A-Portfolio

The company in one breath

CoreWeave, ticker CRWV on the Nasdaq, rents high-performance computing built around Nvidia (NVDA) graphics chips to artificial-intelligence labs and large companies. Customers get the chips, networking, storage, software, and technical help without building the whole data-center stack themselves.

I think of it as an AI landlord. The tenants want enormous blocks of computing power. CoreWeave builds the property, fills it with expensive machinery, and collects rent through long contracts. The difference is that an apartment building can survive one ugly software cycle. A warehouse full of aging chips may not.

Why now

CoreWeave just showed the cleanest version of its bull case and the ugliest version of its balance sheet in the same quarter. Second-quarter revenue hit $2.575 billion, up 112%. That means the company more than doubled sales in one year. Good for a young infrastructure leader would be growth above 50%, so demand is not the problem.

Revenue backlog reached $104 billion. Backlog means contracted service that CoreWeave has not delivered or counted as revenue yet. It equals more than eight times the midpoint of this year's sales forecast. Good backlog gives years of visibility and converts into revenue on schedule. Bad backlog looks enormous while power delays, customer changes, or financing keep the company from serving it.

The stock closed Wednesday at $80.925. That leaves it 102% above its $40 IPO price, which means the public market has doubled the original bill, but 47% below its $153.09 peak, which means the summer mania already lost almost half its height. Good entry quality means the price falls faster than the business weakens. Here, the stock cooled, but the valuation still assumes a lot goes right.

The debate is brutally simple. Bulls see the fastest-growing scaled AI cloud outside the giant platforms. Bears see $35.1 billion of debt, enormous equipment spending, and customers that can build their own capacity. Both sides have receipts.

The numbers, translated

• Second-quarter revenue was $2.575 billion, up 112%. Revenue is money earned before expenses. Good for this stage is growth above 50%, so CoreWeave cleared the growth bar by a mile.
• Revenue backlog was about $104 billion. This is contracted service still waiting to become reported sales. Good means the company converts backlog into revenue without cancellations or construction delays, and a backlog above three times annual sales already counts as strong. CoreWeave sits above eight times this year's forecast midpoint, which is extraordinary and demanding.
• Management expects $12.4 billion to $13.2 billion of 2026 revenue. The $12.8 billion midpoint is management's best middle estimate for the year. Good means beating that midpoint while keeping margins intact, not reaching it by signing expensive capacity that barely earns a return.
• Standard-accounting operating loss was $49 million, a negative 1.9% margin. Operating margin shows what remains after normal operating costs but before interest and taxes. Good infrastructure economics should eventually keep at least 10 cents from each sales dollar, so losing about 2 cents says scale has not reached the owners yet.
• Adjusted operating income was $128 million, a 5% margin. This version removes selected costs to show management's preferred operating picture. Good would be a margin above 10% that also appears on the standard-accounting statement, so today's adjusted profit is progress, not proof.
• Net loss was $626 million, a negative 24.3% margin. The bottom line lost roughly 24 cents for every dollar of sales after interest and taxes. Good would be a loss below 10 cents per sales dollar and shrinking, so the financing bill still eats the operating story alive.
• Net interest expense was $640 million, equal to 24.9% of revenue. Interest is the rent CoreWeave pays lenders for borrowed money. Good would be below 10% of revenue and falling, so almost one sales dollar in four going to lenders is a red alarm.
• Debt totaled about $35.1 billion against $5.524 billion of cash. Subtracting cash leaves roughly $29.5 billion of net debt, which is the burden the business must carry after using its available cash. Good for a capital-heavy growth company would be net debt below two times annual revenue. CoreWeave sits near 2.3 times this year's revenue forecast midpoint, before counting lease obligations.
• First-half operating cash flow was $3.663 billion, but equipment purchases reached $14.117 billion. Operating cash flow is cash generated by the business, while equipment purchases fund chips and data centers. The simple gap was negative $10.454 billion. Good means operations eventually fund expansion, so a gap this wide explains why CoreWeave keeps returning to debt and equity markets.
• Full-year capital-spending guidance is $35 billion to $39 billion. Capital spending buys long-lived assets such as servers and infrastructure. At the $37 billion midpoint, the company plans to spend almost three times its expected annual revenue. Good means each dollar of new equipment produces durable revenue and cash before the hardware ages.
• Active power reached 1.5 gigawatts and contracted power reached 3.7 gigawatts. Active power runs revenue-producing machines today, while contracted power covers capacity CoreWeave expects to bring online later. Good means the 2.2-gigawatt gap closes on time without costs outrunning contracts.
• The two largest customers supplied 62% of second-quarter revenue. That means two buyers controlled almost two-thirds of the sales line. Good diversification would push the pair below 40%, so the direction has improved but the dependency remains severe.
• Common shares rose from about 502 million at year-end to 551 million at June 30, up 9.8%. A rising share count makes every old share own less of the company. Good would keep annual dilution below 3%, so private placements and employee shares took a meaningful bite.
• At $80.925, the stock market values the common equity near $44.6 billion. Add debt and subtract cash, and the operating business costs about $74.2 billion, called enterprise value. That equals roughly 5.8 times this year's expected sales. Good value for a company growing this fast could sit below 5 times sales, but only if future margins and cash conversion become real.

What I like

The growth is not a PowerPoint trick. CoreWeave more than doubled quarterly sales and added nearly 500 megawatts of active power during the quarter. That means management brought a huge block of physical capacity online, not just another chatbot demo. Good execution turns powered buildings into billed usage quickly, and the third-quarter revenue forecast of $3.45 billion to $3.6 billion suggests that conversion has started.

The backlog gives the company rare visibility. $104 billion of contracted work can support years of expansion if customers honor commitments and CoreWeave delivers the machines. Good category leaders win before competitors finish debating whether the category exists. CoreWeave already has contracts with Microsoft (MSFT), Meta Platforms (META), and OpenAI (private), plus expanding enterprise demand.

The infrastructure has a focused design. CoreWeave built for dense AI workloads rather than adapting a general-purpose cloud. Good specialization should deliver faster deployment, stronger performance, and better technical support than a broad platform offers. Customers keep signing huge agreements, so the pitch has teeth.

Founder and chief executive Michael Intrator has shown genuine nerve. He helped pivot the business from crypto mining into AI infrastructure, then financed a buildout few operators could attempt. Good management sees the turn early and executes before everyone else. He did that. The next test is whether the same appetite for risk can produce owner cash instead of a larger mountain of obligations.

What worries me

The capital machine depends on outside money. CoreWeave generated $3.663 billion of operating cash in the first half and spent $14.117 billion on equipment. Good self-funding businesses reverse that relationship over time. CoreWeave still needs lenders and new shareholders to keep the construction line moving.

Interest already exceeds adjusted operating profit by a brutal margin. The quarter produced $128 million of adjusted operating income and $640 million of net interest expense. Good operating profit should cover interest at least three times. Here, adjusted operating profit covers only one-fifth of it. That is not a cushion. That is a floor mat.

The largest customers also own the option to become competitors. Microsoft (MSFT), Meta Platforms (META), Alphabet (GOOGL), and Amazon (AMZN) can build their own AI capacity with balance sheets CoreWeave cannot match. Good customer concentration falls as new industries arrive. If the giants internalize more work, today's backlog may not guarantee tomorrow's renewal economics.

Graphics chips age fast. The debt lasts longer than the glamour around any single chip generation. Good financing matches the useful life of the asset with the life of the customer contract and leaves room for replacement. If older chips lose rental value faster than lenders assumed, the collateral can shrink while the debt stays stubbornly adult-sized.

The accounting controls need work. The latest filing still identifies material weaknesses in financial reporting controls. A material weakness means the process may fail to prevent or catch a significant accounting mistake. Good public-company controls deserve a clean opinion and timely remediation, especially when the balance sheet contains this many financing structures.

What would make this a buy

This remains research, not a signal. I want a cheaper stock or a safer balance sheet, preferably both.

The price door opens around $60 to $65. At $60, the operating business would cost about 4.9 times the midpoint of 2026 sales guidance. At $65, it would cost about 5.1 times. Good valuation for a triple-digit grower can stretch to 5 times sales when backlog is durable, so this range gives some room for mistakes without pretending the debt disappeared.

The proof door opens if third-quarter revenue beats $3.6 billion, adjusted operating margin clears 7%, and net interest falls below 20% of revenue. Those numbers would mean capacity converts faster, core economics improve, and lenders take a smaller cut of each sales dollar. Good would then repeat for two quarters, because one clean print can wear rented clothes.

I also want the top two customers below 55% of revenue, year-end active power above 1.85 gigawatts, and no increase beyond the $39 billion top end of capital-spending guidance. That combination would show diversification, delivery, and budget discipline. Good growth earns more when every new contract does not require a fresh financing circus.

What makes me walk away

If revenue growth falls below 50% while capital spending stays above $35 billion, I walk. That would mean the buildout keeps sprinting while customer conversion slows. Good capital intensity falls as scale rises.

If net interest rises above 30% of revenue, the lenders own too much of the economics. Good interest burden moves toward 10%, not away from it.

If the top two customers climb back above 70% of revenue, diversification has failed. Good category leadership attracts new industries instead of deepening dependence on the same giants.

If the share count rises more than 15% in a year without a matching jump in per-share value, I walk. Good financing grows the pie faster than management cuts it into new slices.

If the material weaknesses survive the next annual filing, management loses trust points. Good finance teams fix known control gaps before complexity creates an expensive surprise.

IPO Watch

• Gravitics (private), expected September 3 at $14 to $17. The company designs large space structures, orbital carriers, cargo spacecraft, and station modules. I would write it up if the offering closes, Nasdaq trading begins under planned ticker GVTX, and the final filing shows enough cash to reach meaningful production without another near-term raise.
• Siyata PTT (private), expected September 8 with no underwritten price range. The highest recent private cash sale was $7, which gives context but does not guarantee where public trading starts. The company sells push-to-talk communications software and devices for workforces that need instant group coordination. I would write it up if Nasdaq approves the direct listing, recurring software revenue is real, customers are diversified, and the balance sheet does not require immediate dilution.

Earnings On Deck

• UiPath (PATH), September 3 after the close. Watch revenue above $400 million, annualized recurring revenue above $1.934 billion, customer retention above 110%, and a higher annual forecast. Good means the turnaround lasts beyond one quarter. Bad means the summer rally outran the customer contracts.
• Zscaler (ZS), September 3 after the close. Watch signed future business, large customer wins, and free cash flow. Good means contracted demand grows faster than reported sales while cash stays strong. Bad means competition slows deals or discounts buy the growth.

The verdict

Score: 67 out of 100, watch. CoreWeave earned nearly every available point for growth and many for category position, founder execution, backlog, customer wins, and near-term revenue catalysts. It lost heavily because official profit remains negative, interest consumes a quarter of sales, first-half equipment spending exceeded operating cash by $10.454 billion, customer concentration stays high, shares multiplied, controls remain weak, and the current valuation still expects elite execution.

The bucket is watch. Candidate status starts at 80. CoreWeave may become essential infrastructure for the AI economy, but common shareholders stand behind a financing stack built like a Manhattan skyscraper. I want proof that the elevators work before I pay penthouse rent.

The DBTech read

Capital intensity measures how much money a company must pour into physical assets to create revenue. CoreWeave spent $14.117 billion on equipment during six months while earning $4.653 billion of revenue, or roughly $3 of equipment for every $1 of sales. Good capital intensity falls as existing assets produce more revenue and operating cash funds a larger share of the next build. High growth can hide this burden for a while because every new contract looks exciting. The bill becomes obvious when growth slows and the machines still need replacing.

Sources and as-of date

Research completed September 3, 2026 at 7:01 a.m. Eastern.

The $80.925 closing price came from Alpaca's IEX broker feed at 3:59:53 p.m. Eastern on September 2, 2026. IEX is a narrower exchange feed than the full consolidated market, so small differences from Nasdaq's official close can occur. The $153.09 peak came from split-adjusted daily bars on the same IEX feed for June 5, 2026.

Second-quarter revenue, backlog, margins, cash flow, debt, cash, capital spending, customer concentration, share count, power capacity, guidance, control weaknesses, and risk disclosures came from CoreWeave's Form 10-Q filed with the U.S. Securities and Exchange Commission on August 12, 2026 and its earnings exhibit filed on August 11, 2026: https://www.sec.gov/Archives/edgar/data/1769628/000176962826000366/crwv-20260630.htm and https://www.sec.gov/Archives/edgar/data/1769628/000176962826000362/coreweave2q26earningspress.htm

The $40 IPO price, offering size, and Nasdaq listing details came from CoreWeave's final prospectus filed on March 28, 2025: https://www.sec.gov/Archives/edgar/data/1769628/000119312525067651/d899798d424b4.htm

Current company positioning and Michael Intrator's background came from CoreWeave's official website, retrieved September 3, 2026: https://www.coreweave.com/ and https://www.coreweave.com/leadership/mike-intrator

The IPO calendar and terms came from Renaissance Capital, Investing.com, and Siyata PTT's SEC registration statement, retrieved September 3, 2026. Gravitics still depends on closing conditions and Nasdaq approval, so its date can move. Siyata PTT proposes a direct listing rather than an underwritten IPO, so it has no offering range: https://www.renaissancecapital.com/IPO-Center/Calendar, https://www.investing.com/ipo-calendar/, and https://www.sec.gov/Archives/edgar/data/2110025/000121390026085987/ea0300166-f1_siyata.htm

The earnings calendar came from UiPath Investor Relations and Nasdaq, retrieved September 3, 2026: https://ir.uipath.com/news/detail/456/uipath-announces-second-quarter-fiscal-2027-financial-results-conference-call and https://www.nasdaq.com/market-activity/earnings

Community research covered August 4 through September 3, 2026 across Reddit, YouTube, Hacker News, and GitHub. It returned 8 Reddit threads, 6 YouTube videos, 1 Hacker News item, and 11 GitHub items. The GitHub matches were mostly unrelated name collisions, and the usable conversation centered on debt, backlog, power delivery, and customer concentration. I used community chatter to map the debate, not to prove financial facts.