be temporary. Customers may have workarounds. Or the cost to solve the problem might be higher than what customers can (or will) pay.
If you're a founder weighing an infrastructure idea, start with a concrete test: Who has the problem, how does it disrupt their operations, and what are they doing today? Interviews and pilots can help - but they need to reveal actual buying conditions, not just enthusiasm about AI growth.
Next, test the basics of deployment. Can the product be installed, integrated, and maintained in the environments customers actually use? In physical infrastructure, even a technically solid solution can fail if deployment takes too long, depends on scarce components, or assumes an operating model customers can't sustain.
Finally, ask whether the problem will last. A company built around a temporary gap may have to reinvent itself once supply improves. A company built around a durable operational need has a better shot at repeatable demand - assuming it can deliver reliably and earn acceptable margins.
New categories may form between established industries
Not every opportunity fits cleanly into existing labels. AI connects software teams, energy providers, data-center operations, grid systems, and equipment vendors. New businesses can appear at those boundaries - especially when teams need to coordinate systems that used to be planned and managed separately.
That's worth watching, but it's not automatic. A new category isn't valuable just because it's new. It has to solve a customer problem someone recognizes, fit into real procurement and operations processes, and offer a meaningful improvement over what's already available.
One reason to bring people from energy and data-center solutions into the same room is that it can reveal how these dependencies interact. For founders, that perspective helps surface problems that get missed when AI is treated only as an "application" or "model" story. For investors, it can help distinguish real operational demand from generic market hype.
What founders and investors should look for
The most useful response to an infrastructure boom isn't trying to chase every area getting investment. It's to look at the system, identify where customers are constrained, and test whether a solution can be delivered repeatedly. It's less flashy than pitching a big market size - but it tends to lead to better decisions.
A practical assessment can start with four questions:
- Is there a specific constraint? Name the resource or process limiting a customer's ability to operate or expand. "AI needs more infrastructure" is a trend, not a customer problem.
- Who controls the budget and the decision? The person feeling the pain might not be the buyer. Figure out who evaluates the solution, who pays, and how procurement actually works.
- Can the solution work in the real environment? Consider site requirements, integration, maintenance, reliability, and the time needed to get from contract to live.
- Does the value hold if conditions change? Pressure-test the business against different demand and supply scenarios. If the shortage ends, what value remains?
Founders: validate the problem before building the category
Founders hear that a market is big and assume customers will rush to buy. In infrastructure, that's a dangerous assumption. Projects can require planning, physical installation, multiple stakeholders, and coordination with existing systems. The sales cycle and delivery cost can look very different from software-only products.
Start with a narrow customer problem. Learn how often it happens, what it costs, what workaround exists, and what would make someone actually switch. Then test delivery, not just interest. A pilot that shows measurable operational impact is more useful than broad support for the AI infrastructure idea.
Also, be clear about what's proprietary. A company can build a valuable service or deployment model without owning every component in its system. That can still be a real business - but you have to understand the economics and defensibility, not assume them. The goal isn't to force every company into a software-style growth model. It's to build something customers will keep paying for and a team can run sustainably.
Investors: assess durability, not just demand
Investors should look beyond forecasts of rising compute needs. Demand is only one side. The other sides include customer economics, how quickly solutions can actually be deployed, how available alternatives are, and whether the company can deliver consistently as it scales.
A good diligence conversation asks what needs to be true for the business to scale. Does success depend on a handful of big projects, or can delivery become repeatable? Are customers paying for a real problem happening now, or for a promise tied to uncertain future growth? And what happens if the constraint shifts elsewhere in the infrastructure stack?
None of these questions argues against investing in physical infrastructure. They just make the pitch more specific. Strong opportunities can exist in capital-intensive markets - but the economics, execution risk, and time horizon should be evaluated on their own terms, not borrowed from software assumptions.
Technology leaders: plan for dependencies early
For technology leaders, infrastructure constraints can turn into product constraints. If a roadmap assumes computing capacity that won't be available on time, product plans and customer commitments may have to change. That makes infrastructure planning relevant well before you reach a crisis.
Teams can improve decisions by making assumptions explicit: what capacity is needed, when it's needed, which dependencies could affect delivery, and what alternatives exist. That doesn't eliminate uncertainty - it just makes it visible enough to plan around.
The same discipline applies to claims about AI's physical footprint. Good planning should start with actual workloads and deployment needs, not a generic belief that "all AI use" will create the same infrastructure demand. The requirements for any product depend on how it's built, how it's used, and how it's operated.
What to expect at Disrupt 2026
At TechCrunch Disrupt 2026, Ben Longmier of Ambrosia Energy and Bill Thayer, SVP and Head of Datacenter Solutions at Bloom Energy, are scheduled to join the Smart Systems Stage for a session titled "Where the AI Infrastructure Boom Is Creating Winners." The announced discussion will focus on constraints that create business opportunities, emerging categories, and what founders can build as AI moves further into the physical world.
This session isn't only for energy or data-center specialists. Founders can use it to find problems beyond the application layer. Investors can use it to think through where infrastructure demand could support defensible businesses. Technology leaders can use it to connect product growth with the physical systems that make deployment possible.
It also helps to keep expectations realistic. A conference session can surface questions, compare perspectives, and point you toward what to investigate. It can't prove that every infrastructure category will grow at the same pace - or that a specific company will benefit. That needs customer evidence, operational data, and a clear understanding of the economics.
The announcement says Disrupt 2026 includes more than 250 speakers and 200 sessions. It also states that attendees can save up to $100 by registering before rates rise at the door on October 13, and get 50% off a second pass of the same ticket type. The event is scheduled to open at Moscone West at 8 a.m. PT on October 13. Check the original event announcement for current registration details and terms.
Conclusion
AI infrastructure is turning into a business question because software growth increasingly depends on physical systems that take time and capital to build. Power, data centers, cooling, grid connections, and the equipment to run them aren't optional side issues for AI deployment. They shape where and how deployment can happen.
But seeing a constraint isn't the same thing as finding an opportunity. Durable businesses still need a real customer problem, a deployment path that works, and economics that still make sense when market conditions change. That's the framework to bring into the Disrupt 2026 conversation with Ambrosia Energy and Bloom Energy.
If you're attending, use the session to pressure-test your assumptions about where demand is likely to persist - and where the "bottleneck" is just a temporary bottleneck that shifts elsewhere. Read the event details and registration offer before you book. The point isn't just to collect a list of hot sectors. It's to understand which physical-world problems are actually worth solving - and what it takes to solve them repeatedly.