Google and Blackstone-backed Crux AI has secured a massive $22 billion in bank loans, marking a landmark moment where the surging demand for AI computing power has tapped directly into the banking sector's financing channels. The emergence of Astra, along with the unprecedented AGI discussions it has sparked across the AI industry, has injected fresh momentum into the computing capacity expansion cycle. The robust AI computing demand tied to the infrastructure supply chain is already clearly reflected in the strong earnings of leading companies, South Korea's record-breaking semiconductor exports, and long-term capacity expansion agreements centered around new cloud players.

According to the latest media reports, Crux, supported by Google and Blackstone, secured this financing to purchase Google's Tensor Processing Units (TPUs), with the loans backed by customer commitments and the acquired chip assets. The company's initial phase plans to bring 500 megawatts, or 0.5 gigawatts, of computing capacity online by 2027. From a financing mechanism perspective, customer commitments provide the basis for future revenue, while chip assets offer another layer of support for the financing, enabling demand to convert into actual capacity along the path of "customer commitments — bank financing — chip procurement — infrastructure delivery." Crux's integrated approach to software orchestration, chips and networking, data centers, and power infrastructure also highlights its goal of delivering a complete AI computing resource operating platform.

Google-backed new cloud service provider Crux AI has obtained $22 billion in bank loans to purchase AI chips. The newly established cloud computing company Crux AI has secured $22 billion in bank loans, a joint venture founded by Google and Blackstone Group. Media reports citing informed sources indicate these loans will be used to purchase Google's proprietary Tensor Processing Units, or TPUs, the AI chip computing hardware. The loans will be backed by Crux AI's customer commitments and the value of the purchased chips. Banks involved in providing the loans include Goldman Sachs, Sumitomo Mitsui Banking Corporation, Barclays, BNP Paribas, and Scotiabank.

Crux AI plans to provide massive cloud-based AI inference computing infrastructure resources for the world's top AI laboratories such as OpenAI and Anthropic. It will compete with other new cloud service providers like CoreWeave and Nebius. The company was founded just last week and plans to bring 500 megawatts of computing capacity online in 2027 in its first phase, according to its development roadmap.

"AI has brought about a generational technological shift, and with it comes the opportunity to fundamentally reimagine the infrastructure that underpins this technology," said Benjamin Treynor Sloss, CEO of Crux AI. "This is exactly what we are building at Crux AI: an integrated platform designed end-to-end to operate reliably at hyperscale. Our goal is to become an infrastructure company trusted with the most critical workloads." Before founding Crux AI, Sloss spent more than 20 years at Google.

Crux AI is a new cloud infrastructure company (Neocloud) specifically tailored for hyperscale AI workloads. Its core business is not traditional general-purpose cloud computing, but rather integrating 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 computing power for AI laboratories like OpenAI and Anthropic, tech enterprises, and government clients. The initial phase plans to launch 500MW of TPU capacity by 2027, expanding toward multi-gigawatt scale.

The biggest difference from GPU-native new clouds like CoreWeave, backed by Nvidia, is that Crux AI is a "TPU-native Neocloud" from inception: Google directly provides TPUs, software, and services, while Blackstone handles capital and data center infrastructure. It therefore functions more like an independent TPU computing distribution channel outside of Google Cloud. CoreWeave, by contrast, is a typical GPU-native AI cloud, built around Nvidia AI GPU clusters, high-performance networking, storage, and orchestration software, and has already achieved more mature multi-customer commercial scale. CoreWeave's core moat lies in GPU cluster operational efficiency and its AI cloud software stack, while Crux AI seeks to establish a vertically integrated "TPU + power + data center + capital" model. The former is closer to a mature AI computing service provider, while the latter resembles a hyperscale TPU infrastructure platform incubated jointly by Google's technology stack and Blackstone's infrastructure capital.

Tech Giants Lock in Long-Term Contracts as Banks Supply Capital: The New Cloud Expansion Wave Is Unstoppable

Global AI computing demand is expanding explosively through "multi-year long-term contracts + cross-platform procurement." Neocloud platforms are no longer just absorbing temporary overflow demand but are part of tech giants' long-term infrastructure strategies. In April this year, Meta signed an approximately $21 billion expansion agreement with CoreWeave, extending computing supply through December 2032, with a focus on supporting inference workloads. Anthropic subsequently reached a multi-year agreement with CoreWeave to secure additional computing power for the development and deployment of Claude.

Meta's five-year agreement with Nebius, announced in March this year, could reach up to $27 billion, with $12 billion corresponding to dedicated capacity and the remaining up to $15 billion for procurement arrangements of available capacity on specific clusters. External computing procurement also extends to SpaceX. Media reported on September 11 that Google's agreement signed in June is valued at approximately $920 million per month, while Anthropic's agreement is $1.25 billion per month, lasting through May 2029.

Extrapolating from these procurement arrangements, even tech giants with massive self-built capabilities are locking in production-grade computing through external platforms in advance. The competitive focus is shifting from "how many AI chips can be purchased" to "how quickly can you access computing clusters that are already powered, deployed, and stable in operation" — which is precisely the fundamental basis for specialized new cloud platforms to secure long-term customer relationships.

OpenAI's recently launched GPT-6 Astra large model, along with the RSI technical pathways that AI leaders are focusing on, are expected to be two core drivers of exponential AI computing demand growth. More powerful AI models, broader adoption of AI application tools, and next-generation AI training pathways requiring even more computing power are strengthening the case for continued growth in AI infrastructure demand. The investment significance of Astra lies in improving the success rate and economic feasibility of complex tasks, prompting enterprises to deploy more agents and handle more specialized tasks. Morgan Stanley's recent emphasis on "shifting from demand debates back to the physical supply constraints of the AI theme" reflects the new round of AI computing resource demand expansion triggered by the cutting-edge Astra model.

OpenAI's product head's statement that demand is so unprecedented that the company may pause new Pro subscriptions is a significant signal of current AI computing service capacity pressure. The latest major signal from the AI demand side undoubtedly comes from Astra: OpenAI's product head described its demand as "unprecedented," and OpenAI officially confirmed that starting September 10, new registrations and upgrades for the $200-per-month Pro plan would be suspended. The service demand pressure brought by model capability upgrades is no longer just an abstract market expectation; it has added real-world inference demand catalysts from actual usage to the multi-year expansion wave already underway.

The core bullish thesis for new cloud players like Crux AI and CoreWeave is converting long-term customer commitments, available power, and full-stack delivery capabilities into revenue growth, rather than just holding more GPUs. CoreWeave's second-quarter revenue reached $2.575 billion, up approximately 112.5% from $1.212 billion in the same period last year. As of the end of June, contracted revenue backlog stood at approximately $104 billion, excluding over $25 billion in net new customer commitments added in early Q3. Power capacity already in use reached 1.5 gigawatts. These customer commitments provide the demand foundation for future capacity deployment, with related revenue gradually recognized as delivery and service conditions are met.