Bitcoin miners are being hit from two sides. The April 2024 halving cut crypto mining rewards directly from 6.25 BTC to 3.125 BTC. Then came a drop in bitcoin prices, making an already difficult position worse. JPMorgan’s data lays out the severity: after the halving, mining companies’ revenue plunged 46%, while gross profit fell 57%. In February alone, monthly gross profit dropped another 9%.

Under that pressure, mining companies have almost simultaneously turned their attention to AI.

Companies such as Riot and Cipher have led the shift. They are converting their crypto mining computing power into high-performance computing resources and leasing it to AI companies that need capacity, or selling ASIC chips directly. Cipher has even secured an $8.5 billion AI hosting contract, using cash flow from its crypto mining business to fund the transition.

Their confidence comes from assets they already have: low-cost power and data center facilities. Those are exactly the core resources AI computing power needs most urgently.

But the transition story is starting to look less straightforward. In January, Chinese company DeepSeek said its models cost far less than OpenAI’s. That claim immediately made the market question how many data centers are really needed. JPMorgan also believes that even if miners move into AI, they cannot escape the broader industry’s pressures.

The more uncomfortable issue is that valuations contain a bubble.

JPMorgan values mining companies using an SOTP model, adding together crypto mining cash flow, power assets and the value of coins held. By that calculation, miners such as Hut 8 that have picked up the AI narrative are valued well above peers.

At the same time, the bank upgraded Cipher but lowered its price targets for MARA and Riot, explicitly pointing to the risk of shareholder dilution.

The anchor for mining-company valuations appears to be loosening. The standard is slowly shifting from how many coins a company can mine to how much potential its AI computing power may have.

In that process, the value of power assets is being overstated, while no one has yet proven how real the AI demand actually is.

If revenue from crypto mining keeps shrinking and the AI business fails to live up to expectations, how should these companies selling computing power be valued?