Compal, Verda in $2.1B AI Cloud Deal; Anthropic Bets $200B on Google

Compal Electronics will supply next-generation GPU server systems to Verda, a European AI cloud provider headquartered in Helsinki, Finland, in a partnership that accelerates AI infrastructure build-out across Europe and Asia-Pacific. The deal, valued at up to $2.1 billion, marks one of the largest hardware procurement agreements between an Asian manufacturer and a European cloud operator, and it signals a strategic pivot for Compal as it expands manufacturing capacity in Taiwan, Vietnam, and the United States. Separately, Anthropic committed to spend $200 billion on Google's AI cloud services and TPU chips, while also signing a compute deal to use resources from SpaceXAI's Colossus 1 supercomputer. This move places the startup at the center of a multi-cloud, multi-hardware strategy. The compute crunch and data center delays that have plagued the industry are now forcing every major AI player to lock in capacity years in advance, and these two transactions show how the market is fragmenting into distinct hardware and geographic camps. Why this matters now: the race for AI compute is no longer just about who builds the best model. It is about who controls the physical infrastructure, the chip supply chains, and the strategic alliances that will determine which companies survive the next wave of consolidation.

Where the $2.1B Flows: Compal’s Manufacturing Expansion

The Compal-Verda agreement is structured as a multi-year supply contract for next-generation GPU server systems, with Compal committing to ramp production across three geographies. The company is expanding its existing facilities in Taiwan and Vietnam, and it is building new capacity in the United States. This move aligns with Washington’s push to onshore critical AI hardware manufacturing. Verda, which operates AI cloud services primarily in Europe, will use the servers to build out data centers in the Nordics, Germany, and Singapore, targeting enterprise customers that require low-latency inference and training workloads. The $2.1 billion figure includes both hardware and integration services, and it represents roughly 15% of Compal’s annual revenue, according to analysts who track the contract manufacturer. The deal also includes a provision for joint R&D on next-generation server architectures, which gives Compal an edge in designing systems optimized for Nvidia’s upcoming GPU generations. For Verda, the partnership solves a critical bottleneck: securing enough GPU server supply to compete with larger cloud providers like Google Cloud and Microsoft Azure, which have been hoarding Nvidia’s H100 and B200 chips. Verda’s CEO Alan Chang has publicly stated that the company aims to double its compute capacity within 18 months, and the Compal deal provides the manufacturing backbone to achieve that target. Compal’s expansion in the United States will create new jobs in Arizona and Texas, where the company is building assembly lines for high-density server racks. The Vietnam facility, located near Ho Chi Minh City, is expected to begin production by early 2027, serving customers across Southeast Asia.

Anthropic’s $200 Billion Google Commitment Locks In TPU Supply

Anthropic’s decision to commit $200 billion to Google’s AI cloud services and TPU chips is the largest single cloud spending pledge in the industry’s history, surpassing even Microsoft’s multi-year commitments to OpenAI. The deal guarantees Anthropic access to Google’s latest TPU v6 and v7 chips, which are purpose-built for large-scale training and inference workloads. For Google, the commitment transforms its cloud business into a dedicated AI foundry, with Anthropic effectively becoming a tenant on Google’s internal TPU cluster. The $200 billion figure is spread over a 10-year period, with annual payments starting at $15 billion and escalating as Anthropic’s compute needs grow. This structure gives Google predictable revenue growth while locking Anthropic into Google’s hardware ecosystem. This is a strategic win for Google, which has been losing ground to Microsoft in the enterprise AI cloud market. Anthropic’s CEO Jorge Santos has framed the deal as a hedge against GPU shortages, arguing that TPU chips offer better price-performance for certain workloads, particularly those involving large language models with long context windows. The commitment also includes a clause that allows Anthropic to sublease unused compute capacity, giving it flexibility to manage demand fluctuations. Industry observers note that the deal effectively makes Anthropic Google’s anchor AI tenant, similar to how OpenAI anchors Microsoft Azure.

Competitive Reshuffle: SpaceXAI Gains a New Tenant

Anthropic’s separate deal to use computing resources from SpaceXAI’s Colossus 1 supercomputer reshapes the competitive landscape in unexpected ways. SpaceXAI, formed earlier this year through the merger of SpaceX and xAI, is seeking to go public as soon as next month, and the Anthropic deal provides a marquee customer that validates its supercomputing capabilities. Colossus 1, which was originally built by xAI to train its own models, has excess capacity that SpaceXAI is now monetizing through third-party compute contracts. The deal gives Anthropic access to a different hardware architecture, one based on Nvidia’s H100 and B200 GPUs, complementing its TPU-based workloads on Google. Colossus 1’s cluster currently spans over 200,000 GPUs, making it one of the largest single AI compute installations in the world and an attractive resource for a model lab running long training runs. This multi-hardware strategy reduces Anthropic’s dependence on any single chip vendor and allows it to optimize training costs across different model architectures. For SpaceXAI, the deal signals that its supercomputer is not just a tool for internal AI development but a commercial asset that can generate revenue ahead of its IPO. The partnership also creates an awkward dynamic for xAI, which competes with Anthropic in the large language model market. Both companies are now sharing the same compute infrastructure, raising questions about data isolation and competitive intelligence. SpaceXAI has stated that it will use hardware-level isolation to prevent cross-tenant data leakage, but the arrangement remains unprecedented in the AI industry.

Downstream Effects: Hyperscalers, Fabs, and Enterprise Buyers

The Compal-Verda and Anthropic-Google deals have immediate downstream consequences for hyperscalers, chip fabs, and enterprise buyers. For Nvidia, the deals represent a mixed signal: while Compal’s servers will use Nvidia GPUs, Anthropic’s $200 billion commitment to Google’s TPU chips reduces the startup’s reliance on Nvidia hardware. This diversification pressures Nvidia to maintain its pricing power and supply allocation, especially as Google and Microsoft push their own custom chips. For Taiwan Semiconductor Manufacturing Co., which fabricates both Nvidia GPUs and Google TPUs, the deals guarantee sustained demand for advanced packaging and high-bandwidth memory capacity. Enterprise buyers, particularly in Europe and Asia-Pacific, will benefit from Verda’s expanded cloud footprint, which offers an alternative to US-based hyperscalers. The deal also accelerates Compal’s US manufacturing expansion, which reduces supply chain risks for American customers. However, the compute crunch and data center delays that have plagued the industry remain unresolved. Verda’s expansion timeline depends on securing permits and power for new data centers, a process that has slowed projects across Europe. Similarly, Anthropic’s $200 billion commitment assumes that Google can deliver the required TPU capacity on schedule, which is not guaranteed given ongoing supply chain constraints. The US government is also stress-testing AI models from Google, xAI, and Microsoft, adding regulatory uncertainty to the infrastructure build-out.

Policy and Strategy Signal: The Fragmentation of AI Compute

The two deals signal a broader strategic shift in the AI industry: compute is fragmenting along hardware, geographic, and political lines. Anthropic’s dual commitment to Google TPUs and SpaceXAI GPUs reflects a deliberate strategy to avoid vendor lock-in, while Verda’s partnership with Compal shows that European cloud providers are willing to bypass US hyperscalers to build their own infrastructure. This fragmentation creates opportunities for contract manufacturers like Compal, which can serve multiple cloud providers without being tied to a single platform. It also pressures regulators to define rules for cross-border data flows and chip exports, especially as Compal expands manufacturing in the US and Vietnam. The SpaceXAI IPO, expected next month, will test investor appetite for pure-play AI infrastructure companies, and the Anthropic deal provides a strong revenue anchor for its prospectus. For Google, the $200 billion commitment locks in a major customer at a time when its cloud business faces increasing competition from Microsoft and Amazon. For Anthropic, the deals provide compute diversity and long-term capacity, but they also create financial obligations that will constrain its ability to pivot to new hardware architectures. The industry is entering a phase where infrastructure commitments of $2 billion and $200 billion are becoming routine, and the companies that can secure the most favorable terms in pricing, geography, and hardware diversity will have a structural advantage in the next generation of AI development.

The next 12 months will test whether these infrastructure bets pay off. Compal must execute its manufacturing expansion across three continents while managing rising labor and component costs. Verda must demonstrate that it can compete with hyperscalers on latency and reliability, particularly in the European market where data sovereignty regulations give it an edge. Anthropic faces the challenge of managing two massive compute contracts simultaneously, optimizing workloads across TPUs and GPUs while keeping training costs under control. SpaceXAI’s IPO will set a valuation benchmark for AI infrastructure assets, and the deal with Anthropic provides a tangible revenue stream that analysts can model. The broader lesson from these transactions is that the AI industry is moving beyond the model-building phase into an infrastructure arms race, where capital allocation and supply chain management matter as much as algorithmic innovation. Companies that can secure diverse compute resources, build geographically distributed data centers, and navigate regulatory scrutiny will emerge as the dominant players in the next cycle. The compute crunch is not easing. It is deepening, and these deals are the first major moves in a game that will define the industry for the next decade.