Original Title: Neocloud Economics
Original Author: APP ECONOMY INSIGHTS

Editor's Note: Generative AI is driving a continuous increase in computing power demand, and the industry discussion is shifting from "Is GPU enough?" to "Who can, at a lower capital cost, transform electricity, chips, and data centers into usable computing power." As order growth and computing power shortages gradually become consensual, a more critical question emerges: for new cloud providers who must invest billions of dollars before revenue realization, are they building the next-generation AI infrastructure or preemptively overspending future demand with debt and capital expenditures?

Under App Economy Insights' "How They Make Money," CoreWeave, Nebius, and Cerebras' latest financial reports are dissected, breaking down three new cloud models:

CoreWeave relies on long-term contracts to lease NVIDIA GPUs, with a backlog of orders reaching $104 billion, but high capital expenditure and interest costs continue to drag down profits.

Nebius gains stronger pricing power through its self-built AI cloud platform, with customer prepayments and a shorter payback period improving capital efficiency.

Cerebras has shifted to cloud-based inference with its proprietary chip, where the cloud business has surpassed hardware, but order conversion is still constrained by capacity build-out.

These three companies face the same set of contradictions: demand far outstrips supply, but capacity must be built out in advance, resulting in simultaneous increases in capital expenditures, debt, and depreciation. A massive backlog of orders does not equate to revenue, let alone cash flow.

As tech giants like Meta expand their self-designed chips and GPU clusters, new cloud providers need to prove they are not just temporary fill-ins for the computing power gap. The next phase of competition will not only look at revenue growth rates but also focus on three key factors: capacity online speed, unit capital return, and financing costs.

Demand is already locked in, and capital efficiency will determine who ultimately prevails.

The following is the original text:

AI Has Created a Massive Computing Power Gap

The market's demand for computing power has exceeded the scale that large cloud computing companies can provide.

This gap has driven the rise of Neocloud (new cloud providers). They are specialized service providers built around AI infrastructure, focusing on acquiring power, dedicated data center capacity, and AI accelerator clusters, rather than replicating the vast software ecosystems of AWS, Azure, or Google Cloud.

This week, three different modes of new cloud providers announced their performances, allowing us to clearly see how they are addressing the computing power bottleneck through different strategies:

CoreWeave: Specializes in renting out NVIDIA GPU clusters and relies on long-term contracts with enterprise customers to scale up.

Nebius: Built from scratch on international data center infrastructure, creating an AI cloud platform.

Cerebras: With its own chip, is transitioning to provide cloud-based high-speed inference services.

The business logic of these three companies is somewhat counterintuitive. Computing facilities must invest capital and complete construction months in advance before generating revenue, so free cash flow is often negative. The importance of debt, leasing, depreciation, and customer concentration is almost as critical as revenue growth.

Let's take a look at the information disclosed this week.

CoreWeave's Recent Computing Power Nearly Sold Out

CoreWeave is the purest representative of the new cloud model. It purchases NVIDIA GPUs, deploys them in data centers, and rents out the computing power to customers such as OpenAI, Microsoft, and Meta.

Most of CoreWeave's computing power has already been booked by customers. Long-term committed contracts contributed 98% of second-quarter revenue, while on-demand usage only contributed 2%.

The company's revenue grew 112% year-over-year to $2.6 billion, but the gross margin decreased by 8 percentage points to 66%, as the growth rate of data center rent, power, and other expansion costs exceeded revenue.

CoreWeave recorded a $49 million operating loss and a $626 million net loss. Of these, $640 million in interest expenses directly related to GPU collateralized debt financing weighed heavily on profits.

And the income statement only reflects part of the expenses. CoreWeave's second-quarter capital expenditure reached $9.4 billion, more than three times the revenue for the quarter.

What Do These Numbers Mean?

Demand growth continues to outpace capacity expansion: The company's order backlog reached $104 billion, a 246% year-over-year increase; entering the early third quarter, the company also secured $25 billion in customer commitments. Recent capacity is effectively sold out.

Profit margins are expected to reach an inflection point: CoreWeave endured a gross margin decline to expedite new capacity online, but the adjusted operating profit margin has increased from 1% in the previous quarter to 5%. It is expected that the profit margin contribution of new contracts in the second quarter will be 5 to 10 percentage points higher than recent contracts, which does not yet consider the approximately 25% price increase in July.

Revenue Mix Improving: Annual recurring revenue from Storage, CPU, Networking, and Software businesses has surpassed $400 million. Annual recurring revenue from Hosted Inference business has jumped from $10 million to over $1 billion within a quarter.

Growth Still Highly Expensive: CoreWeave has raised its 2026 capital expenditure outlook to $35 billion to $39 billion, aiming to increase its green power capacity to over 1.85GW by the end of the year.

Key Takeaway: With a backlog of $104 billion compared to the company's targeted annualized revenue run rate of $19 billion by the end of 2026, it seems almost outrageously large. However, order conversion is still constrained by physical capacity. CoreWeave's investment thesis ultimately hinges on: whether it can convert power and GPUs into revenue fast enough while avoiding financing costs eroding margin improvement.

Nebius Gains Pricing Power

Nebius was not initially a typical AI infrastructure startup.

Spun off from Yandex: Nebius emerged in 2024 from the split of the Russian tech giant Yandex. Its Dutch holding company, which listed on Nasdaq, sold the Russian business for $5.4 billion and retained a smaller set of international operations, which later formed Nebius.

Fresh Start as a Public Company: The remaining entity retained its Nasdaq listing as Nebius Group and was steered by Yandex's co-founder Arkady Volozh.

Shift to AI Infrastructure: Instead of reconstructing the past internet conglomerate, Nebius leveraged existing engineering talent, cloud computing expertise, and capital to build a cloud platform tailored for AI.

Nebius rents out GPU computing power through its proprietary cloud platform. In the second quarter, AI Cloud business revenue reached $575 million, accounting for 98% of total revenue.

Company revenue grew 454% year-over-year to $582 million, with a gross margin increase of 6 percentage points to 77%; adjusted EBITDA reached $236 million, corresponding to a 41% profit margin.

Nebius recorded an operating loss of $176 million. With the addition of new billion-dollar infrastructure values to the P&L, depreciation and amortization alone amounted to $260 million.

What Do These Numbers Mean?

Hashrate Prices Are Rising: The average value of the four newly signed AI cloud contracts exceeds $1 billion, with a contract value per megawatt ranging from $20 million to $25 million. Prices for shorter-term hashrate contracts have reached as high as $40 million to $50 million per megawatt.

Investment Payback Accelerating: Management anticipates that contracts signed in the second quarter will require around 22 months to recoup the corresponding capital expenditure and operating costs, a significant reduction from the previous two to three years payback period.

Huge Expansion Scale: Nebius' second-quarter capital expenditure reached $5.7 billion, nearly 10 times the quarterly revenue. The company still expects full-year capital expenditures to reach $20 billion to $25 billion.

Customers are Helping with Financing: Nebius expects to receive over $9 billion in customer prepayments by 2026, covering around 50% to 60% of the associated capital expenditures.

Key Takeaway: Nebius is investing funds at an unprecedented scale, but the rising prices, shortened payback period, and customer prepayments are enhancing the economic value of each megawatt of added capacity. In the long run, this capital efficiency may be more critical than the 454% revenue growth rate.

Cerebras Shifting to the Cloud

Cerebras is a unique player among the new cloud vendors. Instead of purchasing NVIDIA GPUs, it designed its own wafer-scale processor and commercialized it in two ways: by selling systems and by renting out compute power through Cerebras Cloud.

The company's revenue structure is rapidly evolving. Second-quarter revenue grew 74% year-over-year to $180 million. Cloud and other service revenue increased by 281% to $126 million, while hardware revenue declined by 23% to $54 million.

The company reported an operating loss of as much as $477 million, but this number needs to be viewed in context. Cerebras went public in May, incurring significant stock-based compensation expenses. Excluding these factors, its core operating loss was only $34 million, far below the $477 million under US GAAP.

Core performance excludes stock-based compensation, customer warrant costs, and certain pass-through items. As the equity incentives granted at the IPO settle over time, this expense pressure is expected to normalize in the coming quarters.

Under the company's reporting standards, the gross margin is only 14%, but the core gross margin reaches 41%, representing an approximately 9 percentage point increase year-over-year, although slightly lower than the 46.5% reported in the first quarter. One reason is that, to meet the demand for cloud services, Cerebras temporarily needs to pay to rent back systems that were previously sold.

What Do These Data Mean?

Cloud services have become a growth engine: Core cloud business revenue has nearly tripled to $128 million, surpassing hardware revenue for the first time.

Performance expectations have improved: Cerebras has raised its FY 2026 core revenue guidance to $880 million to $890 million, while also increasing its gross margin and operating profit margin expectations.

Capacity remains a key bottleneck: By 2027, the company has commissioned or contracted data center capacity exceeding 600MW. The core gross margin is expected to bottom out in the third quarter, with a subsequent rebound in the margin as new capacity comes online and reliance on high-cost leased computing power diminishes.

Demand far exceeds current revenue: The company's remaining performance obligations amount to $25.4 billion. While OpenAI remains its primary customer, converting these backlog orders into revenue will require Cerebras to invest heavily in infrastructure.

Key Takeaway: Cerebras is transitioning from a chip vendor to a high-speed inference cloud service provider. Equity incentives and other accounting adjustments have masked this progress, but the real test is whether, as new capacity comes online, the company can convert the substantial backlog orders into revenue while repairing its profit margin.

What's Next?

Three new cloud providers are all facing the same dilemma: market demand has outstripped available computing power, yet to meet this demand, significant capital must be invested before revenue arrives.

A computing power shortage is also driving large tech companies to build their own capacity. Meta is expanding its in-house chip technology and a multi-gigawatt-scale GPU cluster, while SpaceX has begun selling rights to use its Colossus cluster externally.

The longer-term question is: once large tech companies' AI capacity is fully online, will new cloud providers continue to exist as indispensable infrastructure partners, or are they merely temporary expedients to fill the short-term computing power gap?

The next stage of competition will hinge on three factors: capacity, profit margin, and financing capability. New cloud providers need to convert contracted demand into fully operational infrastructure, improve investment returns as utilization increases, and also raise funds for the next round of expansion to avoid debt or equity dilution undermining the economic viability of the entire business model.

The demand has already been locked in, and capital efficiency will determine who comes out on top.

[Original Article Link]

Welcome to join the official BlockBeats community:

Telegram Subscription Group: https://t.me/theblockbeats

Telegram Discussion Group: https://t.me/BlockBeats_App

Official Twitter Account: https://twitter.com/BlockBeatsAsia