Concentration of AI stocks in S&P 500 reaches dot-com bubble peak, Bitcoin miners busted

The top 10 AI stocks currently account for about 41% of the S&P 500, according to a chart from BofA Global Research circulated online.

This will bring the AI ​​basket to the same concentration level that technology and communications reached around the time of the dot-com peak. According to BofA charts, Nifty Fifty accounted for 40% in the 1970s, and Japan accounted for 44% in the late 1980s.

The comparison turned stock market concentration warnings into a stress test for a corner of cryptocurrencies that has been pitching a new identity to investors over the past year.

Market concentration triggers stress. Miner disclosure and mining reports provide exposure maps.

Public Bitcoin miners increasingly trade as hybrid infrastructure companies with BTC exposure. Many companies have entered into AI and high-performance computing deals, raised capital for higher-density data centers, converted premium power sites, or shifted investor attention to the economics of long-term leases.

As the premium for AI infrastructure fades, these companies will face a different kind of pressure. Risks range from hash prices alone to debt, contract durability, construction execution, and equity multiples.

At the same time, Bitcoin undergoes a second test. A weaker AI build could ease the competition for power, rack space, interconnects, cooling, and GPUs.

While this will be a blow to miners whose new valuations depend on AI growth, it could help the remaining miners by making it easier to secure scarce infrastructure.

Miners are re-evaluating their own prices around AI

Miner pivots can now be measured in revenue forecasts. The projected revenue mix cited by S&P Global Market Intelligence shows publicly traded miners such as IREN, Riot Platforms, Core Scientific, HIVE, Cipher, and TeraWulf moving toward AI and HPC workloads.

The projected earnings mix is ​​already large enough to change the way these companies are valued.

Visible Alpha expected HPC to account for 71% of 2026 revenue at IREN and Core Scientific, 70% at TeraWulf, 34% at Cipher, 15% at HIVE, and 13% at Riot.

This spread shows that the sector is divided into several groups. Some miners are becoming exposed to Bitcoin and becoming data center operators.

Some companies keep mining as their core business while maintaining the option of AI on sites with access to electricity and the grid.

Its scale is reflected in the economics of the minor. According to CoinShares, public miners have announced AI/HPC contracts totaling more than $70 billion.

The company also stated that WULF, Core Scientific, Cipher, and Hut 8 are effectively becoming the data center operators that still mine Bitcoin.

This will change the market connection from the fall in AI stocks. Any decline in AI multiples will flow into miner stocks as investors assign value to HPC pipelines.

Lower AI demand will also put pressure on financing for projects built around long-term tenants, high-density cooling, and premium grid positions.

Mining margin still depends on BTC price and difficulty, but for stocks there are other variables.

Leverage data points in the same direction. According to CoinShares, several miners have taken on large amounts of debt to build AI, including $3.7 billion in convertible debt from IREN, $5.7 billion in total debt from WULF, and $1.7 billion in senior secured notes issued by Cipher.

crypto slate independently tracked how miners were borrowing and funding AI Pivot while selling BTC. Simply put, the AI ​​pivot added a credit cycle to a business that was already living with the Bitcoin cycle.

The table below combines 2026 revenue projections, 2025 company disclosures, and contract renewals, so each row represents exposure over various time periods.

Cipher’s rebrand for HPC adds another example of change. TeraWulf’s Fluidstack expansion demonstrates how miners have been combining large power portfolios with AI tenant and credit support.

Risk lies in the site, contract and capital stack

Core Scientific is the clearest example of the shift from mining emphasis to running infrastructure. The company said in its fourth-quarter 2025 financial results that it has delivered approximately 350 MW of power under the CoreWeave contract and plans to deliver approximately 590 MW by early 2027.

It also reported that colocation revenue increased to $31.3 million in the fourth quarter from $8.5 million in the same period last year, while digital asset self-mining revenue decreased to $42.2 million from $79.9 million.

That is the core of our management format. Power and buildings, which were once primarily tied to Bitcoin production, are being monetized through colocation.

Core Scientific also said $226.2 million of its $279.2 million in fourth quarter capital expenditures was funded by CoreWeave under existing agreements. While customer funding alleviates some of the capital burden, it also shows how deeply dependent the build is on the AI ​​tenant’s growth path.

This conversion also complicates accounting. Core Scientific said it was restating its historical financial statements after identifying inappropriate capitalization of assets scheduled for demolition during the facility’s conversion from a mine to HPC colocation infrastructure.

Although this issue was company-specific, it illustrates a broader point. The transition from mining holes to dense AI infrastructure is more than a marketing buzzword.

Core Scientific’s cancellation of the CoreWeave merger agreement shows that the value associated with AI is already embedded in shareholder decision-making.

CoreWeave’s 2025 Form 10-K adds counterparty context, including disclosed risks related to large power contracts and AI demand.

Therefore, a miner’s exposure is related to both the delivery of the site and the financial health of the AI ​​cloud ecosystem.

TeraWulf shows similar changes at larger downscaling scales. In its full-year 2025 results, the company reported long-term data center lease agreements totaling 522 significant IT MW, long-term credit enhancement customer commitments totaling over $12.8 billion, and long-term financing of $6.5 billion.

The company said that while it continues to operate its traditional mining infrastructure opportunistically, HPC hosting has become its primary growth engine.

CoinShares reported that WULF mined 262 BTC in the fourth quarter along with $9.7 million in HPC lease revenue. The report said WULF’s cost per BTC figure was distorted by the company’s transition, including interest, SG&A expenses, and depreciation and amortization related to new infrastructure.

This distinction is very important. As miners become AI infrastructure companies, cost per BTC metrics can distort the business unless the balance sheet is separated from the rest of the mining fleet.

Riot’s Corsicana decision illustrates how AI optionality could change Bitcoin’s capacity path before there is a final AI contract. The company’s Corsicana update said it will evaluate the use of AI/HPC on approximately 600MW of its remaining power capacity, halt its previously announced 600MW Phase II Bitcoin mining expansion, and reduce its expected self-mining capacity by the end of 2025 from 46.7EH/s to 38.4EH/s.

IREN adds another exposure type. The October 2025 AI Cloud Update targets more than $500 million in annual AI cloud revenue from 23,000 GPUs by the end of Q1 2026, with 11,000 GPUs contracted at an average 2-year ARR of approximately $225 million.

This creates a faster repricing channel than long-term co-location. GPU cloud economics can change as hardware supply, utilization, and customer budgets change.

Power shortages determine aspects of Bitcoin transactions

The Bitcoin side of the transaction is less straightforward. A weakening AI infrastructure cycle will initially put pressure on miners exposed to AI through equity valuations, funding costs, and contract expectations.

The Bitcoin network will feel the change through the industrial base competing for the same power and sites.

The link between AI and mining is physical. While Bitcoin mining remains a large total revenue pool in the dominant BTC price scenario, AI poses an immediate economic risk to the network’s industrial security foundations.

AI and mining compete for land, grid interconnections, substations, cooling designs, funding, and management attention.

Energy demand from AI explains why competition persists. The IEA estimates data center electricity consumption in 2024 to be approximately 415 TWh, and in the reference case predicts that global data center consumption will approximately double to 945 TWh by 2030.

AI-powered accelerated servers are a big part of the increase. Data centers can be built faster than power systems can add transmission, substations, and generation, making space and access to the grid at a premium.

The North American Data Center Trends report adds market bottlenecks behind that discussion. Low free capacity and high pre-release rates increase the value of power-ready capacity.

For miners, the scarce asset is often a powered site, and the ASIC fleet is just one part of the stack.

According to Bitcoin market data at the time of writing, BTC is trading around $76,800, with a market capitalization of approximately $1.5 trillion, a current block reward of 3.125 BTC, and a network hash rate of over 1.1 ZH/s.

igcurrencynews’s general market page shows Bitcoin’s dominance at about 60% of the $2.6 trillion cryptocurrency market. These numbers put pressure on miner economics before competition from AI is taken into account.

BTC price, fees, difficulty, and energy costs determine how much security Bitcoin can support.

A cooler AI cycle could relieve some of that pressure. If hyperscaler demand, GPU shortages, or data center pre-lease weakens, miners that stay close to Bitcoin may have fewer power sites and infrastructure to contend with.

If the capacity finishes mining, the difficulty will be adjusted and the hash price may increase for the remaining operators. That mechanism is evident in igcurrencynews’s analysis of miners as AI utilities.

There are limits to that relief. Considering the fee and cost situation, with fees close to zero and cost pressure reaching nearly $80,000 per BTC, there is plenty of upside room.

Merely alleviating hardship will leave the weakness of the minor economy unresolved. Long-term AI leases, customer-funded build-outs, interconnection agreements, equipment specialization, and site conversion costs also cause delays.

AI unwinding may unevenly free up capacity, leaving some of it unavailable for mining at attractive returns.

Two outcomes that depend on the demand for AI

The market risk indicated by the AI ​​concentration chart has two different outcomes for miners.

First, the demand for AI applies. Public miners with high-quality power campuses continue to sign HPC contracts because AI tenants can provide longer-term revenue visibility than Bitcoin mining.

As premium sites continue to migrate to AI, mining is concentrated around flexible power, demand response, stranded energy, and regions where interruptions are acceptable.

In that scenario, public miner stock would be less reliable for BTC because the company’s value would come as much from leasing and running data centers as from mined Bitcoin.

The second is the price of AI infrastructure. Miners most exposed to AI growth face pressures from leverage, equity multiples, contract preconditions, construction pipelines, and more.

If the expected revenue from AI declines, it will become more difficult to shoulder the debt raised to expand data centers. GPU cloud contracts with shorter durations can be reset more quickly.

Long-term colocation leases may offer greater protection, but they also lock the site into a path that can take years to recover.

Bitcoin’s potential gains are downstream from its damage. The benefits include alleviation of scarce input resources, reduced power competition, and a better hash price environment for operators still focused on mining.

This is an industrial security discussion, and BTC price is outside the scope of the direct argument.

This is why AI concentration charts are included in the Bitcoin and miner balance sheet discussion. Looking at this chart, there is a growing possibility that AI trading is becoming crowded.

Miner data shows which crypto companies have built their trades around them. The unresolved test is whether these AI/HPC contracts remain durable enough to justify the transition, or whether the same infrastructure that drove public miners away from Bitcoin will become a source of stress.

In the case of Bitcoin, the result is a mix rather than a clean one. Repricing could weaken some of the best-capitalized public miners while making energy and data center inputs less scarce for the remaining miners.

The next signals will come not from AI rhetoric, but from loan terms, tenant delivery schedules, new power contracts, and hashed prices. These are variables that indicate whether miners have purchased a stronger business model or introduced a second cycle to Bitcoin’s security infrastructure.