Bernstein affirms Outperform ratings on TeraWulf

Bernstein maintained its Outperform rating and $36 price target on TeraWulf after the Bitcoin miner’s second-quarter results showed high-performance computing, or HPC, had become the company’s main source of revenue. TeraWulf reported $32 million in HPC revenue, representing 71% of total quarterly revenue, as it begins converting power capacity once used primarily for cryptocurrency mining into infrastructure for artificial-intelligence computing customers.

TeraWulf shares closed at $18.07 on Aug. 5, leaving Bernstein’s target almost double the closing price. The firm’s bullish case rests less on Bitcoin mining and more on contracted data-center leases with Anthropic, Fluidstack and Core42, which together account for more than $27 billion in contracted revenue, according to TeraWulf.

Bernstein’s valuation assumes TeraWulf’s mining business is removed from its model entirely by 2028. That approach places the company alongside emerging AI data-center infrastructure providers rather than publicly traded miners whose earnings remain closely tied to Bitcoin’s price and network economics.

The shift is already visible in TeraWulf’s operating figures. The company said it had delivered 102 IT megawatts of revenue-generating capacity, linked to roughly $180 million in annualized recurring revenue. It expects contracted capacity to reach 839 IT megawatts during the first half of 2028.

Long-term leases drive the valuation case

The three major customer agreements are structured as modified gross leases, under which TeraWulf retains responsibility for some operating expenses while receiving contracted lease payments. The company expects stabilized net operating income margins of roughly 85% under those arrangements.

Across its full contracted order book, TeraWulf forecasts average annual revenue of more than $1.8 billion and average annual net operating income above $1.5 billion. Those projections depend on the sites being built, energized and fitted with the specialized equipment required for HPC workloads.

Bernstein, whose coverage is led by analyst Gautam Chhugani and began in June, values TeraWulf at 21 times one-year forward enterprise value to EBITDA based on steady-state 2030 earnings, discounted back to the present. EBITDA measures earnings before interest, taxes, depreciation and amortization, and is commonly used to compare infrastructure companies with differing financing structures.

The firm’s model gives greater weight to the predictability of long-duration customer contracts than to mining output. That distinction could reduce the sensitivity of TeraWulf’s projected revenue to Bitcoin market cycles, though it also makes execution on construction, financing and customer delivery schedules central to the company’s outlook.

TeraWulf recently expanded one of its largest agreements. The company amended its Fluidstack lease, increasing total contracted revenue by about $300 million to $7.2 billion over 10 years. TeraWulf also agreed to contribute $150 million toward tenant fit-out costs, adding to the capital required before the customer’s operations are fully online.

Rising build costs test projected returns

TeraWulf’s biggest challenge is the cost of converting available power into operational AI data-center capacity. Bernstein’s updated development-cost guidance puts capital expenditure at $10 million to $12 million per IT megawatt, above the earlier $8 million to $10 million range associated with the first Fluidstack contracts.

The increase is also apparent at the company’s Lake Mariner campus in New York. TeraWulf’s cost there rose to about $9.1 million per megawatt during the quarter, compared with roughly $8.6 million per megawatt that had previously been financed. A prior brownfield development approach had been associated with costs around $7.2 million per IT megawatt.

Higher construction costs do not automatically undermine the lease economics, particularly if contracted rent and high projected operating margins are achieved. They do narrow the room for delays, equipment overruns or more expensive financing. The $150 million Fluidstack fit-out contribution illustrates how a large revenue contract can also require substantial upfront spending from the infrastructure provider.

TeraWulf’s next major development is the 1-gigawatt Muskie campus in Kentucky. The company plans a first 500-megawatt phase for the fourth quarter of 2028, with full build-out targeted by 2030. Muskie sits within a broader pipeline totaling roughly 3.6 gigawatts of gross power across five sites in Kentucky, Maryland and New York.

That pipeline gives TeraWulf a potentially large supply of power for future AI infrastructure contracts, but it also extends its exposure to development timelines. Large-scale HPC facilities require more than land and grid interconnection: developers must secure equipment, build substations and cooling systems, and meet tenants’ technical requirements before lease revenue begins.

Lake Mariner contracts remain outside New York restriction

TeraWulf has said a temporary New York data-center moratorium does not affect capacity already leased to Core42 and Fluidstack at Lake Mariner because that capacity is permitted. The distinction protects the company’s existing contracted build-out at the site, though future expansion in New York could face a less predictable policy environment.

Bernstein identified customer concentration as a downside risk to its price target. Anthropic, Fluidstack and Core42 account for the company’s contracted revenue book, so delays, changing computing requirements or financial stress at any one customer could have an outsized effect on expected cash flow.

The research firm also disclosed that Chhugani holds long positions in various cryptocurrencies. Bernstein’s disclosures state that SG and/or its affiliates beneficially own 0.5% or more of TeraWulf’s issued share capital or another class of common equity securities with a net long or short position.

TeraWulf’s second-quarter results show the company is moving from a mining-led revenue base toward a concentrated, capital-intensive AI hosting model. The $27 billion-plus contract book gives that transition unusually clear commercial backing for a former miner, while rising per-megawatt costs and reliance on three customers will determine whether its projected 2030 data-center earnings can be delivered.

Disclaimer: The content on this page is provided for general informational purposes only and does not represent the views or financial advice of Toobit. We make no guarantees regarding the accuracy or completeness of this information and shall not be held liable for any errors, omissions, or outcomes resulting from its use. Investing in digital assets involves risk; users should independently evaluate their financial situation and the risks involved. For further details, please consult our Terms of Service and Risk Disclosure.