AI Demand Rewrites Bitcoin Mining Economics and Site Values

For AI operators, a 10- to 15-year tenant lease can make a long, fixed-price power contract a natural hedge, according to Jeff Ko, chief analyst at ViaBTC.

Ko said AI data-center expansion is raising the opportunity cost of power for Bitcoin miners. Mining decisions now require more than cheap electricity: operators must compare hashing returns with AI and high-performance computing economics, including financing, hedging and the site's alternative value.

Ko said sites with strong grid access, fiber and high uptime increasingly hold value as data-center assets, while mining shifts toward cheaper, remote, intermittent or interruptible power. He does not expect AI to displace mining outright.

He said: “In short, I don’t think AI will simply displace Bitcoin mining. It is forcing miners to be much more disciplined about where each megawatt is best deployed.”

What can a mining site reuse for AI and HPC?

Ko drew a line at the main transformer. Upstream assets can often be reused, while much of the downstream data-center layer must be rebuilt. The grid connection is the prime asset: utility agreement, substation, interconnection position, land and permits.

Ko said large-load connections can take years to secure, so an energized site carries significant value for AI or HPC. ASIC miners have no practical AI use, and mining racks, containers and basic air-cooling systems generally cannot support GPU clusters.

AI and HPC sites require more fiber, cooling, redundancy and uptime, often including liquid cooling, UPS systems and backup generation. Ko said conversion requires substantial redesign and capital investment, not a simple ASIC-for-GPU swap.

Ko's conclusion: the transferable asset is mainly power access and land; mining hardware has little conversion value.

Which returns should miners compare?

Ko said the key test is risk-adjusted return per megawatt and per dollar of capital, rather than whichever business produces more revenue. Bitcoin mining metrics include hashprice, machine efficiency, all-in power cost, uptime and ASIC payback periods.

Miners must stress-test returns against Bitcoin price, network difficulty and future halvings, Ko said. For AI and HPC, he cited capex per megawatt, utilization, power reliability, cooling, network needs, financing costs, and customer-contract quality and duration.

AI may produce more revenue per megawatt, Ko said, but requires significantly more upfront capital and longer development. The sharper comparison is return after those costs and risks.

Why do power contracts and site location matter?

Ko said the strategic asset is deliverable capacity at a specific place and time. For miners, the more important contract terms are often flexibility rights, including curtailment, resale during price spikes and assignment to a third party, rather than the headline price.

AI demand is absorbing prime grid-connected sites in mature markets, Ko said. Miners are moving toward sites near power plants, stranded gas, curtailed renewables and surplus state-utility capacity, including parts of Africa, Latin America, Central Asia and the Middle East.

Ko said the shift is producing more modular, geographically diversified expansion rather than concentration in a few large sites. In his view, deliverability and schedule determine value, not just power price.

Where can AI improve mining operations?

Ko identified operational optimization as the clearest application. Mining sites collect real-time data on power use, machine temperature, hashrate, fan performance and failures while balancing power prices, hashprice and machine condition. With thin margins, a few percentage points of energy or uptime gains can materially lift profitability.

He said better forecasting works best alongside traditional optimization. At mining pools, behavioral models can detect account takeovers, suspicious payout-address changes, malicious firmware that redirects hashrate and attacks on pool infrastructure.

Ko disputed the idea that mining cooling offers a large unclaimed gain. Immersion-cooled, hydro-cooled and cold-climate air-cooled sites already use little energy beyond the machines, leaving less room to optimize.

Where does mining go as AI takes the best sites?

Ko traced mining from hydro-rich Chinese provinces to Texas and the Nordics. AI now competes for North America's best grid-connected sites. Unlike the 2021 migration out of China, driven by a sudden regulatory ban, this shift is gradual and market-driven. Ko expects it to be more orderly than the 2021 migration.

He expects AI and other higher-value grid users to push mining toward stranded hydro, curtailed renewables, excess generation and weaker-fiber areas. State-led energy monetization is also expanding across parts of Africa, Central Asia, the Middle East and South Asia.

“Cheap power alone is not enough,” Ko said. “Miners still need regulatory stability, grid reliability, logistics and access to capital.” He expects diversification to strengthen Bitcoin’s resilience, while miners balance lower power costs against political, operational and infrastructure risks.

Can miners complement AI data centers on the grid?

Ko said Bitcoin mining is highly interruptible: machines can cut load or shut down when prices spike or the grid comes under stress, then restart when conditions normalize. AI facilities need continuous, reliable power because workloads, customers and service-level agreements depend on uptime, backed by stronger redundancy, backup systems and stable grid capacity.

Ko said the requirement makes mining potentially complementary. Miners can absorb excess generation during low-demand periods and curtail when power becomes scarce, helping balance variable renewable supply.

Ko cautioned that technical flexibility does not guarantee economic willingness. Miners must still weigh lost mining revenue against the value of shutting down.

What would prove the convergence is structural?

Over the next 12 to 24 months, Ko said he would look beyond the first wave of deals. Renewals, capacity expansions and new long-term leases with investment-grade tenants would show that AI demand is durable rather than tied to the current investment cycle.

His second test comes if AI capital spending slows. Ko said contracts that remain intact without widespread renegotiations, cancellations or tenant defaults would point to a structural shift. A return to Bitcoin mining would suggest the change was cyclical.

Ko previously worked at J.P. Morgan Private Bank and founded data-visualization startup Chartipedia in Finland before entering crypto.

Under Ko's test, the next proof point is contract performance: durable AI demand should show up in renewals, expansions and investment-grade tenants after the capital spending cycle cools.