The intense compute demand generated by generative AI is fundamentally reshaping the infrastructure landscape. The industry's focus has shifted from a simple shortage of GPU supply to determining which players can convert power, chips, and data centers into usable computing capacity at the lowest capital cost.

An analysis of the latest earnings reports from CoreWeave, Nebius (NBIS.US), and Cerebras reveals the central tension facing Neocloud providers as they build next-generation AI infrastructure. These companies are grappling with the conflict between debt-fueled expansion and the delayed realization of contracted revenue.

CoreWeave's model of borrowing against future demand

As the purest representation of the new cloud model, CoreWeave's business logic is built on purchasing Nvidia (NVDA.US) GPUs, deploying them in data centers, and then renting that capacity to major clients such as OpenAI, Microsoft (MSFT.US), and Meta (META.US). In the second quarter, long-term committed contracts contributed 98% of the company's revenue, with on-demand usage accounting for just 2%, while total revenue grew 112% year-over-year to $2.6 billion. However, rising data center rents, electricity costs, and other expansion expenses outpaced revenue growth, causing gross margins to fall 8 percentage points to 66%. The company recorded an operating loss of $49 million and a net loss of $626 million, with $640 million in interest expenses directly tied to GPU-backed debt financing representing a heavy burden.

More critically, the income statement fails to fully reflect the scale of its spending, as second-quarter capital expenditures reached a staggering $9.4 billion, more than triple its quarterly revenue. Despite demand vastly exceeding capacity, with backlog reaching $104 billion, up 246% year-over-year, and an additional $25 billion in customer commitments signed at the start of the third quarter, the inflection point in profit margins has not yet been fully established. Adjusted operating margins did improve from 1% in the prior quarter to 5%, and management expects contribution margins on newly signed second-quarter contracts to be 5 to 10 percentage points higher than recent agreements. Combined with roughly 25% price increases implemented in July, the revenue mix is improving. Annual recurring revenue from storage, CPU, networking, and software businesses has exceeded $400 million, while contracted annual recurring revenue from managed inference services jumped from $1 million to over $100 million in a single quarter.

Yet the cost of growth is immense. CoreWeave has raised its 2026 capital expenditure guidance to between $35 billion and $39 billion, targeting active power capacity above 1.85GW by year-end. Relative to its $19 billion target annualized revenue run-rate by the end of 2026, the $104 billion backlog appears substantial, but converting that backlog remains constrained by physical capacity. The investment thesis ultimately depends on how quickly the company can transform power and GPUs into revenue while preventing financing costs from eroding margin improvements.

Nebius takes a different route to capital efficiency

Nebius (NBIS.US) has followed a distinctly different path to growth. The company originated from the 2024 split of Russian tech giant Yandex. Its Nasdaq-listed Dutch holding company sold the Russian operations for $5.4 billion, and the retained international business subsequently formed Nebius (NBIS.US), with Yandex co-founder Arkady Volozh returning to lead the company. Rather than rebuilding an internet conglomerate, Nebius (NBIS.US) leveraged its existing engineering talent and cloud computing expertise to build a cloud platform purpose-built for AI. In the second quarter, AI Cloud revenue reached $575 million, representing 98% of total revenue, with overall revenue surging 454% year-over-year to $582 million. Gross margins improved 6 percentage points to 77%, and adjusted EBITDA reached $236 million, corresponding to a 41% margin. However, the company still reported an operating loss of $176 million, primarily due to $260 million in depreciation and amortization from new infrastructure hitting the income statement.

Notably, compute pricing is on the rise. The average value of four newly signed AI cloud contracts exceeded $1 billion each, with contract values reaching $20 million to $25 million per megawatt. Short-term contracts have commanded prices as high as $40 million to $50 million per megawatt. Management estimates that contracts signed in the second quarter will recover capital expenditures and operating costs in approximately 22 months, a significant improvement from the previous two-to-three-year cycle. Although second-quarter capital expenditures reached $5.7 billion, nearly ten times quarterly revenue, and full-year capital spending is projected at $20 billion to $25 billion, customer prepayments provide strong support. Nebius (NBIS.US) expects to receive more than $9 billion in customer prepayments in 2026, covering approximately 50% to 60% of related capital expenditures. This improvement in capital efficiency significantly enhances the economic value of each additional megawatt of capacity, making it strategically more significant in the long run than the 454% revenue growth rate.

Cerebras bets on custom silicon and a cloud pivot

Cerebras stands apart among neocloud providers, as it does not purchase Nvidia (NVDA.US) GPUs. Instead, the company designs its own wafer-scale processors and commercializes them through system sales and its Cerebras Cloud rental service. In the second quarter, revenue grew 74% year-over-year to $180 million, with cloud and other services revenue surging 281% to $126 million, while hardware revenue declined 23% to $54 million. The company recorded an operating loss of $477 million, but this was primarily driven by large stock-based compensation expenses related to its May IPO. Excluding these factors, core operating losses were just $34 million. Reported gross margins stood at 14%, but core gross margins reached 41%, up approximately 9 percentage points year-over-year, though down from 46.5% in the first quarter, partly due to paying to lease back previously sold systems to meet cloud service demand. Core cloud revenue nearly tripled to $128 million, surpassing hardware revenue for the first time.

Cerebras has raised its fiscal 2026 core revenue guidance to between $880 million and $890 million, while also improving gross margin and operating margin expectations. The company has over 600MW of data center capacity either operational or under contract through 2027, and expects core gross margins to recover after bottoming out in the third quarter. Despite remaining performance obligations reaching $25.4 billion and OpenAI remaining its primary customer, converting backlog into revenue still requires substantial infrastructure investment. The true test lies in whether the company can repair its margins while effectively converting its massive backlog as new capacity comes online.

A structural dilemma: spending before earning

These three neocloud providers collectively face a structural contradiction: market demand has outpaced available compute, but meeting that demand requires enormous capital investment before revenue arrives. This upfront capital spending means free cash flow is often negative, making debt, leases, depreciation, and customer concentration nearly as important as revenue growth. Massive backlogs do not equal revenue, let alone cash flow. Capacity must be built in advance, causing capital expenditures, debt, and depreciation to climb simultaneously, while revenue realization lags behind. This timing mismatch forces neocloud providers to constantly balance capital chain security against the pace of expansion.

Meanwhile, the trend of tech giants building their own capacity poses a long-term competitive threat. Meta (META.US) is expanding its self-developed chips and multi-GW-scale GPU clusters, and SpaceX has begun selling access to its Colossus cluster. These giants possess stronger financial resources and vertical integration capabilities. Once their AI capacity comes fully online, neocloud providers could find themselves relegated from indispensable infrastructure partners to temporary stopgaps for short-term compute gaps. This shifting competitive landscape demands that neocloud providers not only maintain technological leadership but also prove their long-term value through sustainable business models.

The next phase of competition will hinge on three core variables: capacity, margins, and financing capability. Neocloud providers must convert contracted demand into truly operational infrastructure, improving investment returns as utilization rates rise. Simultaneously, they must raise capital for the next round of expansion without allowing debt or equity dilution to undermine the economics of their business models. The speed of capacity conversion determines the pace of revenue realization, margin repair capability reflects operational efficiency, and financing capability concerns the survival floor. These three factors are mutually constraining and collectively determine the ultimate fate of neocloud providers in this fiercely competitive landscape.

Demand is locked in, but capital efficiency will decide the winners. In an environment where compute shortages are widely acknowledged, the decisive factor for neocloud providers is no longer merely technological sophistication or order book size, but rather how efficiently they can achieve capital turnover and profit conversion under enormous capital expenditure pressure. Those able to optimize capital structures, shorten payback periods, and effectively control financing costs will secure advantageous positions in this infrastructure race. Conversely, companies unable to resolve capital efficiency issues may find themselves trapped in a debt spiral despite holding billions in backlog. This is the crux of the new round of reshuffling in the AI infrastructure sector, following the era of traditional cloud computing giants' dominance.