Bitcoin's total network hash rate drops below the annual average: What kind of game is the miners' AI transformation playing?

In May 2026, the Bitcoin network experienced a noteworthy technical signal. According to CryptoQuant analysts monitoring the network, the total network hash rate has fallen below the annual average level, a situation not seen since 2021. Since the winter storms in the United States, the hash rate has remained below the 1 ZH/s threshold, reflecting significant changes in the competitive landscape among miners.

From a broader timeline perspective, after reaching a peak of approximately 1,160 EH/s at the end of 2025, Bitcoin’s hash rate saw its first quarterly decline in six years during the first quarter of 2026, with a decrease of about 4% to 10%. This rare consecutive downward trend is not due to technical failures in the network itself but is a reflection of collective business decisions by miners. Some mining companies have chosen to temporarily shut down operations, while others are adjusting their business models to improve profitability. Bitfarms’ recent operational adjustments serve as an example.



## Why High Mining Costs Lead to Continuous Losses Compared to Coin Prices

The direct driver of the hash rate decline is the fundamental deterioration of the mining economic model. Data from February 2026 shows that the average total cost to produce one Bitcoin across the network has risen to $87k, while the coin price has fluctuated repeatedly around $60k—meaning miners are losing nearly $20k on each block mined.

This “loss on every unit produced” situation is extremely rare in industry history. Hash price—the metric measuring revenue per unit of hash power—dropped to between $28 and $33 per PH per day in early 2026, hitting the lowest level since the halving. The weighted average cash cost for publicly listed miners approaches $80k per Bitcoin, with about 15% to 20% of the global mining fleet operating at a loss, especially older mid-generation equipment. An internal industry “Miner Profitability Sustainability Index” has fallen from a healthy level of 100 to 21, indicating that, aside from a few top players with the most efficient power and hardware, the vast majority of miners are operating with negative cash flow.

Under such profit squeeze, the network has experienced consecutive negative difficulty adjustments, with a 7.76% reduction in mining difficulty in March 2026— the largest adjustment in over a year. While automatic difficulty adjustments are a system balancing mechanism, they also indirectly confirm that a large number of ASIC miners are exiting the network.

## Why Miners Are Collectively Transitioning to AI Computing Infrastructure

Against the backdrop of ongoing losses in traditional mining operations, the entire mining industry is undergoing an unprecedented shift. In February 2026, Bitdeer liquidated all its Bitcoin holdings, selling not only the 189.8 BTC mined that week but also clearing out the 943.1 BTC stored in its treasury, realizing about $63 million. This move broke the long-held industry rule of “mining equals hoarding.”

IREN Limited completed a $3 billion convertible senior bond issuance, netting $2.96 billion, to accelerate its transition from cryptocurrency mining to AI infrastructure services. Prior to this, the company had signed a $9.7 billion AI cloud hosting agreement with Microsoft and reached an agreement with NVIDIA to deploy up to 5 GW of AI data center capacity globally.

Globally, several publicly listed mining companies have announced AI and HPC contracts totaling over $70 billion. These long-term hosting contracts are usually priced in fixed USD amounts, providing highly predictable and high-margin cash flows, contrasting sharply with the volatile Bitcoin block rewards.

## Can the Financial Performance of AI Computing Business Offset Mining Losses?

The effects of this transition are gradually reflected in financial data. For some leading miners, AI-related revenue now accounts for about 30% of total income, expected to rise to as high as 70% by the end of 2026. AI hosting services typically achieve operating profit margins of 80% to 90%, far exceeding traditional mining profitability levels.

Diversification of revenue streams provides miners with a hedge against the cyclical volatility of cryptocurrencies. Cipher Digital signed its third large-capacity leasing agreement in Q1 2026, leasing out its power and data center resources. Core Scientific has transformed its mining facilities into cloud and AI shared cabinets and data centers, actively expanding hosting and co-location services. Riot Platforms reported $33.2 million in data center revenue during the same period, indicating that “hash power hosting” has moved from a narrative stage to actual revenue generation.

However, this transition also involves significant financial pain. In Q1 2026, several listed miners reported “grim” results: Hut 8 posted a net loss of $253.1 million, Core Scientific’s net loss reached $347.2 million, and Riot Platforms’ quarterly loss exceeded $500 million. Much of these losses stem from asset impairments of mining hardware and related equipment in a low-price environment, as well as capital expenditures for the transition. IREN’s financing case illustrates that such a shift requires not only external capital injections but also solutions to complex issues like debt restructuring and shareholder dilution.

## What Does the Rise of Miner Reserves to $140 Billion Mean?

While hash rate continues to decline, miner reserves have seen a remarkable increase. As of April 2026, reserves reached $140 billion, the highest level since February 2026. This data reveals a seemingly contradictory signal: on one hand, miners are reducing hash rate investments and operational scale; on the other, their holdings of Bitcoin are increasing.

This situation can be understood from two perspectives. From the supply side, miners who have shut down operations no longer need to sell newly mined Bitcoin frequently to cover costs, reducing market liquidity. From the demand side, some miners have liquidated old equipment and completed financing, gaining sufficient capital for their transition and are not eager to sell their existing Bitcoin holdings. After the $3 billion bond issuance, IREN’s debt structure improved, easing cash flow pressures. This provides miners with more room to hold their coins.

It’s important to note that this increase in reserves does not necessarily indicate bullish market sentiment. The motivation for miners to hold coins is fundamentally a transitional state during their business model shift. As more miners shift focus from “mining for coins” to “hash power leasing,” the structural changes in reserves need to be observed over a longer time horizon.

## Does the Transition to AI Computing Signal a Long-term Structural Change?

The migration of miners into AI infrastructure is essentially a revaluation of resources in the digital infrastructure sector. Traditional data center construction cycles typically take 3 to 5 years, but miners already possess land, power contracts, and grid connection permits, significantly shortening the transformation cycle. During peak AI demand, the most valuable assets for miners are no longer the mining hardware but “the speed of power connection.”

In capital markets, this transition has fundamentally changed pricing logic. Through “recognition agreements,” tech giants like Google and Microsoft provide financial guarantees for rent payments on transformed data centers, turning high-risk miner leasing contracts into low-risk credit agreements with tech giants. These assets can then access bond markets at approximately 7.125% interest rates, attracting large institutional capital from Vanguard, Oaktree, and others. As a result, the valuation metrics for mining farms have shifted from “hash rate scale” to “available power capacity and interconnection points.”

However, the path is not smooth. Physical infrastructure upgrades are far more challenging than replacing chips: AI data centers require N+1 or even 2N power redundancy, millisecond-level switching, direct-to-chip liquid cooling systems, and strict environmental controls. A single cabinet can consume up to 120 kW, and traditional air cooling cannot support high-density AI clusters. Riot Platforms’ chief data center officer resigned during the transition process, forfeiting potential gains of about $18.7 million, illustrating the complexity of this transformation.

## How Do Declining Hash Rate and Rising Miner Reserves Affect Market Structure?

The apparent contradiction between declining hash rate and rising reserves points to deep structural changes in the Bitcoin market. The loss of hash rate indicates a reduction in active miners, temporarily easing competitive pressure. For remaining miners, difficulty adjustments improve marginal profitability.

From a liquidity perspective, high reserves suggest that short-term selling pressure from miners may decrease. Historically, large miner sell-offs tend to occur near market bottoms; conversely, accumulated reserves could imply a longer-term bullish outlook. However, caution is warranted: not all of the $140 billion reserves are “liquid assets,” as some are locked as collateral for financing or part of transition capital.

Models from institutions like CoinShares project that total network hash rate could rebound to 1.8 ZH/s by the end of 2026 and reach 2 ZH/s by March 2027, heavily dependent on Bitcoin prices returning close to or above $100k. If prices fail to recover effectively, the structural loss of hash rate could accelerate, pushing more small and medium miners out of the market and further altering the industry’s concentration.

## What Risks and Constraints Do Miners Face in Transitioning to AI Computing?

The transition from mining to AI infrastructure is not without limits. First, the capital barrier: building an AI-ready facility now costs approximately $8 million to $11 million per MW, a high threshold that many small and medium miners cannot afford. Second, technical constraints: AI workloads require 99.999% uptime and latency below 10 milliseconds, while traditional mining power architectures typically only provide N-level redundancy, necessitating a complete overhaul of power distribution systems.

Market competition is also intense. As miners shift to AI infrastructure, they will compete with traditional data center operators, cloud providers, and tech giants. In NVIDIA- and AMD-dominated AI chip supply chains, miners lack priority in procurement. Moreover, whether AI revenue growth can sufficiently offset ongoing mining losses remains uncertain.

From a cybersecurity perspective, the reduction in hash rate and network decentralization could lower the network’s resistance to attacks. Although no direct risks have materialized yet, the trend toward concentration among a few large players conflicts with Bitcoin’s core value of decentralization.

## Summary

In Q1 2026, the Bitcoin mining industry is undergoing its most profound structural adjustment since the halving. The hash rate falling below the annual average directly reflects miners’ collective business decisions, driven by the inverted relationship between mining income and costs—losing nearly $20k per Bitcoin mined compels many to seek new profit avenues. The shift toward AI infrastructure is essentially a re-pricing of core assets like power capacity, land, and grid permits. This transition has already shown initial financial results, with some miners’ AI revenue accounting for over 30%. Miner reserves have risen to $140 billion, influenced both by capacity contraction on the supply side and by the willingness to hold coins after financing. However, the high costs, technical barriers, and fierce market competition mean that this process will continue to involve significant pain in the short term. The structural changes in hash rate and the game of reserves will remain key indicators for observing supply-demand dynamics in the Bitcoin market in 2026.

## FAQ

Q1: Does declining hash rate mean the network’s security is decreasing?

Declining hash rate does not directly pose a major risk to network security. The difficulty adjustment mechanism automatically recalibrates mining difficulty to adapt to hash rate changes, maintaining the 10-minute block time target. However, in the long term, if hash rate loss becomes a structural trend, increased centralization could theoretically impact network security.

Q2: How does the rise of miner reserves to $140 billion affect prices?

An increase in reserves suggests that miners’ short-term selling pressure may decrease, reducing liquidity in the market. But this should be considered alongside miners’ cash flow status and financing structures; it is not a straightforward bullish signal.

Q3: Will miners transitioning to AI still hold Bitcoin?

Strategies vary among miners. Some choose to liquidate their Bitcoin holdings to fund their transition; others retain reserves and raise capital through financing. Changes in miners’ holding intentions require case-by-case analysis.

Q4: Can small and medium miners keep up with this AI transition?

It’s challenging. Upgrading infrastructure for AI data centers requires substantial upfront capital, technical expertise, and supply chain resources—barriers that are significant hurdles for smaller miners.

Q5: Is the long-term economic model of mining still valid after the halving?

The sustainability of the post-halving mining economic model is under serious question. Based on current costs above $80k and coin prices around $60k, pure mining operations are not sustainable. This is a key driver behind the large-scale shift to AI. The ultimate equilibrium will depend on the interplay of Bitcoin prices, energy costs, and AI revenue streams.