ViaBTC CEO Haipo Yang: Will AI Push Bitcoin Mining out of the Market?

By Haipo Yang, Founder and CEO of ViaBTC

Bitcoin mining has gone through a noticeable adjustment over the past year. In October 2025, network hashrate climbed above 1.1 ZH/s before trending lower, and this year it has fallen back toward 900 EH/s on several occasions. Mining difficulty dropped 11.16% in a single adjustment in February, followed by another 10.09% decline in June—moves rarely seen since 2021.

At the same time, another shift has attracted growing attention: more mining companies are moving their focus toward AI and high-performance computing (HPC). Core Scientific reported a negative 56% gross margin for self-mining in the second quarter, while its data center colocation business generated nearly $80 million in gross profit. At TeraWulf, HPC leasing already accounted for around 71% of total revenue during the same period. Companies that once built their businesses primarily around Bitcoin mining are increasingly redirecting sites, power, and capital toward AI.

Put these two trends together, and an obvious conclusion seems to follow: AI is taking hashrate away from Bitcoin. Take that argument one step further, and another concern emerges—if network hashrate keeps falling, could Bitcoin’s security eventually come under pressure?

There is a certain logic to that concern. AI is indeed absorbing some of the resources that used to go into mining. But I think the more useful question is which resources AI can actually take, which ones it cannot, and whether what remains is enough for Bitcoin mining to continue making economic sense.

What Are AI and Bitcoin Mining Actually Competing For?

AI compute hardware and Bitcoin ASICs are not interchangeable computing resources. Mining Bitcoin with GPUs today is generally uneconomical, while an ASIC designed for SHA-256 cannot be repurposed to run large AI models. The real competition happens further upstream: chip capacity, capital, land, power, and existing data center infrastructure. Among these, one resource is becoming particularly scarce—the ability to access large amounts of reliable power quickly.

For an AI company, two plots of land of the same size can be completely different assets. One may already have substations, grid capacity, fiber connectivity, and other infrastructure in place, allowing it to support AI workloads after the necessary upgrades. The other may sit right next to a power plant but still require years of work to secure power and build the supporting infrastructure.

Time matters enormously in AI. AI chips and models are evolving quickly, while substations and other major power infrastructure are built on timelines measured in years. For companies racing to add compute capacity, power that is already available also means time already saved. Many mining companies happen to have completed precisely this slowest part of the process: they secured land, substations, and grid capacity years ago, when those resources were far less competitive than they are today. AI companies are now willing to pay considerably more for the same infrastructure.

That is why the logic behind the mining industry’s pivot is relatively straightforward: what these companies are offering AI is not simply electricity, but power access that is already in place.

What Aren’t AI and Bitcoin Mining Competing For?

Move away from large-scale data centers, however, and a very different type of power demand appears. Over the years, we have worked with many small and mid-sized miners. Some operate in settings where a small factory has connected a dozen or so Bitcoin mining rigs to its rooftop solar system. The factory’s own production takes priority, and any excess solar generation can then be used for mining. When surplus power is available, the rigs run; when it falls, the operation scales down or shuts off. The heat produced by the rigs can even be recovered for hot water used by the factory.

These operators did not install solar panels specifically to mine Bitcoin. They are simply finding another use for electricity that would otherwise have very little economic value. In this kind of setup, mining does not need to maximize 24/7 utilization. The goal is to extract more value from electricity that would otherwise sit idle. Once normal household or industrial demand has been met, surplus power that earns little when sold back to the grid—or cannot be exported because of local transmission constraints—can instead be absorbed by mining rigs.

From the perspective of the power system, mining rigs have a useful characteristic: they are highly adjustable loads. They can run when surplus power is available, scale down when supply falls, and shut off when necessary. They do not require that electricity to appear at exactly the same time every day, nor do they require an expensive high-availability power system simply to keep a long-running computation alive.

AI effectively says, “When I need power, you need to have it.” This type of mining says, “When you have cheap surplus power, I can run.”

AI can certainly use solar and wind power, and many operators are already building infrastructure around renewables. But if intermittent solar or wind generation is going to support high-availability training or inference workloads, it usually needs storage, grid power, or another stable energy source to turn it into continuously available electricity. Turning intermittent electricity into dispatchable electricity comes at a cost. A kilowatt-hour of excess rooftop solar at midday and a kilowatt-hour available on demand in the evening are physically the same amount of energy, but economically they can be worth very different amounts.

The need to make use of constrained or otherwise underutilized power is not limited to households or small energy projects. This year, energy group ENGIE said it was evaluating battery storage or Bitcoin mining facilities at its large Assú Sol solar project in Brazil, where grid constraints mean not all available solar generation can always be absorbed.

From a rooftop system supporting a dozen mining rigs to a solar project measured in hundreds of megawatts, the underlying question is the same: what can you do with a unit of electricity that nobody needs at that moment and that cannot easily be sent somewhere else?

AI is willing to pay more for stable, highly available power. Mining can make use of electricity that is less convenient. Curtailed wind and solar, associated gas at oil fields, remote small-scale hydropower, and negative-price periods may all be cheap for very different reasons—intermittency, geography, or transmission constraints. Those characteristics can make them unsuitable for continuously operating data centers while still making them viable for mining.

So while AI will raise the value of premium sites, it will not compete equally for every source of energy. The future advantage of Bitcoin mining may increasingly lie closer to the energy source itself rather than in standardized data centers.

Mining Rigs Will Find New Owners, and Hashrate Will Move

As large mining companies scale back self-mining, some of their mining rigs will gradually make their way into the secondary market. A machine that no longer makes economic sense in a high-electricity-cost data center may still work elsewhere. Sell it at a lower price and move it to a site with cheap hydropower, surplus solar, or another low-cost energy source, and the economics can change again.

Of course, cheaper hardware and cheaper electricity solve different problems. Whether a mining rig should stay online is still largely determined by its energy efficiency and electricity rate. A lower purchase price instead reduces capital requirements, shortens the payback period, and eases financing pressure. For the factory using surplus solar in the example above, the rigs may never be expected to run around the clock. If the electricity itself is cheap enough, lower-priced second-hand machines may make sense despite weaker energy efficiency because the upfront capital cost is smaller and the operation can tolerate lower equipment utilization. Different generations of mining rigs can therefore end up matching different electricity-rate ranges.

Over the past few years, one of the defining trends in mining has been institutionalization. Large mining companies raised capital from public markets, built data centers with capacities measured in hundreds of megawatts, and brought a growing share of network hashrate into listed corporate structures. That path may now be becoming less one-directional. Public mining companies will remain important participants, but future hashrate growth may not come primarily from them. Private operators with direct access to inexpensive energy, small and mid-sized miners, and energy projects themselves may once again find more room to compete.

In the ten years we have operated ViaBTC Pool, we have never served only large institutions. There have always been many smaller miners operating under very different energy conditions. They do not publish earnings reports or hold investor calls, so they rarely appear in public discussions. But as mining expands further into distributed solar, remote energy sources, and other fragmented low-cost power environments, it becomes clear that these miners never actually left the market.

The more diverse and distributed the miner base becomes, the more important mature infrastructure becomes as well. Large mining companies can build their own operations, finance, and asset-management teams. Smaller miners care about simpler things: stable service, transparent reward calculation, flexible payment methods, and keeping the cost and friction of connecting and managing assets as low as possible. A mining pool’s job is to standardize as much of that complexity as possible, so miners of different sizes can focus on what they do best—finding the right machines and the right electricity.

Taken together, the impact of AI looks less like a simple displacement of mining and more like a reallocation of resources. Premium sites, power, and capital move toward the workloads that value them most, while mining rigs continue searching for energy sources and operators that can make mining viable. Rather than saying AI is pushing Bitcoin mining out of the market, I see it more as a process of sorting: the sites best suited to AI move toward AI, while electricity better suited to mining continues to find mining rigs.

Difficulty Adjustment Will Keep Pushing the System Toward a New Equilibrium

So if hashrate falls, does the system eventually become unstable?

Suppose a group of miners shuts down at the same time. Network hashrate falls and blocks come in more slowly. At the next difficulty adjustment, mining difficulty falls accordingly. All else being equal, the same mining rig can then produce more BTC over the same amount of time. That can lower its mining break-even price (shutdown price), making some machines that had previously been switched off economically viable to operate again.

Bitcoin does not stop miners from leaving. Its difficulty adjustment mechanism simply changes the economics after hashrate moves, forcing the market to run the numbers again.

The hashrate that comes back may come from a large mining farm with long-term low-cost power, or from smaller miners using hydropower, surplus solar, or other unconventional energy sources. As hashrate exits and re-enters, the system continues moving toward a new equilibrium shaped by electricity rates, machine efficiency, and the Bitcoin price.

Difficulty adjustment, of course, does not mean that lower hashrate has no implications for security. Total hashrate still matters because it affects the cost of attacking the network. But it is equally simplistic to assume that a temporary decline in hashrate automatically means the network is heading toward a security crisis.

The fifth Bitcoin halving in 2028 will be a more predictable stress test. Nobody can know with certainty how AI demand, Bitcoin prices, or regulation will develop. What we do know is that the fixed BTC reward created with each block will be cut in half again. That is written into the protocol. The math is straightforward: the fixed reward drops by half. If higher Bitcoin prices, greater transaction fee revenue, and subsequent difficulty adjustments fail to make up enough of that gap, miners will naturally face greater pressure on earnings.

By then, competition may be less about sheer scale or who owns the latest machines, and more about overall cost discipline: who can secure cheaper and more flexible energy, acquire equipment at a lower cost, and operate with stronger cash flow and greater efficiency.

Bitcoin has already gone through four halvings. Over the past decade, we have seen hashrate, prices, and the structure of the mining industry rebalance again and again. I do not expect the fifth halving to be any different.

Final Thoughts

So, back to the original question: will AI take hashrate away from Bitcoin? I think some premium power and site capacity will naturally move toward AI businesses that are willing to pay more for those resources. But mining rigs will continue searching for new owners and new sources of power.

On one side, a mining company converts a grid-connected site with hundreds of megawatts of capacity to support AI workloads. On the other, a small factory with rooftop solar switches on a dozen mining rigs when it has surplus power. AI will make the best sites more expensive, and it will make some of today’s mining models uneconomical. But mining has one unusual advantage: as long as there is a unit of electricity that is cheap enough and difficult for other industries to use efficiently, someone will run the numbers again.

One of the things I find most interesting about Bitcoin is that the system was created without any knowledge that AI would ever exist. It does not know who is shutting down, who is building solar capacity, or where a second-hand mining rig is being moved. It simply looks at the previous 2,016 blocks and adjusts one parameter. The rest is left to the market—and to every miner making their own decision.

A system that knows nothing about prices, participants, or what is happening outside the network can still allow thousands of independent actors to reallocate resources on their own.

That design has been running for seventeen years. I think it will run for much longer.