The artificial intelligence race is no longer being fought exclusively inside computer laboratories, semiconductor factories or software companies. An increasingly important battle is taking place somewhere less glamorous: the energy industry.
Anthropic, the company behind Claude, has reportedly agreed to spend approximately $3.4 billion securing 1.7 gigawatts of natural gas-fired power generation equipment for its expanding data centre operations. The deal, first reported by the Financial Times on November 8, represents one of the clearest signs yet that leading AI companies are moving beyond purchasing computing capacity and taking greater control of the infrastructure that keeps it running.
But the most interesting part of this agreement isn’t simply its price.
Anthropic is effectively securing a critical piece of the energy supply chain before electricity shortages, equipment backlogs and competition from other technology companies make expansion even more difficult.
The move also raises a bigger question: If the future of artificial intelligence depends on enormous quantities of electricity, will the companies building AI eventually become some of the world’s biggest private power operators?
Anthropic’s latest agreement suggests that future is arriving faster than many people expected.
Table of Contents
1. What exactly did Anthropic buy for $3.4 billion?
The Financial Times report contains an important detail that many initial headlines have overlooked.
Anthropic did not simply sign a conventional electricity-purchasing agreement with a utility company. It reportedly secured commitments for 1.7 gigawatts of heavy-duty Caterpillar gas-fired generators originally ordered by a Texas-based supplier, OnePWR Solutions.
The transaction was agreed in late September 2026.
According to the report:
- OnePWR Solutions had previously ordered the generators from Caterpillar.
- Anthropic initially planned to purchase electricity from OnePWR.
- It subsequently decided to take over the existing generator orders.
- The reported $3.4 billion arrangement also includes engineering and construction capabilities needed to install the equipment.
- Anthropic did not acquire OnePWR itself, and the supplier’s employees did not transfer to Anthropic.
This distinction matters. Anthropic is securing equipment and the capabilities required to deploy it, rather than purchasing an established energy company.
Reuters separately reported the transaction but said it could not independently verify the Financial Times account. Neither Anthropic nor OnePWR publicly confirmed the arrangement in the reporting. Financial Times+1
How much electricity is 1.7 gigawatts?
The number sounds abstract until it is put into perspective.
A 1.7 GW generation portfolio is a substantial amount of power infrastructure. It is comparable in nameplate capacity to a large power-generation complex.
However, there is an important distinction between installed generating capacity and electricity actually delivered. The generators must be installed, connected to fuel supplies, permitted and operated. Their real output will depend on operating conditions, maintenance, fuel availability and other constraints.
The agreement does not mean Anthropic already has 1.7 GW of operational electricity.
It means the company is attempting to secure the equipment needed to build that capacity before competitors take the available supply.
2. Why Anthropic is bypassing traditional electricity suppliers
For years, the conventional model of building a data centre was relatively straightforward.
A technology company would find suitable land, arrange construction, obtain a grid connection and purchase electricity from a utility.
That model is becoming increasingly difficult for large AI facilities.
Modern AI data centres require enormous amounts of electricity to operate thousands of specialised processors. Training advanced models and serving millions of user requests create a continuous demand for computing power.
The problem is that electricity infrastructure cannot expand at the same speed as AI software.
Building a new power plant, extending a natural gas pipeline, constructing transmission infrastructure and obtaining regulatory approvals can take years.
Anthropic directly acknowledges this problem in its November 2 SEC registration filing.
The company warns that the US electricity grid was not designed to accommodate the rapid increase in demand associated with AI development. It also identifies long construction and interconnection timelines as major obstacles to expanding its computing infrastructure.
This creates an uncomfortable situation for AI companies.
They can potentially purchase more advanced chips, hire additional researchers and develop more powerful models. But none of those investments can deliver their intended value if the data centres housing the hardware cannot obtain enough electricity.
The bottleneck is shifting from computing hardware to the physical infrastructure that supplies it with power.
Anthropic’s decision is a response to that bottleneck.
Instead of waiting entirely for traditional energy suppliers to expand their capacity, the company is moving closer to the source of electricity itself.
3. The overlooked detail: Anthropic is not just buying generators. It is buying time.
One of the most important aspects of this transaction is its timing.
The generators involved in the agreement had already been ordered by OnePWR from Caterpillar. Taking over existing equipment commitments potentially gives Anthropic access to manufacturing capacity that would otherwise have been difficult to secure.
This matters because the global supply chain for large power-generation equipment is under pressure.
Data centre developers are competing for generators, turbines, transformers, switchgear and other specialised electrical components.
Wärtsilä, a major manufacturer of power-generation engines, reported in its October 2026 update that strong demand was pushing equipment delivery times higher. The company also identified data centres as an important source of demand. wartsila.com
The wider grid problem is equally revealing.
According to Lawrence Berkeley National Laboratory’s 2026 interconnection research, projects seeking connections to the US electricity grid face lengthy approval processes. For projects built in 2025 in regions with available data, the median time between submitting an interconnection request and commercial operation exceeded five years. Energy Markets & Planning+1
That is a major obstacle for a company trying to expand its AI infrastructure quickly.
Anthropic is effectively paying to reduce one part of that waiting period.
But this is not a guaranteed shortcut. Buying generators does not automatically secure natural gas, pipeline connections, land, environmental permits or completed construction.
It solves an equipment-access problem, not the entire electricity problem.
4. Why natural gas instead of waiting for renewable energy or nuclear power?
This is where the deal becomes particularly interesting.
Natural gas is not necessarily the cleanest or cheapest long-term solution for powering AI infrastructure. However, it offers something that is extremely valuable to data centre operators: dispatchable electricity.
Unlike solar and wind generation, which depend on weather conditions, natural gas generators can generally be operated when electricity is needed, subject to fuel availability and equipment constraints.
For data centres that require dependable power around the clock, this flexibility is attractive.
There are three important considerations.
First, reliability and control.
A data centre cannot simply stop processing workloads whenever solar generation falls or wind conditions change. Batteries and grid connections can help manage fluctuations, but they introduce additional infrastructure requirements.
Natural gas generation offers a controllable source of electricity that can operate independently of the immediate availability of renewable generation.
Second, deployment speed.
New nuclear power projects involve lengthy development, regulatory and construction processes. Large renewable projects also need suitable locations, transmission connections and supporting storage or balancing arrangements.
Natural gas engines can offer a more immediate alternative in certain locations, particularly where suitable fuel infrastructure already exists.
Third, energy security.
Securing dedicated generation can reduce dependence on the availability of electricity from an overloaded local grid.
However, it does not eliminate dependence on external infrastructure. Natural gas still has to reach the generators through pipelines or other fuel-delivery arrangements.
This is a critical distinction that the headline alone does not explain.
Anthropic is not escaping the energy supply chain. It is changing which parts of that supply chain it wants to control.
5. The environmental contradiction behind the AI boom
Anthropic’s deal also creates an uncomfortable question for the AI industry.
Can companies build increasingly powerful artificial intelligence systems while relying on a growing fleet of fossil-fuel-powered generators?
Natural gas is often described as a cleaner-burning alternative to coal. However, that does not make it emissions-free.
Burning natural gas releases carbon dioxide. Gas production and transportation can also result in methane leakage, an important contributor to climate change.
There are additional local environmental concerns.
Large gas-fired generator installations can produce nitrogen oxides and other air pollutants. Their environmental impact depends on the engine technology, emissions controls, operating hours and local conditions.
The US Environmental Protection Agency specifically identifies stationary combustion turbines and engines as power sources subject to Clean Air Act emissions requirements relevant to data centres. US EPA
There is another issue that deserves attention: local communities.
A large generation installation can bring employment, construction investment and supporting infrastructure. But nearby residents may also face concerns about air quality, noise, land use and the expansion of gas pipelines.
These questions become especially important when a data centre and its electricity generation are located close to residential areas.
The 1.7 GW figure should therefore not be viewed only as a technology investment. It represents a potentially significant expansion of physical energy infrastructure.
Is natural gas a temporary bridge or a long-term commitment?
That question is particularly important because data centre infrastructure is built to operate over many years.
If natural gas becomes the fastest way to satisfy immediate AI electricity demand, companies may find themselves locked into fossil-fuel infrastructure even as renewable generation, battery storage and other technologies become more competitive.
Anthropic’s own SEC filing acknowledges that certain dedicated power and data centre arrangements can have initial terms of 20 years, with annual price increases and fuel costs passed through without caps. SEC+1
That makes the environmental question inseparable from the financial one.
A generator purchased to solve today’s electricity shortage could influence a company’s energy costs and emissions profile for years.
6. The $895.4 billion question: How much infrastructure is Anthropic committing to?
Perhaps the most revealing information about Anthropic’s energy strategy is not contained in the $3.4 billion generator agreement.
It is in the company’s SEC registration filing.
In its November 2, 2026 amended S-1 filing, Anthropic disclosed approximately $895.4 billion in aggregate purchase and lease commitments, substantially all related to computing and infrastructure.
The filing breaks the figure down further:
| Commitment period | Disclosed amount |
|---|---|
| Total aggregate commitments | $895.4 billion |
| Expected before December 31, 2030 | $447.5 billion |
| Due in the 12 months following September 30, 2026 | $69.8 billion |
| Most remaining commitments | 2031–2045 |
Source: Anthropic’s SEC registration statement.
These numbers need careful interpretation.
The $895.4 billion is not an amount Anthropic has already spent, nor is it simply a pile of debt. It represents aggregate contractual commitments, including purchases and leases, some of which have not yet commenced.
It also should not automatically be added to the $3.4 billion generator deal. The filing does not separately identify that transaction as an additional amount outside the disclosed commitments.
Nevertheless, the scale is remarkable.
Anthropic is making infrastructure decisions that extend far beyond its immediate computing requirements.
Why this changes the financial risk
AI companies generally generate revenue by selling access to models, APIs and related services.
But the infrastructure required to support those services often needs to be secured years in advance.
This creates a potential mismatch:
- Infrastructure suppliers expect long-term payments.
- Data centres require large upfront investments.
- Electricity and fuel prices can change.
- Customer demand and AI revenue can fluctuate.
- Computing hardware may become outdated before long-term commitments expire.
Anthropic explicitly warns investors that some of its contracts contain minimum purchase commitments and take-or-pay provisions, meaning it may have to pay for resources even when they are not fully utilised.
It also acknowledges that its revenue expectations may not materialise quickly enough to cover its infrastructure obligations. SEC
This is an important part of the story.
Anthropic is not simply investing in infrastructure because it expects AI demand to grow. It is accepting substantial long-term financial obligations on the assumption that the demand will justify them.
If that assumption proves correct, securing infrastructure early could provide a major competitive advantage.
If demand grows more slowly than expected, the same commitments could become a financial burden.
7. Anthropic Direct: The strategy behind owning more of the AI infrastructure
The generator agreement makes more sense when viewed alongside another initiative disclosed in Anthropic’s IPO filing: Anthropic Direct.
Traditionally, Anthropic has relied heavily on cloud providers and infrastructure partners to access computing resources.
Under its newer approach, the company is moving towards a hybrid infrastructure model.
Rather than relying exclusively on cloud providers to supply computing capacity, Anthropic can:
- Secure land and dedicated data centre facilities through partners.
- Arrange access to electricity and supporting infrastructure.
- Lease or procure specialised computing equipment directly.
- Have third-party operators host and manage the infrastructure.
- Deploy its own workloads inside those dedicated facilities.
The important innovation is the separation of two major investments: securing the data centre and deciding how many expensive AI accelerators to install inside it.
Anthropic explains that data centres take longer to prepare than the computing equipment deployed within them. Securing facilities early allows the company to delay some accelerator spending until its computing requirements become clearer.
The company believes this approach could provide long-term economic advantages and greater flexibility.
This is more than a procurement strategy.
It represents a change in how an AI company thinks about its business.
Anthropic increasingly wants control over the physical foundations of its products, even when third-party companies still own or operate important parts of the infrastructure.
Why this matters to the AI industry
The biggest AI companies are discovering that relying on external infrastructure providers has limitations.
Cloud providers can offer enormous computing capacity, but they also serve other customers competing for the same chips, data centre locations and electricity.
Securing dedicated facilities gives an AI company greater control over where and when its computing resources become available.
It can also reduce exposure to some forms of supply uncertainty.
However, this approach requires more capital, more complex contracts and greater operational responsibility.
Anthropic is effectively moving closer to the business of building and operating an industrial-scale computing infrastructure network.
8. Could this affect electricity bills for ordinary households?
This is perhaps the most important question for people living near large data centre developments.
AI companies need enormous quantities of electricity. When a large facility connects to a local grid, utilities may need to invest in additional substations, transmission lines and other infrastructure.
If those costs are allocated across the wider customer base, ordinary households could potentially end up paying part of the bill.
This concern has become increasingly important in the United States, where data centre expansion is creating new challenges for electricity planners and regulators.
Anthropic has attempted to address the issue directly.
The company has publicly committed to:
- Paying for 100% of the grid upgrades needed to interconnect its data centres.
- Bringing new electricity generation online to match its data centre demand.
- Working to cover demand-driven electricity price increases where new generation is not yet available.
- Investing in systems that reduce electricity consumption during periods of peak grid demand.
- Addressing environmental impacts and supporting local communities.
These are meaningful commitments, but they need to be evaluated through actual implementation and independent verification.
There is also an important distinction between paying for grid upgrades and guaranteeing that electricity bills will never rise.
Electricity prices are influenced by many factors, including fuel costs, weather, transmission constraints, generation availability and demand from other industries.
Anthropic’s own SEC filing acknowledges that its fuel costs can be passed through without cost caps under certain arrangements.
The company also warns that growing competition for natural gas and pipeline capacity could push prices higher in some locations.
In other words, Anthropic may be willing to pay for the infrastructure it needs, but that does not make the underlying energy system immune to price increases.
The real test will be whether AI companies can expand their electricity consumption without transferring a disproportionate share of the costs to the communities around them.
9. The hidden vulnerability: A gas generator still needs gas
There is another aspect of this deal that deserves more attention.
Buying 1.7 GW of generators does not guarantee that 1.7 GW of electricity will always be available.
The generators require a dependable supply of natural gas.
That means Anthropic must also consider:
- Natural gas pipeline availability.
- Long-term fuel transportation agreements.
- Regional gas prices.
- Pipeline construction and connection approvals.
- Fuel supply during periods of extreme weather.
- Maintenance and equipment reliability.
Anthropic’s SEC filing explicitly warns that some planned data centre facilities rely on dedicated on-site natural gas generation.
It says these arrangements expose the company to fuel price volatility and supply disruptions. It also identifies pipeline capacity and firm fuel transportation as potential constraints.
More importantly, Anthropic warns that competition for gas supplies and pipeline capacity could increase prices or prevent planned projects from moving forward.
This creates an interesting contradiction.
Anthropic is trying to escape electricity shortages by securing its own generation equipment. But the solution introduces a new dependency on the natural gas industry.
The bottleneck could simply move from electricity connections to fuel transportation.
This is why the $3.4 billion agreement should be understood as one part of a much larger infrastructure strategy rather than a complete solution to Anthropic’s power requirements.
10. What does this mean for Anthropic’s potential IPO?
The timing of the agreement is also noteworthy.
Anthropic is preparing for a public listing, and its SEC registration statement provides investors with a closer look at the financial commitments behind its growth strategy.
The company’s infrastructure expansion presents two competing investment narratives.
The optimistic case: Anthropic is securing scarce resources before its competitors. If demand for Claude and other AI services continues to grow, early access to power and dedicated computing facilities could help the company expand faster and potentially improve its long-term infrastructure economics.
The cautious case: The company is committing enormous sums to infrastructure that may take years to become fully operational. Electricity costs, fuel prices, construction delays, changes in AI hardware and slower-than-expected customer demand could weaken the economics of these investments.
The prospectus itself identifies these risks.
It warns that delays in completing facilities, securing energy and bringing computing capacity online could leave Anthropic with infrastructure that is underutilised or does not generate revenue quickly enough to justify its costs.
This is why investors should look beyond the headline revenue growth of an AI company.
The more important question is whether its future revenue can support the long-term costs of the infrastructure required to generate that revenue.
Anthropic’s $3.4 billion generator agreement makes that question even more relevant.
11. Anthropic is not alone: AI companies are becoming energy buyers
Anthropic’s move is part of a wider transformation across the technology industry.
AI developers are increasingly competing for resources that traditionally belonged to utilities, industrial energy consumers and power-generation companies.
The emerging strategies include:
| Strategy | Why companies are pursuing it |
|---|---|
| On-site natural gas generation | Obtain power without waiting entirely for grid expansion |
| Renewable energy agreements | Secure longer-term electricity supplies and support cleaner generation |
| Battery storage | Manage electricity demand and provide flexibility |
| Dedicated data centres | Gain greater control over computing infrastructure |
| Nuclear energy partnerships | Explore dependable low-carbon electricity for long-term demand |
This is not necessarily a choice between one energy source and another.
A future AI data centre could use a combination of grid electricity, renewable power, battery storage and natural gas generation.
The important development is that technology companies are becoming more directly involved in deciding how electricity is produced, delivered and financed.
That could reshape the relationship between the technology and energy industries.
The next major competitive advantage in AI may not come exclusively from designing a better model or securing a more advanced chip.
It could come from securing the electricity needed to run those chips.
12. What happens next? Five developments worth watching
The $3.4 billion agreement is only one part of Anthropic’s infrastructure expansion. Several developments will determine whether this strategy succeeds.
1. Actual generator installation
The first question is when the generators will become operational. Securing equipment is not the same as delivering electricity. Construction, engineering and commissioning will determine how quickly Anthropic benefits from the transaction.
2. Natural gas availability
Anthropic will need dependable fuel supplies and pipeline connections. Competition from other data centre developers could make these resources more expensive.
3. Environmental approvals
The locations, emissions permits and operating arrangements for the installations will be important. The public still lacks a complete picture of where the reported generator capacity will be deployed.
4. The economics of AI infrastructure
Anthropic must demonstrate that revenue from Claude and its enterprise AI services can justify its enormous long-term infrastructure commitments.
5. Whether competitors follow the same strategy
If securing dedicated power becomes a competitive necessity, other AI companies could increasingly purchase generation equipment, sign direct fuel agreements and invest in dedicated energy infrastructure.
That would represent a significant change in the relationship between technology companies and traditional energy providers.
The bigger picture: AI’s next battle may be fought in the energy industry
Anthropic’s reported $3.4 billion natural gas agreement reveals something fundamental about the future of artificial intelligence.
The industry is discovering that developing advanced AI is not simply a software challenge. It requires enormous physical resources, including land, electricity, cooling systems, specialised hardware and reliable supply chains.
For years, the AI conversation focused on which company had the best model, the most advanced chips or the largest computing clusters.
Now another question is becoming increasingly important:
Who has enough electricity to keep those systems running?
Anthropic appears willing to spend billions securing an answer.
Its move could help it overcome some of the electricity constraints limiting AI expansion. But it also exposes the company to a different set of risks involving fuel prices, environmental regulations, construction delays and enormous long-term financial commitments.
There is a broader lesson here for the entire AI industry.
The companies that dominate the next phase of artificial intelligence may not necessarily be the ones that develop the most impressive algorithms.
They could be the companies that successfully secure the energy and infrastructure required to turn those algorithms into commercially useful products.
The AI race is becoming an energy race. And increasingly, the winners may be determined not just by what they can build, but by whether they can keep it powered.
What readers should remember
Anthropic’s $3.4 billion deal is not simply a purchase of generators. It is a sign that AI companies are moving deeper into the physical infrastructure business.
The real story involves three interconnected challenges:
- Technology: Securing enough electricity to run increasingly powerful AI systems.
- Economics: Making enormous long-term infrastructure commitments financially sustainable.
- Environment: Meeting rising energy demand without creating unacceptable costs for communities and the climate.
How Anthropic manages these three challenges could become a blueprint—or a warning—for the next generation of AI companies.
Also Read: