Crux AI, a new cloud service provider backed by Google, has secured a $22 billion bank loan to fund its chip procurement. The core bullish rationale behind emerging cloud players like Crux AI and CoreWeave is to convert long-term customer commitments, available power capacity, and full-stack delivery capabilities into revenue growth—rather than simply holding more GPUs.
Zhitong Finance APP has learned that U.S. tech giant Google and leading alternative investment firm Blackstone have secured bank loans totaling up to $22 billion, marking a landmark case of banks opening their financing channels amid the current surge in AI computing demand. The emergence of Astra has sparked another wave of AI application enthusiasm, while the very launch of Astra has ignited unprecedented industry-wide discussions on AGI, injecting fresh momentum into the cycle of computing capacity expansion. Strong demand for AI computing power—linked to the broader AI computing industry chain—is already clearly reflected in the robust financial performance of industry leaders, South Korea's record-breaking semiconductor exports, and long-term agreements to expand computing capacity among emerging cloud players.
According to the latest media reports, Google and Crux, backed by Blackstone, have secured this round of funding to purchase Google Tensor Processing Units (TPUs), with the loan collateralized by customer commitments and the acquired chip assets; the company's initial phase plans to bring 500 megawatts—equivalent to 0.5 gigawatts—of computing capacity online by 2027.
From the perspective of the financing mechanism, customer commitments serve as a basis for future revenue, while chip assets provide an additional layer of collateral, enabling demand to translate into actual production capacity along the sequence: "customer commitment—bank financing—chip procurement—infrastructure delivery." Crux's proposed integrated deployment of software scheduling, chips and networking, data centers, and power infrastructure further underscores its goal of delivering a comprehensive AI computing‑resource operating platform.
Crux AI, a new cloud service provider backed by Google, has secured a $22 billion bank loan to procure AI chips.
Recently established cloud computing company Crux AI has secured a $22 billion bank loan; the company is a joint venture founded through a joint investment by Google and Blackstone.
According to media reports, sources revealed that these loans will be used to purchase Google's proprietary Tensor Processing Unit, or TPU, an AI chip-based computing hardware. The loans will be backed by Crux AI's customer commitments and the value of the chips purchased. Banks participating in providing these loans include Goldman Sachs, Sumitomo Mitsui Banking Corporation, Barclays, BNP Paribas, and the Bank of Nova Scotia.
Crux AI plans to provide unparalleled, massive cloud-based AI inference computing infrastructure resources to some of the world's top AI labs, including OpenAI and Anthropic. It will compete with other emerging cloud service providers such as CoreWeave (CRWV.US) and Nebius (NBIS.US).
The company was established just last week and, in accordance with its development roadmap, plans to bring 500 megawatts of computing capacity online in the first phase by 2027.
"Artificial intelligence has ushered in a generational technological transformation, and with it comes the opportunity to fundamentally reimagine the infrastructure that underpins this technology," said Benjamin Triner Sloss, CEO of Crux AI. "That's exactly what we're building at Crux AI: an end-to-end‑designed, integrated platform engineered to deliver reliable performance at hyperscale. Our goal is to become an infrastructure company that can be trusted to handle the most mission‑critical workloads." Prior to founding Crux AI, Sloss spent more than 20 years at Google.
Crux AI, a new cloud‑infrastructure company (Neocloud) specializing in ultra‑large‑scale AI workloads, does not offer traditional general‑purpose cloud services. Instead, it integrates power, data‑center capacity, high‑speed networking, Google TPU‑accelerated computing, and software and operations into an end‑to‑end platform, providing large‑scale, dedicated compute resources to AI labs such as OpenAI and Anthropic, technology firms, and government clients. The initial phase plans to bring 500 MW of TPU capacity online by 2027, with expansion toward multi‑gigawatt scale.
The biggest difference between Crux AI and new cloud providers like CoreWeave, which is backed by NVIDIA, is that Crux AI was conceived from the outset as a "TPU-native Neocloud": Google directly provides TPUs, software, and services, while Blackstone handles capital and data-center infrastructure. As a result, it functions more like an independent TPU compute‑distribution channel outside of Google Cloud. By contrast, CoreWeave is a quintessential GPU-native AI cloud, built around NVIDIA's AI GPU compute clusters, high‑performance networking, storage, and scheduling software, and has already achieved a more mature, multi‑customer commercial scale.
CoreWeave's core competitive advantages lie in the operational efficiency of its GPU clusters and its AI cloud software stack, while Crux AI seeks to establish an integrated vertical model encompassing "TPU + power + data centers + capital." The former is closer to a mature AI computing power service provider, whereas the latter resembles a hyperscale TPU infrastructure platform co-incubated by Google's technology stack and Blackstone's infrastructure capital.
Industry giants lock in long-term contracts, and banks provide financing! The new wave of cloud expansion in the AI era is unstoppable.
Global demand for AI computing power is experiencing explosive growth, driven by "multi-year long-term contracts plus cross-platform procurement." The emerging Neocloud platform now handles not just temporary overflow workloads, but has become an integral part of tech giants' long-term infrastructure strategies.
In April this year, Meta signed an expanded agreement with CoreWeave worth approximately $21 billion, extending computing power supply through December 2032 and prioritizing inference workloads. Subsequently, Anthropic also reached a multi-year deal with CoreWeave to secure additional compute capacity for the development and deployment of Claude. Meanwhile, Meta and Nebius announced a five-year agreement in March, valued as high as $27 billion—$12 billion for dedicated capacity and up to $15 billion for purchasing remaining available capacity in specific clusters. External compute procurement has also extended to SpaceX: according to media reports on September 11, Google's related agreement signed in June is valued at roughly $920 million per month, while Anthropic's deal amounts to $1.25 billion per month and runs until May 2029. Judging from these procurement arrangements, even tech giants with substantial in-house capabilities are proactively locking in production‑grade compute through external platforms; the competitive focus is shifting from "how many AI chips can we buy" to "how quickly can we access compute clusters that are already powered, deployed, and capable of stable operation"—a key foundation for professional new cloud platforms to build long‑term customer relationships.
OpenAI's recently unveiled GPT‑6 Astra large model, along with the RSI technology roadmap that leading AI players are focusing on, is poised to become two of the key drivers behind the exponential expansion of AI computing demand. More powerful AI large models, broader adoption of AI application tools, and next‑generation AI training pipelines that demand even greater computational power are all reinforcing the robust, sustained growth in demand for AI‑specific computing infrastructure.
The investment implications of Astra lie in enhancing the success rate and economic feasibility of complex tasks, encouraging enterprises to deploy more intelligent agents and tackle a wider range of specialized workloads. Wall Street financial giant Morgan Stanley has recently underscored the shift "from debating demand back to the physical supply constraints underlying the AI agenda," which precisely reflects the new round of expanded demand for AI computing resources driven by cutting-edge large models like Astra. Meanwhile, the OpenAI product lead's remark that demand has reached such unprecedented levels that the company may temporarily suspend onboarding new Pro subscribers serves as a significant signal of mounting pressure on AI computing service capacity.
The latest major signal on the AI computing‑power demand side undoubtedly comes from Astra: OpenAI's head of product described the demand as "unprecedented." OpenAI has also officially confirmed that, effective September 10, it will suspend new sign-ups and upgrades for its $200‑per‑month Pro plan. The pressure on service demand driven by model‑performance upgrades is no longer just an abstract market expectation; it has added a catalyst in the form of real‑world inference workloads to the multiyear capacity‑expansion wave that was already underway.
The core bullish thesis of new cloud players like Crux AI and CoreWeave is to convert long-term customer commitments, available power capacity, and full-stack delivery capabilities into revenue growth, rather than simply holding more GPUs. In the second quarter, CoreWeave's revenue reached $2.575 billion, up roughly 112.5% from $1.212 billion in the same period last year. As of the end of June, its revenue backlog stood at approximately $104 billion, not yet including over $25 billion in net customer commitments added at the start of the third quarter; meanwhile, its deployed power capacity had reached 1.5 gigawatts. These customer commitments provide a solid demand base for future capacity expansion, with related revenue expected to be recognized gradually as delivery and service milestones are met.