Crux AI, a new cloud supported by Google, received a loan of 22 billion US dollars, and the “wave of new cloud power expansion” in the AI reasoning era is unstoppable

The Zhitong Finance App learned that US tech giant Google and US alternative investment leader Blackstone (Blackstone) have obtained bank loans of up to 22 billion US dollars, which can be described as an iconic example of this round of AI computing power demand expansion connecting bank financing channels. Another round of AI application boom brought about by the rise of Astra and even the AI industry brought about by the advent of Astra have never been discussed before. It can be said that it has injected new catalysts into the computing power expansion cycle. Strong AI computing power demand support linked to the AI computing power industry chain level has been clearly reflected in the strong performance of industry chain leaders, South Korea's continued record semiconductor exports, and long-term computing capacity expansion agreements surrounding new cloud forces.

According to recent media reports, Crux, which is supported by Google and Blackstone, received this financing to purchase a Google Tensor Processing Unit (TPU). The loan is supported by customer promises and purchased chip assets; the company plans to launch 500 megawatts, or 0.5 gigawatts of computing power capacity, in 2027 in the first phase.

Looking at the financing mechanism, customer promises to provide a basis for future revenue, and chip assets provide another layer of support for financing, so that demand can be transformed into actual production capacity along “customer promises - bank financing - chip procurement - infrastructure delivery”. The integrated layout of software scheduling, chip and network, data center and power infrastructure proposed by Crux also highlights its goal of delivering a complete AI computing power resource operation platform.

Crux AI, a new Google-backed cloud service provider, received a $22 billion bank loan to purchase AI chips

Crux AI, a recently established cloud computing company, has received a $22 billion bank loan. The company is a joint venture between Google and Blackstone Group.

According to media reports, people familiar with the matter revealed that these loans will be used to purchase tensor processing units exclusively built by Google (Google), that is, TPU, AI chip computing power hardware. These loans will be backed by Crux AI's customer commitments and the value of the chips purchased. Banks involved in providing loans include Goldman Sachs Group, Sumitomo Mitsui Bank, Barclays, BNP Paribas, and Scotiabank Canada.

Crux AI plans to provide the world's top artificial intelligence laboratories such as OpenAI and Anthropic with enormous cloud AI inference infrastructure resources. It will compete with other new cloud service providers such as CoreWeave (CRWV.US) and Nebius (NBIS.US).

The company was founded just last week and plans to put 500 megawatts of computing power into operation in the first phase in 2027 in accordance with its development roadmap.

“Artificial intelligence has brought about an intergenerational technological transformation, and with it, an opportunity to fundamentally reimagine the infrastructure that supports this technology,” said Benjamin Trainer Sloss, CEO of Crux AI. “This is exactly what we're building with Crux AI: an all-in-one platform designed from end to end and designed to run reliably at very large scales. Our goal is to be an infrastructure company that can be trusted and entrusted with the most critical workloads.” Before founding Crux AI, Sloss worked at Google for over 20 years.

Crux AI is a new cloud infrastructure company (Neocloud) that targets hyperscale AI workloads. Its main business is not a traditional general cloud, but rather integrates power+data center capacity+Google TPU accelerated computing (Accelerated Computing) + software and operation and maintenance into an end-to-end platform to provide large-scale dedicated computing power for AI laboratories, technology companies, and government customers such as OpenAI and Anthropic; the first phase plans to launch 500MW TPU in 2027 capacity, and expanded to multi-gigawatt scale.

The biggest difference from new clouds such as CoreWeave, which Nvidia invests in, is that Crux AI has been a “TPU native Neocloud” since its inception: Google directly provides TPU, software, and services, and Blackstone is responsible for capital and data center infrastructure, so it is more like an independent TPU computing power distribution channel outside of Google Cloud; CoreWeave is a typical GPU-native AI cloud, with the core centered around the NVIDIA AI GPU computing power cluster (NVIDIA) GPU clusters), high-performance networks, storage and scheduling software have been built, and a more mature multi-customer commercialization scale has been formed.

Coreweave's core barriers are GPU cluster operation efficiency and AI cloud software stack. Crux AI is trying to establish a vertical integration model integrating “TPU+power+data center+capital”; the former is closer to a mature AI computing power service provider, while the latter is more like a large-scale TPU infrastructure platform co-incubated by Google's technology stack and Blackstone infrastructure capital.

Giants lock in long orders and bank funds! The wave of new cloud expansion in the AI era is unstoppable

Global demand for AI computing power is exploding through “multi-year long orders+cross-platform procurement”. The new cloud platform (Neocloud) is taking on not only temporary spillover requirements, but part of the long-term infrastructure layout of tech giants.

In April of this year, Meta and CoreWeave signed an expansion agreement of about 21 billion US dollars. The supply of computing power continued until December 2032, focusing on supporting inference workloads; Anthropic then reached a multi-year agreement with CoreWeave to obtain additional computing power for Claude's development and deployment. The five-year agreement between Meta and Nebius announced in March this year reached a maximum of 27 billion US dollars, of which 12 billion US dollars correspond to dedicated capacity, and the remaining up to 15 billion US dollars correspond to procurement arrangements for the remaining usable capacity of specific clusters. External computing power procurement also extended to SpaceX: According to media reports on September 11, the relevant agreement signed by Google in June was worth about US$920 million per month, while Anthropic's agreement was US$1.25 billion per month, and will last until May 2029. Deriving from these procurement arrangements, even tech giants with huge self-construction capabilities are locking in production-level computing power in advance through external platforms; the competitive focus is shifting from “how many AI chips can be bought” to “how quickly they can obtain computing power clusters that are already powered, deployed, and can operate stably”. This is an important foundation for professional new cloud platforms to obtain long-term customer relationships.

The GPT-6 Astra model recently launched by OpenAI and the RSI technology path focused on by AI leaders are expected to become the two core driving forces driving the exponential expansion of AI computing power demand, namely the AI big model with better performance, the use of a wider range of AI application tools, and the next generation AI training path with stronger computing power requirements, which are an important basis for the continued growth in AI computing power infrastructure demand.

The investment significance brought by Astra is to improve the success rate and economic viability of complex tasks, so that companies are willing to deploy more agents and handle more professional tasks; Wall Street financial giant Morgan Stanley's recent emphasis on “shifting from demand debates to physical supply constraints on AI themes” is a new round of AI computing power resource demand expansion mechanism brought about by Astra, the most advanced model. The statement by the OpenAI product manager about unprecedented demand that the company may suspend new Pro subscriptions can be described as an important sign that AI computing power service capacity is under pressure recently.

The latest major signal on the AI computing power demand side undoubtedly comes from Astra: OpenAI product leaders described their demand as “unprecedented”; OpenAI also officially confirmed that new registrations and upgrades to the $200 per month Pro package will be suspended from September 10. The pressure on service demand brought about by the upgrading of model capabilities is no longer just an abstract market expectation; it has added inference demand catalysts from the actual user side to the wave of expansion that began over the years.

The core bullish logic of new cloud forces such as Crux AI and CoreWeave is to turn long-term customer promises, available power, and full-stack delivery capabilities into revenue growth rather than just holding more GPUs. CoreWeave's second-quarter revenue reached US$2,575 billion, an increase of about 112.5% over the same period of the previous year; by the end of June, the contract revenue reserve/backlog (Revenue Backlog) was approximately US$104 billion, which did not include net customer commitments of more than US$25 billion added at the beginning of the third quarter; the electricity capacity already in use reached 1.5 gigawatts. These customer promises provide a demand basis for subsequent production capacity investment, and related revenue can be gradually realized as delivery and service conditions are met.