The Next Great AI Winner May Have Almost Nothing to Do with AI

The AI investment boom has so far followed a relatively familiar path.

Hyperscalers committed enormous sums to building data centres. That expenditure became revenue for semiconductor designers. Demand for more powerful processors then spread into high-bandwidth memory, DRAM, advanced packaging and semiconductor equipment.

Investors followed the money. Nvidia surged. Memory manufacturers rerated. Cooling systems, transformers, turbines and electrical equipment were subsequently identified as the next beneficiaries.

But that raises a more interesting question.

If cash flow has already transferred from the hyperscalers to semiconductor and memory companies, where does it go next?

The answer may not lie in finding the “next Nvidia”. It may instead lie in identifying the next constraint capable of slowing the AI infrastructure boom.

Follow the constraint, not the narrative

AI infrastructure is not a single industry. It is a chain of interdependent systems.

Chips need memory. Memory requires fabrication capacity. Fabrication plants need water, gases, chemicals and electricity. Data centres require power, cooling, fibre, permits, engineering expertise and regulatory approval.

As one constraint is resolved, another emerges.

The economic benefits should therefore migrate towards whichever resource is in shortest supply. Companies controlling that resource may be able to raise prices, secure longer contracts and generate higher returns on capital.

This leads to a different approach to investing in AI.

Instead of searching for companies with “AI” prominently featured in their investor presentations, investors could look for businesses that satisfy four tests:

• Their product or service has become scarce.
• Customers have few practical alternatives.
• Revenue is recurring or contractually protected.
• AI is not yet the principal reason investors own the shares.

The stranger the business sounds, the greater the possibility that the connection has been overlooked.

Selling time to the hyperscalers

Power has already been identified as one of the principal constraints on AI expansion.

The less obvious opportunity, however, is not simply generating more electricity. It is providing electricity before the grid is ready.

A new data centre may have land, buildings, customers and processors in place but remain unable to operate because its grid connection will not arrive for several years.

In those circumstances, the most valuable commodity is not electricity. It is time.

This is driving demand for bridge power: modular gas generation, microgrids and behind-the-meter energy systems that allow facilities to operate before permanent infrastructure has been completed.

Potential beneficiaries include companies traditionally associated with gas compression, generators, engines and industrial maintenance. Their role is evolving from supporting conventional energy infrastructure to accelerating the commercial deployment of AI capacity.

The customer is effectively paying to begin generating revenue sooner.

That can create an attractive business model combining equipment sales, long-term maintenance, fuel infrastructure and recurring service income. It is also less dependent on which semiconductor architecture or hyperscaler ultimately succeeds.

One of the strangest AI beneficiaries may therefore be a company that compresses natural gas.

AI’s growing thirst

AI does not only run on chips and electricity. It also runs on water.

Data centres require increasingly sophisticated cooling systems. Semiconductor factories consume enormous quantities of ultrapure water. Both generate wastewater, while chip production also produces complex chemical and hazardous waste streams.

This creates an opportunity extending far beyond conventional water utilities.

Potential beneficiaries include companies providing:

• Water recycling and closed-loop cooling
• Filtration and membrane technology
• Pumps and leak-detection systems
• Ultrapure water infrastructure
• Wastewater treatment
• Hazardous-waste collection and disposal

These services may be especially attractive because demand is not limited to the construction phase.

Once a facility is operating, water must continue to be treated, monitored, recycled and disposed of. The original infrastructure project creates an installed base capable of generating recurring revenue for many years.

Water may also become an increasingly important planning constraint. A proposed facility that places unacceptable pressure on local resources may face delays, political opposition or additional regulatory scrutiny.

Technology that reduces water consumption could therefore improve the likelihood of gaining approval as well as lowering operating costs.

The eventual AI winner may not manufacture processors. It may manufacture membranes.

The gases behind the chips

Industrial gases provide another largely invisible link in the AI supply chain.

Advanced semiconductor fabrication plants require continuous supplies of ultra-high-purity nitrogen, oxygen, argon, hydrogen and specialist gases. Even a brief interruption can disrupt production and cause substantial financial losses.

Suppliers will often construct dedicated facilities beside semiconductor plants and operate them under long-term agreements.

This creates a very different investment profile from owning a chipmaker.

Semiconductor companies remain exposed to selling prices, technology cycles and competitive leadership. Industrial-gas suppliers are primarily exposed to the quantity of manufacturing capacity installed and the volume operating through it.

The upside may be less dramatic, but the revenue can be more durable and contractually supported.

This also highlights an important distinction. A company can be overlooked as an AI beneficiary without being undervalued as a share.

High-quality industrial businesses are often already priced accordingly. Investors must therefore distinguish between an underappreciated earnings driver and an attractive starting valuation.

Certifying infrastructure nobody has built before

Testing and certification may be the most unconventional opportunity of all.

AI data centres are introducing immersion cooling, direct-to-chip liquid cooling, large battery systems, high-voltage electrical architecture and modular power generation.

All of this must be tested for:

• Fire risk
• Electrical safety
• Reliability
• Environmental compliance
• Regulatory approval

The faster the technology changes, the more equipment requires independent assessment.

Testing companies do not need to predict whether immersion cooling, cold plates or another system ultimately dominates. They can potentially earn revenue from each competing technology as it seeks certification.

This can create an attractive toll-booth business model, combining regulatory barriers to entry, relatively low capital requirements and recurring demand as safety standards continue to evolve.

The next AI winner may be a testing laboratory.

Engineering the impossible

The latest AI campuses are also becoming extraordinarily complex.

They combine computing, power generation, batteries, cooling, water, networking and storage. These systems must operate together, sometimes at a scale that has never previously been attempted.

This creates demand for:

• Engineering consultants
• Electrical contractors
• Grid-connection specialists
• Environmental advisers
• Commissioning engineers
• Digital-twin software

These companies are not selling artificial intelligence. They are selling certainty.

Their value lies in reducing construction delays, avoiding costly mistakes and ensuring that billions of dollars of infrastructure can become commercially operational.

Once again, they are selling time.

What this means for Qualis

This thinking is closely connected to the recent evolution of MGTS Qualis Growth.

We are not arguing that the AI investment cycle is ending. We are questioning whether the companies that financed and initially supplied it will remain its only beneficiaries.

Our response has been to reduce concentrated Nasdaq exposure and broaden the US allocation through the S&P 500. We have also introduced active US smaller-company exposure and increased the portfolio’s allocation to global quality-dividend businesses.

These changes widen the opportunity set beyond the largest technology companies and create greater potential exposure to industrial, infrastructure, energy, engineering and specialist-service businesses.

The objective is not to identify one secret AI stock.

It is to recognise that the profit pool is moving.

The first phase rewarded the companies designing the chips. The second rewarded memory manufacturers and semiconductor equipment suppliers.

The next could reward companies moving gas, purifying water, supplying industrial gases, treating waste and certifying cooling fluids.

The next great AI winner may have almost nothing to do with AI at all.

 

This article is for information only and does not constitute investment advice or a personal recommendation. Capital is at risk, and the value of investments can fall as well as rise.