Why Is CoreWeave Spending Up to $39 Billion on AI Infrastructure?

 

A company with $2.575 billion of quarterly revenue spent $9.4 billion on capital expenditure in that same quarter. That is a ratio of 3.6 times, and next quarter's guidance says it is not coming down.

 

After reporting second quarter 2026 results on August 11, CoreWeave raised its full-year capital expenditure outlook to $35 billion to $39 billion. Per CNBC's report, the prior forecast had been $31 billion to $35 billion, and the company called for more than 1.85 gigawatts of active power by year end. This is the third increase to that guidance this year, starting from an initial range of $30 billion to $35 billion.

 

When a company keeps raising its spending plan there are usually only two explanations: demand exceeding expectations, or costs exceeding expectations. For CoreWeave both apply. Understanding that requires splitting the capital expenditure into three parts: what it buys, why it has to be bought now, and how the cost of financing it is entering the income statement.

 

 

 

Second quarter capital expenditure was $9.4 billion, slightly above the high end of guidance, against quarterly revenue of $2.575 billion, a ratio of roughly 3.6 times.

 

Full-year capital expenditure guidance rose to $35 billion to $39 billion from $31 billion to $35 billion, itself up from $30 billion to $35 billion at the start of the year.

 

Third quarter capital expenditure guidance is $11.5 billion to $13.5 billion against revenue guidance of $3.45 billion to $3.6 billion, a ratio still near 3.5 times, indicating spending intensity is not easing.

 

Third quarter interest expense is guided at $860 million to $940 million against adjusted operating income of $200 million to $260 million, roughly four times higher.

 

Active power reached 1.5 gigawatts in the second quarter after adding nearly 500 megawatts, more than tripling year over year, while contracted power stood at 3.7 gigawatts at quarter end and 4.2 gigawatts as of the earnings call.

 

Revenue backlog was $104.2 billion, up from $99.4 billion at the end of the first quarter, excluding more than $25 billion of new commitments secured in early Q3.

 

Second quarter operating cash flow was $679 million while financing activities supplied $10.071 billion, with $35 billion of debt on the balance sheet at quarter end.

 

Managed inference booked annualised recurring revenue grew from $1 million at launch to more than $100 million, with the company expecting to exit 2026 at at least $250 million.

 

 

 

Where the capital goes can be read directly from the capacity data. Per the earnings call summary published via Yahoo Finance, second quarter capital expenditure was $9.4 billion, slightly above the high end of guidance; active power reached 1.5 gigawatts after adding nearly 500 megawatts in the quarter, more than tripling year over year; and contracted power grew to 3.7 gigawatts in the second quarter and 4.2 gigawatts as of the call.

 

Setting those side by side gives a rough unit reference: $9.4 billion in a single quarter against roughly 500 megawatts of new active power. That figure covers GPUs, servers, networking equipment and data centre facilities rather than pure power interconnection, but it establishes one thing clearly: the investment per megawatt of active capacity runs into the billions.

 

More important is the gap between contracted and active power. Against 4.2 gigawatts contracted, 1.5 gigawatts is active, roughly 36%. The remaining near 2.7 gigawatts is capacity already committed but not yet running, and turning it into revenue-generating assets is precisely what the $35 billion to $39 billion is for.

 

 

A common assumption is that capital expenditure declines relative to revenue as scale grows. Third quarter guidance contradicts that.

 

The same summary shows third quarter guidance of $3.45 billion to $3.6 billion in revenue against $11.5 billion to $13.5 billion of capital expenditure. At midpoints, capex runs roughly 3.5 times revenue, essentially unchanged from the second quarter's 3.6 times.

 

The full-year ratio is lower but still extreme: a $37 billion capex midpoint against a $12.8 billion revenue midpoint is roughly 2.9 times. For comparison, the ratio implied by first quarter guidance was closer to 2.5 times. Spending intensity is rising rather than falling.

 

 

 

The three guidance increases need separating by cause. Per Cryptobriefing's report, when the company earlier raised the lower end of its capex range from $30 billion to $31 billion, it cited rising component costs as the reason.

 

This increase is larger, moving from $31 billion to $35 billion up to $35 billion to $39 billion, a midpoint rise of roughly 12%. Over the same period revenue guidance moved only from $12 billion to $13 billion up to $12.4 billion to $13.2 billion, a midpoint rise of roughly 2.4%.

 

The spending midpoint rose five times faster than the revenue midpoint. That asymmetry indicates the incremental capital serves delivery obligations further out rather than this year's revenue. The company's release does state that revenue backlog comprises remaining performance obligations whose recognition depends on delivering infrastructure and maintaining service availability.

 

 

The other side of the spending is the order book. CNBC reported that during the quarter Meta said it would spend an additional $21 billion with CoreWeave, that the company announced a multi-year agreement with Anthropic, and that it secured a $6 billion commitment from quantitative trading firm Jane Street. Revenue backlog rose from $99.4 billion at the end of the first quarter to $104.2 billion as of June 30, excluding more than $25 billion of new commitments secured in early Q3.

 

The ramp of a newer business line adds corroboration. The earnings call summary shows managed inference booked annualised recurring revenue growing from $1 million at launch to more than $100 million, with the company expecting to exit 2026 at at least $250 million. The absolute scale remains small, but the slope indicates the product mix is extending toward higher value-added services.

 

For investors following both AI compute and crypto, the competition between them for power and high-end chips is worth tracking. On venues offering crypto trading alongside stock futures, such as MEXC, relative performance between the two often signals shifts in capital preference early.

 

 

 

 

This is the most easily overlooked link in the capital expenditure story. Third quarter guidance from the earnings call includes adjusted operating income of $200 million to $260 million alongside interest expense of $860 million to $940 million.

 

At midpoints, interest expense is roughly 3.9 times adjusted operating income. The ratio is improving: per Indmoney's earnings preview, at the second quarter guidance midpoint interest expense would have been more than eleven times adjusted operating income. But even at four times, it establishes that the largest cost in this business right now is capital rather than operations.

 

The same preview explains why EBITDA can look strong while net income stays deeply negative: EBITDA excludes interest and depreciation, the two largest costs in CoreWeave's business, which means adjusted EBITDA functions for this company as a measure of operating scale.

 

 

The second quarter cash flow structure shows how financing underwrites the spending. Per Blockspace's summary, the quarter generated $679 million of operating cash and spent $6.422 billion on property and equipment, while financing activities supplied $10.071 billion including $13.457 billion of debt issuance partly offset by $3.884 billion of repayments.

 

One clarification: the $6.422 billion in the cash flow statement is cash actually paid for equipment, a different measure from the $9.4 billion capital expenditure figure, which typically includes finance leases and amounts incurred but unpaid. The difference between them is itself part of working capital.

 

CNBC reported that as of quarter end the company carried $35 billion in debt on its balance sheet to cover the cost of Nvidia GPUs and other equipment. That figure happens to match the low end of full-year capex guidance, implying that at the current pace the coming year's financing needs sit at the same order of magnitude as the existing debt stock.

 

 

This is the core of the headline question and the hardest part to answer definitively.

 

The supporting evidence is contractual. The overwhelming majority of the $104.2 billion backlog consists of remaining performance obligations tied to committed customer contracts rather than a sales pipeline. Against $37 billion of annual capital expenditure, backlog is roughly 2.8 times that figure. On that basis, the spending has orders behind it.

 

Caution belongs on duration and conversion. Backlog recognition depends on delivery, delivery depends on capacity, and capacity depends on this capital expenditure, forming a closed loop rather than three independent safeguards. Analysis also indicates that roughly 21% of backlog is expected to be recognised beyond four years, double the 10% share a year earlier, meaning the weighted average duration is lengthening.

 

The competitive picture is also shifting. CNBC noted that SpaceX has begun selling excess computing capacity, that Meta has considered launching a cloud business, and that rival Nebius gained 5% in extended trading. CFO Nitin Agrawal responded that even with increased competition the company is seeing demand, pricing and margin all expanding, which he framed as a signal of growth in the product and overall growth within an already massive addressable market. That is management's assessment, and investors should verify it against subsequent pricing and margin data.

 

External constraint is another variable. The earnings call summary notes that regulatory and community opposition to data centre builds could affect future deployment locations, though current guidance remains unaffected. CNBC separately reported that New York's governor signed an executive order in July establishing a moratorium on new large-scale data centres.

 

 

The first risk is capital efficiency. The capex-to-revenue ratio has risen from roughly 2.5 times under first quarter guidance to roughly 2.9 times now, with the third quarter running near 3.5 times. If that keeps climbing, the lag between construction and revenue realisation is lengthening and output per unit of capital is deteriorating.

 

The second is financing dependence and rates. Second quarter operating cash flow of $679 million against $6.422 billion of equipment payments leaves a gap approaching tenfold, filled by roughly $10 billion of financing activity. With interest expense already around four times adjusted operating income, any rise in funding cost directly flattens the slope of margin improvement.

 

The third is customers and delivery. Backlog depends heavily on a few hyperscale customers, with Meta alone adding $21 billion. Roughly 21% of backlog is recognised beyond four years, and longer duration means longer exposure to technological obsolescence.

 

The fourth is the external environment, covering power interconnection progress plus regulatory and community resistance to data centre construction.

 

On scenarios, the base case is capital expenditure deploying as guided, active power exceeding 1.85 gigawatts by year end, revenue accelerating in line with third quarter guidance, and adjusted operating margin improving sequentially from the second quarter's 5% while net income stays weighed down by interest. A second is margin improvement running ahead of expectation with higher value-added lines such as managed inference scaling, shifting the valuation logic from revenue growth to profitability. A third is another capex increase without a matching revenue raise; given the current five-times divergence between the two revisions, that would sharply intensify questions about capital efficiency.

 

Status distinctions: capital expenditure, capacity, backlog and guidance figures come from the company's official release and public call records. Third-party ratio calculations and interpretations are those institutions' judgements. All figures covering the remainder of 2026 are guidance rather than confirmed results.

 

 

The most direct answer to the headline question is this: contracted power stands at 4.2 gigawatts while active power stands at 1.5 gigawatts. That 2.7 gigawatt gap is where $35 billion to $39 billion goes. Contracted power is a right on paper; active power is the revenue source, and converting one into the other requires GPUs, racks, cooling and grid connection, all of which must be paid for before any revenue appears. Understanding that conversion explains why a company with revenue in the low tens of billions is spending close to forty billion.

 

But what deserves scrutiny is not the absolute figure. It is the direction of the ratio. Second quarter capital expenditure of $9.4 billion against $2.575 billion of revenue is roughly 3.6 times; the third quarter guidance midpoint remains near 3.5 times; and the full-year ratio has risen from roughly 2.5 times under first quarter guidance to roughly 2.9 times. Growing scale has not improved capital efficiency. If that ratio still sits near three times through 2027, this is not a temporary construction-phase phenomenon but the normal structure of the business, and the valuation framework needs to shift from growth equity to capital-intensive infrastructure asset.

 

Three misreadings look likely. The first is treating the $104.2 billion backlog as a cushion for capital expenditure. Backlog recognition depends on delivery, delivery on capacity, and capacity on capex, making them a closed loop rather than independent safeguards. The second is evaluating this company on adjusted EBITDA. Second quarter adjusted EBITDA of $1.51 billion looks comfortable, but the measure excludes interest and depreciation, and third quarter guidance puts interest expense at $860 million to $940 million, roughly four times adjusted operating income. Those excluded items are the largest costs. The third is underweighting what the capex increase signals. The capex midpoint rose roughly 12% while the revenue midpoint rose roughly 2.4%, and a fivefold divergence means the incremental spending serves obligations further out while adding depreciation and interest burden in the near term.

 

Three verifiable metrics deserve tracking. First, active power as a share of contracted power, currently around 36%, the only physical route by which capital expenditure becomes revenue, with its rate of improvement determining the payback cadence. Second, the quarterly capex-to-revenue ratio: a fall below three times in the fourth quarter would indicate the construction peak is passing, while holding near 3.5 times would indicate expansion is still accelerating. Third, the multiple of interest expense to adjusted operating income, which has fallen from above eleven times under first quarter guidance to roughly four times in the third quarter guide. The slope of that improvement matters more than its absolute level.

 

The cross-asset lesson is that for companies in a heavy-asset expansion phase, the valuation anchor belongs on capital efficiency rather than revenue growth. CoreWeave and bitcoin mining companies share a financial structure closely: both must deploy equipment and power before revenue appears, both fill cash flow gaps with debt and equity, and both are constrained by the physical ceiling of power supply. Crypto has repeatedly demonstrated across cycles that the operator leading during expansion is not necessarily the eventual winner, and that long-run outcomes are determined by the combination of capital cost per unit of compute and cost of financing. When evaluating names like this, dividing capital expenditure by incremental active capacity and comparing that against financing cost gets closer to the truth than reading any order announcement. From a risk management standpoint, while capital expenditure runs at several times revenue, controlling position size is worth more than calling direction.

 

This analysis rests on the company's official release, public call records and credible reporting available now. Subsequent operating results, the financing environment and construction progress could each change the conclusion, and no single scenario should be treated as a fixed expectation.

 

 

 

Second quarter capital expenditure was $9.4 billion, slightly above the high end of guidance. Full-year guidance has been raised to $35 billion to $39 billion from $31 billion to $35 billion, itself up from $30 billion to $35 billion at the start of the year. Third quarter guidance is $11.5 billion to $13.5 billion. At midpoints, full-year capex runs roughly 2.9 times the $12.8 billion revenue guidance midpoint.

 

 

Primarily the conversion of contracted but non-operational power capacity into revenue-generating assets. Active power rose to 1.5 gigawatts in the second quarter after adding nearly 500 megawatts, while contracted power stood at 4.2 gigawatts as of the earnings call, leaving a gap of roughly 2.7 gigawatts. The spending covers GPUs, servers, networking equipment and data centre facilities, with the company targeting more than 1.85 gigawatts of active power by year end.

 

 

Both demand and cost. When the lower end moved from $30 billion to $31 billion, the company cited rising component costs. This latest move from $31 billion to $35 billion up to $35 billion to $39 billion came alongside Meta's additional $21 billion commitment, a multi-year agreement with Anthropic and a $6 billion commitment from Jane Street, with backlog rising to $104.2 billion.

 

 

Not currently. Second quarter capital expenditure ran roughly 3.6 times revenue, the third quarter guidance midpoint is near 3.5 times, and the full-year ratio has risen from roughly 2.5 times under first quarter guidance to roughly 2.9 times. Growing scale has not reduced spending intensity. Whether this is a temporary construction-phase effect or the normal structure requires watching the ratio through 2027.

 

 

It has become the largest single expense. Third quarter guidance puts interest expense at $860 million to $940 million against adjusted operating income of $200 million to $260 million, roughly 3.9 times at midpoints. That ratio is improving, having exceeded eleven times at the second quarter guidance midpoint. The balance sheet carried $35 billion of debt at quarter end, covering the cost of GPUs and other equipment.

 

 

By size yes, but the two are not independent. Against a $37 billion annual capex midpoint, backlog is roughly 2.8 times that figure. The complication is that backlog recognition depends on delivery, delivery on capacity, and capacity on capital expenditure, forming a closed loop. Analysis also indicates roughly 21% of backlog is expected to be recognised beyond four years, double the 10% share a year earlier, meaning duration is lengthening.

 

 

Competition is intensifying. Reporting indicates SpaceX has begun selling excess computing capacity and Meta has considered launching a cloud business. The CFO responded that even with increased competition the company sees demand, pricing and margin all expanding. As evidence from newer lines, managed inference booked annualised recurring revenue grew from $1 million at launch to more than $100 million, with the company expecting to exit 2026 at at least $250 million. These are management disclosures and should be verified against subsequent data.

 

 

Mainly power interconnection and construction permitting. The earnings call noted that regulatory and community opposition to data centre builds could affect future deployment locations, though current guidance remains unaffected. Reporting also indicates New York's governor signed an executive order in July establishing a moratorium on new large-scale data centres. Such constraints act directly on how fast capacity comes online, and capacity is the only route by which capital expenditure becomes revenue.

 

 

This article is provided for informational and research purposes only and does not constitute investment advice, financial advice, legal advice, tax advice or any recommendation to transact. The financial figures, capacity data and guidance referenced here come from the company's official release, public earnings call records and credible reporting; all figures covering the remainder of 2026 and beyond are company guidance rather than confirmed results and actual outcomes may differ materially; third-party analyst ratio calculations, views and interpretations are those institutions' judgements rather than statements of fact; and every calculation from public data in this article is a simple arithmetic illustration. Prices of crypto assets, equities and other related financial instruments can move sharply over short periods, and shares of rapidly expanding, not-yet-profitable companies are especially volatile around earnings windows, with the possibility of total loss of principal. Futures contracts referencing a stock carry leverage risk that magnifies losses proportionally, holding such contracts is not equity ownership and confers no shareholder rights, and the risk profile differs fundamentally from holding shares directly. Historical performance, technical indicators and on-chain data cannot guarantee future outcomes. Readers should conduct their own independent research, verify official information directly, and evaluate any decision against their own financial circumstances, investment objectives, experience and risk tolerance, consulting a qualified professional adviser where appropriate. The MEXC Crypto Pulse team accepts no liability for any direct or indirect loss arising from use of or reliance on the information in this article.

 

 

James Mitchell specializes in technical analysis, market trends, and trading strategies for both Bitcoin and altcoins. Based in London, he has over 10 years of experience in financial markets. Before joining MEXC Learn, James worked as a senior analyst at a leading European investment firm, where he developed expertise in risk management and quantitative trading. His transition to cryptocurrency markets began in 2017, and he has since become recognized for his data-driven approach. He holds a Master's degree in Financial Economics from the London School of Economics. His analytical approach combines traditional technical analysis with on-chain metrics to provide readers with actionable insights.

 

Areas of Expertise:

 

 

Want the fastest access to MEXC's latest updates? Join our official Telegram group now!

MEXC Platform: Download App | Zero Fee Trading | Buy USDT Now

Join MEXC Community: X (Twitter) | Telegram | Discord

Get Insights: MEXC Learn | Terms of Service | MEXC Blog | MEXC News

Account Verification: Understand KYC | How to Complete KYC

External Content Platforms: Substack | Medium | Paragraph | LinkedIn | X(News)