TeraWulf's $19 Billion Anthropic Deal Sounds Like a Transformation - Until You Read the Balance Sheet

I always keep an eye out for false narratives that take the market by storm, and TeraWulf's Q2 earnings week has delivered one of the most seductive I've seen in a while. The headline reads like a tech IPO prospectus: a 20-year, $19 billion lease with Anthropic for 401 megawatts of AI data center capacity in Kentucky. As of early July, the stock was up more than 80% this year on the promise. But there is a massive gap between a signed lease and cash in the bank - and it's a gap that matters enormously for anyone holding the stock today.

Here is the reality underneath the headline. TeraWulf generated $44.8 million in revenue last quarter, missing analyst estimates of $50 million. That's right - the company whose biggest talking point is a $19 billion contract brought in $45 million last quarter. Of that, $31.9 million came from high-performance computing (HPC) lease revenue, which was 71% of total revenue and represents a real shift from the bitcoin mining business it used to be. HPC lease revenue was zero a year ago, so the trajectory is directionally correct. But the scale is still tiny.

The reported loss of $1.94 per share looks grotesque - it missed estimates of negative $0.27 by more than 600%. But the headline loss is almost entirely non-cash, driven by a $755.7 million warrant fair value charge. That charge arises from convertible notes the company issued and the warrants attached to them, which must be marked to market each quarter. When the stock price swings, the accounting swings with it. The warrant charge is a real balance-sheet obligation in economic terms, but it is not a quarterly cash outflow.

The cash outflow problem is different, and it is real. TeraWulf paid $125.7 million in cash interest in the second quarter, up from $5.3 million in the first quarter. The spike reflects the first semiannual interest payment on $2.5 billion of convertible notes at the parent level. Add to that $3.2 billion in senior secured notes at the WULF Compute subsidiary and $100 million in bridge loans at the Kentucky subsidiaries, and you have $5.8 billion in total debt. That is a mountain of fixed obligations for a company generating $45 million per quarter.

So where does the cash come from? TeraWulf ended Q2 with approximately $3.0 billion in cash and restricted cash, down from $3.1 billion at the end of Q1. It also just closed a $250 million revolving credit facility with a syndicate of global banks, which provides a backstop but adds more debt to the structure. The company is also selling its 50.1% majority stake in the Abernathy joint venture - a Texas data center it co-developed with Fluidstack - for approximately $530 million. It originally invested roughly $450 million, so this is a modest premium on invested capital, but it signals that TeraWulf needed to recycle capital. You don't sell a controlling interest in a $9.5 billion contracted-revenue project unless you need liquidity.

The Anthropic deal itself is structurally interesting but temporally distant. First power delivery is expected in the second half of 2027 - more than 12 months away. The $19 billion figure is the sum of 20 years of contracted rent over the initial term, or up to $33 billion if Anthropic exercises both five-year extensions. Annualized, that works out to roughly $950 million per year over the initial term. That would be a transformational revenue stream if it materializes, but it requires TeraWulf to build approximately 400 megawatts of new data center capacity, finance that construction, execute on schedule, and hope that Anthropic - still a relatively young AI company compared to the hyperscalers - remains solvent and committed for two decades.

There is also the question of what the Anthropic deal replaces. TeraWulf is simultaneously acquiring the Muskie Data Campus in Eastern Kentucky with up to 1 gigawatt of contracted electric service from Kentucky Power Company. It still has 336 megawatts under construction at Lake Mariner in New York, where it is delivering capacity to Fluidstack with Google backstopping $600 million of Fluidstack's lease obligations. The pipeline is aggressive. Too aggressive, in my opinion, given the debt load and the revenue scale.

I am not saying the thesis is wrong. The structural logic is sound. Access to power is becoming the binding constraint on AI infrastructure buildout, and TeraWulf's vertically integrated model - controlling energy assets, building on the cheap, and locking in long-term leases with investment-grade or investment-grade-backed counterparties - is exactly the right play for the AI era. HPC lease revenue becoming 71% of the mix, up from zero a year ago, proves the pivot is real, not just marketing.

But I am saying the market has priced in a version of that thesis that the financials cannot yet support. The stock at $18.48 - down 2.1% today despite the Q2 results - had run more than 80% this year as of early July. That multiple expansion has been fueled by signed leases and megawatt announcements, not by free cash flow. There is no dividend. There is no free cash flow generation at current revenue levels. There is a $5.8 billion debt stack. And the biggest contract in the company's history won't produce a dollar until late 2027.

That is the false narrative: that a signed multi-decade lease is equivalent to revenue. It is a promise, not performance. And in the AI infrastructure sector, where construction timelines slip, counterparties renegotiate, and capex requirements escalate, the gap between signing and collecting is where value gets destroyed.

I rate TeraWulf as a Hold. The long-term structural thesis of power-advantaged data center infrastructure has merit, and the company is executing on its Lake Mariner deliveries and customer credit support milestones. But the near-term financials are brutal - $125 million per quarter in interest service against a $45 million revenue base - and the market has already run hard on promises that are still more than a year away from first cash flow. For investors who can tolerate the construction risk and the debt overhang, a small position makes sense as a long-dated AI infrastructure bet. For anyone looking for current cash flow, yield, or earnings power, this is not it yet. The stock needs to deliver on the buildout, convert those contracted megawatts into operating revenue, and prove the model generates free cash flow before the valuation can comfortably sustain the run it has already had.