Why AI Expansion Is Driving Energy Infrastructure Investment

The AI boom is starting to look less like a software cycle and more like an infrastructure buildout. Chips still get the attention, but the harder bottleneck is becoming electricity: powered land, substations, grid connections, cooling, batteries, gas turbines, and data-center campuses that can be delivered fast enough to meet demand.

That is why the chart comparing AI data-center capex with the Marshall Plan, Apollo, Manhattan Project, railroads, and highways matters. The exact six-year estimate should be treated as directional, but the underlying trend is not soft. Epoch AI’s hyperscaler capex tracker says Alphabet, Amazon, Meta, Microsoft, and Oracle have been growing combined capital expenditures at a 72% annualized pace since Q2 2023, nearing half a trillion dollars in 2025, with a trend-line path toward $770 billion in 2026.

The investment case is not simply “AI needs more power.” That is too shallow. The real case is that AI demand is forcing a physical constraint into a digital industry. Every new model, inference product, search assistant, coding agent, autonomous system, and enterprise AI workflow increases the value of reliable electricity and power-ready infrastructure.

For readers following TECHi’s broader AI & Intelligence coverage, this is the next layer of the story: not the model, but the megawatts behind the model. For investors, it also belongs in Markets & Equities because the winners are not only chip companies. Some are energy-storage suppliers. Some are grid and data-center developers. Some are bitcoin miners trying to turn cheap power portfolios into AI-capable campuses.

The AI market is huge. The energy market is still small by comparison

A useful way to read the setup is through market weights. SIFMA’s April 2026 S&P 500 sector table put Information Technology at 35.0% of the index and Communication Services at 11.0%, a combined 46.0% proxy for the public-market weight of AI-adjacent platforms, cloud, semiconductors, software, and digital advertising. Energy was only 3.5%.

That does not mean every technology or communications company is an AI pure play. It does mean the market has already assigned enormous value to the demand side of AI while the physical power side remains a much smaller public-market bucket. That gap is why the energy-infrastructure thesis keeps gaining attention.

The electricity data supports the point. The IEA’s 2026 energy-and-AI update projects data-center electricity consumption roughly doubling from 485 TWh in 2025 to 950 TWh in 2030, with AI-focused data-center electricity demand growing faster than traditional workloads. In the U.S., the Lawrence Berkeley National Laboratory report for the Department of Energy estimated data centers consumed 176 TWh in 2023, or 4.4% of U.S. electricity use, and projected 325 TWh to 580 TWh by 2028, equal to 6.7% to 12.0% of total U.S. electricity consumption.

Those numbers explain why the AI narrative is not cooling off quickly. The spending is not only for experiments. It is flowing into land, electrical gear, power contracts, batteries, construction, and long-lived facilities. Once that cycle starts, it can become self-reinforcing: more AI products create more compute demand, more compute demand creates more power demand, and more power investment makes the next round of AI capacity possible.

TECHi’s earlier analysis asked whether AI data centers are becoming a power stock story. This article narrows that frame to five names you flagged: CleanSpark, MARA, Riot Platforms, Eos Energy, and Tesla.

CleanSpark: bitcoin mining first, AI infrastructure optionality second

CleanSpark (CLSK) is still primarily a bitcoin miner. Its basic business model is to control low-cost power and self-operated data centers, deploy mining machines, produce bitcoin, and manage the treasury and financing around that operating base. That is why the company talks so much about hashrate, fleet efficiency, uptime, and power.

The AI angle is that bitcoin mining and AI data centers both start with scarce electrical infrastructure. CleanSpark’s fiscal 2025 results said the company surpassed 50 EH/s of operational hashrate and generated more than $766 million of revenue. Scale matters because a miner with self-operated data centers, power procurement experience, and a large energy footprint has assets that can become more valuable when hyperscalers are looking for powered sites.

CleanSpark has also started pushing directly into the AI/HPC opportunity. In February 2026, the company announced a major power acquisition near Houston intended to support scaled AI and high-performance-computing development.

The clean version of the thesis is this: CLSK is not a pure AI data-center operator yet. It is a bitcoin miner with power assets that may become more valuable if they can be repurposed, partnered, or expanded for non-bitcoin compute. The risk is equally clear. Bitcoin price, network difficulty, mining economics, construction timelines, and execution in a new AI/HPC market still matter.

MARA: a bitcoin miner turning power into AI-capable infrastructure

MARA is further along in explicitly presenting itself as an energy-and-compute infrastructure company. The legacy business is bitcoin mining: use power-intensive compute to monetize energy and earn bitcoin. The emerging business is to use that same power footprint for AI and high-performance computing.

The February 2026 Starwood partnership is the clearest evidence. MARA and Starwood Digital Ventures said they would develop, finance, and operate AI-capable digital infrastructure across MARA’s power-rich portfolio, targeting about 1 GW of near-term IT capacity with a pathway to more than 2.5 GW.

Then MARA doubled down on the power-control strategy. Its April 2026 agreement to acquire Long Ridge Energy & Power would add a 505 MW combined-cycle gas plant, more than 1,600 acres, and a campus the company says can support more than 1 GW of total potential capacity.

That makes MARA one of the more direct plays on the AI power bottleneck among the bitcoin miners. The company is not just saying AI might create optionality. It is trying to convert owned or controlled energy infrastructure into leaseable digital infrastructure. The upside is a more stable, tenant-backed revenue stream than pure bitcoin mining. The risk is that this requires massive capital, permitting, project execution, and real customer demand at the right price.

Riot Platforms: the AMD lease is the proof point investors needed

Riot Platforms (RIOT) has a similar starting point: bitcoin mining built around large-scale power and data-center assets. But Riot now has a tangible bridge from mining into AI/HPC infrastructure.

In its 2025 results release, Riot described itself as a Bitcoin-driven leader in large-scale data centers and mining applications. It reported $647.4 million of annual revenue, 5,686 bitcoin mined, and said it had begun generating revenue in January 2026 from the first phase of a data-center lease with AMD. Riot also pointed to a nearly two-gigawatt power portfolio across Corsicana, Rockdale, and other sites in its full-year 2025 strategic update.

That AMD lease matters because it moves the story from “bitcoin miners may someday pivot to AI” to “a major chip company is already leasing capacity from a former pure mining infrastructure base.” Riot’s model benefits from the AI supercycle if powered land and electrical infrastructure become the scarce resource hyperscalers and chip companies need.

Riot still carries bitcoin-cycle risk. Mining revenue, bitcoin holdings, depreciation, capex, and power costs can dominate near-term results. But if the AMD lease expands and Riot signs more non-mining tenants, the market may start valuing part of the company less like a miner and more like a power-backed data-center developer.

Eos Energy: batteries for the grid stress AI creates

Eos Energy (EOSE) is not a data-center operator or bitcoin miner. It designs and manufactures zinc-based long-duration battery energy storage systems for utility, industrial, and commercial customers. Its role in the AI cycle is different: data centers need reliable power, and grids with large AI loads need storage to smooth volatility, shift energy, and support resiliency.

That fits the IEA warning that AI workloads can create rapid power swings and make energy storage more important for reliability. Eos is trying to sell into that exact need. In its 2025 results, the company reported $114.2 million of revenue, more than 7x 2024, a $701.5 million backlog representing 2.8 GWh, and a $23.6 billion commercial opportunity pipeline.

The other part of the thesis is U.S. manufacturing. The Department of Energy’s Loan Programs Office closed an up to $305.3 million loan guarantee for Eos to support manufacturing lines for utility- and industrial-scale zinc-bromine battery systems in Pennsylvania.

EOSE is a high-risk infrastructure equity because it is still scaling manufacturing and converting backlog into profitable revenue. But the AI supercycle helps the demand backdrop. If data centers, utilities, and industrial customers need safer, longer-duration storage options near load, Eos has a clearer reason to exist than it did when storage was treated as a generic renewables add-on.

Tesla: the AI story is not only cars and robots

Tesla (TSLA) is usually valued through electric vehicles, autonomy, robotics, and Elon Musk’s broader AI narrative. But the cleaner AI infrastructure angle is Tesla Energy.

Tesla’s energy generation and storage segment sells Megapack, Powerwall, solar, and related services. In its 2025 Form 10-K, Tesla said energy generation and storage revenue increased by $2.69 billion, or 27%, from 2024, primarily because of higher Megapack and Powerwall deployments, while segment gross profit reached $3.8 billion in 2025. Tesla also reported 8.8 GWh of storage deployments in Q1 2026 in its production and deployment update.

For this theme, the important product is Megapack. AI data centers need firm power and grid support, and utility-scale batteries can help absorb volatility, manage peak loads, and make power projects more financeable. Tesla’s TSLA quote page will still move mostly on the auto, autonomy, and robotics narrative, but the energy-storage segment gives the company a real infrastructure exposure that is not dependent on selling another car.

The risk is valuation and mix. Tesla is not a pure grid-storage company. A strong Megapack cycle can be buried by weaker vehicle margins, autonomy execution risk, regulatory noise, or market skepticism about the broader Tesla story. Energy is becoming more important, but it is still one piece of a much larger company.

How the five stocks fit together

The five names are not the same trade.

CLSK, MARA, and RIOT are power-and-compute infrastructure call options wrapped in bitcoin-mining economics. Their advantage is that they already operate energy-intensive sites and understand power procurement. Their weakness is that bitcoin volatility can overwhelm the AI optionality if mining economics turn down.

EOSE is a grid-storage execution story. It benefits if AI load growth makes long-duration storage more valuable, but it has to prove manufacturing scale, margins, project delivery, and financing discipline.

TSLA is a scaled storage and AI platform story. It benefits if Megapack demand keeps expanding alongside data-center and utility investment, but investors have to separate the energy-storage thesis from the much larger auto and autonomy debate.

That is why the strongest version of this idea is not “buy every power stock.” The stronger version is that AI has created a new scarcity map. The companies closest to scarce power, scarce grid capacity, scarce storage, or scarce AI-ready land may capture value before the broader market fully separates them from the old labels of miner, utility supplier, battery maker, or automaker.

The feedback loop investors are watching

The loop is simple, but it is powerful:

1. More AI usage increases demand for compute.
2. More compute requires more data-center capacity.
3. More data-center capacity requires more electricity, storage, cooling, and grid infrastructure.
4. More infrastructure spending makes more AI capacity possible.
5. More capacity lowers bottlenecks and supports the next wave of AI products.

That is the reason the AI narrative can stay hot even when investors worry about model competition or short-term spending discipline. The money is going into physical systems. Those projects do not disappear overnight.

The bear case is also real. If AI revenue fails to justify hyperscaler capex, if electricity prices trigger political backlash, if grid interconnections take longer than expected, or if capital markets pull back, the infrastructure cycle can slow. Bitcoin miners have an extra layer of risk because BTC price and mining difficulty still matter. Battery companies have execution risk. Tesla has valuation risk.

But the direction of demand is hard to ignore. The AI economy is turning intelligence into an energy load. The companies that control the infrastructure behind that load are no longer side stories. They are becoming part of the main AI trade.