Wedbush: SpaceX's rental of computing power may indicate the construction of a new Colossus data center, and AI supply chain tensions may worsen.

After completing the IPO to load the history books, Elon Musk's SpaceX (SPCX.US) is extending its commercial tentacles from interstellar exploration to the most basic infrastructure of global artificial intelligence at an unimaginable speed. According to the latest information from investment banks and the supply chain, SpaceX has successfully rented out its huge computing power to three top technology companies. Wedbush Securities' latest research report pointed out that this series of aggressive computing monetization measures are likely a precursor to SpaceX secretly building another new "Colossus" giant data center. Three sky-high contracts: Who is paying for SpaceX's computing power? In less than a month before and after going public, SpaceX signed sky-high computing power leases with three AI giants: Anthropic pays $1.25 billion a month, Alphabet Inc. Class C pays $920 million a month, and Reflection AI pays $150 million a month. The annual income from these three contracts alone is approximately $26 billion, far exceeding SpaceX's total revenue of approximately $18.5 billion last year. Reflection AI: $6.3 billion "open source bet" On June 22, the open-source AI startup Reflection AI signed a computing power lease agreement with SpaceX. According to the contract obtained by CNBC, Reflection will pay SpaceX $150 million a month starting from July 1, 2026, to use the Nvidia GB300 flagship chip at the Colossus 2 data center. If the contract runs until the end of 2029, the total amount will reach $6.3 billion. Reflection is valued at $25 billion and is committed to building open-source models that can compete with cutting-edge models from OpenAI, Anthropic, and Alphabet Inc. Class C. A company spokesperson said, "Recent events have highlighted the importance of open source to the AI ecosystem. More and more countries and companies are realizing the risks and costs of relying solely on closed models." It is worth noting that Reflection has participated in the U.S. Energy Corp.'s "Genesis Mission" project and the broader AI program of the U.S. Department of Defense. Anthropic: $1.25 billion per month, leasing all of Colossus 1's capacity In May, Anthropic agreed to lease all of the computing power capacity of SpaceX's Colossus 1 data center in Memphis, Tennessee for $1.25 billion per month. The facility has over 220,000 NVIDIA Corporation GPUs (including the H100, H200, GB200, and other generations of chips), which can provide over 300 MW of additional computing power for Claude chat Siasun Robot & Automation. The contract runs until May 2029, with a total value of approximately $40 billion. Alphabet Inc. Class C: $920 million per month, locking in 110,000 GPUs On June 5, Alphabet Inc. Class C signed a cloud service agreement with SpaceX. According to SEC filings, Alphabet Inc. Class C will pay SpaceX $920 million per month from October 2026 to June 2029 to access around 110,000 NVIDIA Corporation GPUs and related computing resources. The agreement includes a clause that allows either party to terminate with 90 days notice. Alphabet Inc. Class C stated that this capacity is aimed at meeting the demand for its Gemini Enterprise AI product suite for large enterprises. The total monthly revenue from these three contracts is approximately $2.32 billion, with an annual revenue run rate of approximately $27.8 billionthis exceeds the annual revenue of many S&P 500 companies. The flip side of renting computing power: The "technical dilemma" of Colossus 1 However, behind SpaceX's large-scale rental of computing power is not purely strategic aggressionpart of it is forced. According to reports, SpaceX initially planned to train its cutting-edge AI models with a large cluster composed of three data center sites. However, there are serious network latency issues between Colossus 1 and the other two sites over 10 miles away, exacerbated by aging network infrastructure. The heterogeneous hardware issue is also significant. Colossus 1 internally deploys multiple generations of NVIDIA Corporation chips, including Hopper and Blackwell systems, as well as some older accelerators. In contrast, Colossus 2 and Colossus 3 have been deployed more uniformly around the NVIDIA Corporation Blackwell chip. In distributed training clusters, older chips can become bottlenecks, forcing faster accelerators to wait, leading to overall performance being limited to the slowest hardware. Faced with these technical limitations, SpaceX ultimately decided: rather than continue to invest resources in solving these problems, it is better to rent out the capacity of Colossus 1 to external customers. Critics point out that this is nothing more than a way to monetize the "mess" of AI due to confusing GPU architectures and stalled model training. This disclosure adds subtle pressure to SpaceX's narrativethe technical dilemma behind renting out Colossus 1, forcing outsiders to take a new look at the company's ability to rapidly expand its data center infrastructure. Wedbush warning: Big players are "full", a new round of computing power infrastructure construction may trigger a supply chain earthquake The exposure of this transaction has quickly led Wall Street to reevaluate Musk's computing power empire. Wedbush's top technology analyst Matt Bryson stated in a blockbuster report to clients that SpaceX's computing pool capacity far exceeds market expectations. Wedbush said: "According to calculations by Wccf, Musk's xAI actually consumes only about 11% of the floating point computing power (FLOPS) of Colossus 1, which means that SpaceX has successfully rented out most of the remaining capacity of Colossus 1. If these intensive rental transactions indicate that another large-scale Colossus data center construction is about to begin, we would not be surprised." However, this grand plan is colliding with the fragile ceiling of the global AI supply chain. Bryson warned that in the past, Musk's companies (such as Tesla, Inc., xAI, SpaceX) have caused severe fluctuations in component demand due to their aggressive procurement, even during periods of excess market supply. Currently, the global supply chain for high-performance chips is already under extreme pressure, even before xAI officially launches the next data center. Whether suppliers can meet SpaceX's next "appetite" is the biggest question in the market. However, this computing power tsunami has also brought epic wealth dividends to specific hardware suppliers. Market analysis indicates that core system suppliers who have been deeply involved in xAI and SpaceX data center construction in the past will be the biggest winners. From rockets to computing power infrastructure: the "second growth curve" of SpaceX after going public In the past, global investors' understanding of SpaceX was mainly focused on the Starlink satellite internet and Starship deep space exploration. However, this glamorous transformation into a "supercomputing contractor" has completely reshaped its valuation logic. These large data center leasing agreements, coupled with SpaceX's recent plans to acquire the popular AI coding platform Cursor, perfectly complement Musk's closed-loop ecosystem of "space network (Starlink) + computing center (Colossus) + AI terminal applications (Cursor/xAI)." Global energy and commodities analysts point out that SpaceX's series of actions indicate that the second half of the big model competition has evolved into a physical melee of "energy and infrastructure." With its strong power acquisition capabilities, land reserves, and synergies with the global Starlink network, SpaceX is becoming a "fourth major computing force" that global AI giants such as Microsoft Corporation Azure, Amazon.com, Inc. AWS, and Alphabet Inc. Class C cannot ignore. With more GB300 clusters delivered in the second half of the year, this multi-billion dollar giant's value in AI is reaching an all-time high.