As Muse and Astra usher in the era of "digital labor," Musk sets his sights on "AI compute supply"! SpaceX (SPCX.US) turns AI competitors into major customers.

Title context: As Muse and Astra usher in the era of "digital labor," Musk sets his sights on "AI compute supply"! SpaceX (SPCX.US) turns AI competitors into major customers. Text: The "AI ambitions" of SpaceX (SPCX.US), the "AI + space exploration" leader founded and helmed by Musk with a market value approaching $2 trillion, appear to be expanding comprehensively from competing for leadership in AI large models to selling or leasing its immensely vast AI compute resources to the global AI industry. The latest report published by The Information on October 1 reveals that, according to a series of sources familiar with the matter, the self-use AI supercomputing cluster originally serving Grok has evolved into a "Neocloud"-type AI compute leasing or selling business that takes on the needs of large AI large-model developers/large AI cloud computing customers such as Anthropic and Alphabet Inc. Class C. It is understood that SpaceX plans to bring approximately 420,000 NVIDIA Corporation GPUs online in November, expand the Memphis Minihard campus, and prepare capacity for a new agreement launching in December worth about $1.1 billion per month; in the summer it also discussed compute leasing with Microsoft Corporation, but the outcome of the negotiations has not yet been clarified. The model race continues to create demand, and SpaceX is trying to convert the highly efficient construction speed of AI compute infrastructure, its strong power resources built around AI systems, and its AI cluster operating capabilities into a powerful revenue curve. This is also why the market places greater emphasis on SpaceX's "AI compute valuation attribute," driving the company's stock price up nearly 40% since its early-August low, with a market value approaching $2 trillion. The participation of Alphabet Inc. Class C under Alphabet is especially a positive signal for the AI compute industry chain: according to disclosed agreements, it obtains approximately 110,000 NVIDIA Corporation GPUs and supporting massive compute resources including CPUs, DRAM memory/NAND data center storage systems, and has agreed to pay $920 million per month from October 2026 to June 2029. A tech giant with self-developed TPUs and massive cloud infrastructure also needs to purchase computing capacity externally, meaning that the compute competition is simultaneously a race to secure available capacity and time-to-online. The latest strong results and future outlook announced by Micron, one of the world's three largest memory chip original manufacturers and a U.S. memory chip super-giant, translate this unprecedented investment expansion around AI infrastructure into the income statement of compute suppliersMicron's fiscal 2026 fourth-quarter revenue was $54.229 billion, up about 379% year over year, with adjusted earnings per share of $33.42; next-quarter revenue guidance is $61.5 billion $1.5 billion. Quarterly data center SSD revenue was close to $10 billion, more than 10 times the same period last year, and the vast majority of 2027 HBM supply has already been contracted with prices significantly higher year over year. Management expects memory supply and demand in 2027-2028 to tighten further compared with the already tight record memory chip supply-demand situation, which is enough to show that AI compute demand is expanding the overall commercial value of the entire AI compute supply system, including the memory hierarchy. The key to SpaceX's transformation is converting previously underutilized resources into sellable services. The latest report from The Information points out that Colossus's utilization rate was once below 40% last year and was still low this spring; Musk initially opposed leasing capacity, but later accepted Anthropic's demand, including fully activating H100 resources that are relatively inefficient for its own Grok training. The Information said this large tech company also solved the secure multi-tenant isolation problem of the original system. Between chip ownership and cloud service revenue, there are engineering thresholds in scheduling, software, security isolation, and customer access; only by crossing these thresholds can locally idle capacity be converted into effective supply urgently needed by the industry. To accelerate delivery, the company stockpiled hundreds of millions of dollars worth of gas turbines in advance, brought in ABBSiasun Robot&Automation to organize cabling, and adopted modular facilities for parallel constructionthe so-called "lock in power in advance, compress the construction period, and fill in reliability"and is accelerating a 500MW data center with Saudi Arabia's Humain. At the same time, drainage and structural issues at Colossus II may cause at least a two-to-three-week delay, prompting SpaceX to mobilize engineers for rotating shifts and strengthen project management and reliability. Together, these details strongly show that what determines the commercial scale of AI compute resources is the complete delivery capability from power supply and construction to stable operations, and this is precisely the efficient AI compute resource integration advantage SpaceX possesses. What customers buy is computing service that can operate continuously, and delivery capability must cover the complete chain from power supply to cluster stability. Anthropic and Alphabet Inc. Class C were for a long time previously among the strongest AI large-model/AI agent competitors of xAI, founded by Musk (xAI has been acquired by SpaceX), and now both have become SpaceX's largest-scale compute resource customers, enabling SpaceX to continuously capture strong AI compute demand growth brought by multiple model ecosystems/agent systems in the great era of AI inference dominated by AI agents, improve AI compute cluster utilization, and broaden revenue sources. Behind AI agents beginning to participate in enterprise business processes: each instruction leverages an entire AI data center The financial disclosures in the latest IPO prospectus of Anthropic, a global leader in AI large models/AI applications, explain from the AI compute demand side why SpaceX is undergoing this transformation, and the most industry-significant aspect is undoubtedly the expansion of actual compute spending. A confidential IPO prospectus previously disclosed by media shows that Anthropic's 2025 revenue was close to $4.6 billion, operating loss was $8.06 billion, and compute and infrastructure spending was $7.33 billion, about three times the previous year and about 58% of total operating expenses of $12.65 billion. The data in this latest IPO prospectus shows that commercialization of frontier models is simultaneously expanding revenue and computing resource consumption, and AI agent/AI large-model training, inference, and infrastructure investment are still significantly ahead of profit realization. A longer-cycle signal comes from Anthropic's infrastructure commitments of at least $518 billion over roughly the next decade, of which about 80% are non-cancelable or require payment regardless of use; Anthropic's arrangement with SpaceX/xAI of up to $84.5 billion through 2029 mostly allows cancellation with 90 days' notice. It is worth noting that these are future contractual obligations and should be understood separately from current-period spending and recognized revenue. However, these latest signs together point to thisthe world's most frontier AI large-model labs are using long-term commitments to exchange for future capacity, and SpaceX is trying to convert construction speed into an increasingly strengthened supply position. Media, citing sources familiar with the matter, reported that Anthropic has signed an agreement to pay up to $84.5 billion to use AI compute resources through 2029 from SpaceX, the "AI + space exploration" leader founded and helmed by Musk. This also means that the compute fees Anthropic has agreed to pay SpaceX are close to twice previous estimates, highlighting that the faster the commercialization of AI large models/AI agents focused on agentic workflows progresses, the more strategically valuable compute clusters that can be delivered on time and operate stably become. Recent technological advances in the most frontier AI agents/AI large models represented by Muse, Astra, and Anthropic Claude can be said to provide an important technological foundation for the large-scale commercial expansion of AI applications into various industries and the continued surge in AI compute demandespecially as the scope of agent applications expands, it is expected to simultaneously increase demand for core AI infrastructure resources such as AI GPUs/TPUs and high-performance CPUs, data center high-performance HBM/DRAM/NAND memory chips, and high-speed optical interconnect components. From the perspective of AI inference system architecture, GPUs and specialized accelerators handle model computation, CPUs are responsible for converting inference results into actual operations, and memory and storage are responsible for saving and retrieving task states. Long context and multi-turn calls increase prefill computation and KV cache demand; browsers, code sandboxes, retrieval, and task orchestration increase server CPU load; files, databases, persistent memory, and cache tiers extend demand to server DRAM, enterprise SSDs, and high-speed networking. NVIDIA Corporation's technical materials have already described AI agent inference as an immensely large systems engineering effort spanning GPU HBM, CPU DRAM, local NVMe and remote storage, and internal high-speed optical interconnects critical for data transmission. In the view of Anthropic's management, the memory chip components of AI data center server clusters, as well as AI GPUs, remain the clearest supply bottlenecks at the AI compute industry chain level. Market research firm TrendForce estimates that in 2026 server DRAM contract prices will cumulatively rise about 270%, and enterprise SSD prices will cumulatively rise about 235%; in 2027 HBM contract prices may still rise 70%-140%, that is, continuing to show doubling growth. These data reflect the combined effect of continued expansion of AI compute demand and memory chip price increases. TrendForce calculations also show that in 2027 NVL72 rack shipments covering the Blackwell and Vera Rubin platforms are expected to grow more than 50% year over year; the accompanying chart in its market research shows that related system output value is expected to rise from about $226 billion in 2026 to $711 billion in 2027, a sharp year-over-year increase of 214%. The core change driven by Muse and Astra is making "digital labor" the new subject of compute consumption. Meta's Muse is powered by Muse Spark, has a dedicated cloud virtual machine, can continue executing tasks after the user leaves the app, and has an independent Sentinel agent review external operations; OpenAI's GPT-6 Astra provides capabilities for dots, with each agent having a cloud computer, browser, and tools, able to handle continuous work and delegate to sub-agents. A single user instruction thus unfolds into a continuous process of planning, retrieval, execution, checking, and re-reasoning. The overall AI compute demand mechanism behind this can be summarized astotal inference workload number of active users task frequency per person number of model calls per task computation per call. Increasing penetration affects the first term, while workflow automation and multi-agent collaboration expand the middle two terms. A previous research report by Anthropic disclosed that agents in its research sample consume about 4 times the tokens of ordinary chat, and multi-agent systems about 15 times; this empirical data cannot be directly applied to Muse or Astra, but it clearly demonstrates the resource change from "answering questions" to "efficiently completing complex human work." A deeper growth driver is that the cost per successful task declines, expanding the range of work AI can economically undertake. Astra improves both capability and token efficiency in some public evaluations, so demand growth does not require each task to become increasingly resource-intensivethat is, as long as the scale of new tasks exceeds the resource savings per unit task, total compute demand will continue to expand. In addition, the open-source and closed-source AI large-model competition trend jointly amplifies the space for AI compute resource demand by improving quality, lowering usage barriers, and expanding scenarios. The "delivery kill line" of the AI frenzy: how SpaceX opens up the imagination space for a $3 trillion valuation The accelerating adoption and penetration of AI agents around Muse and Astra are changing the resource structure of data centers. GPUs handle the main model computation, while CPUs run browsers, code sandboxes, tool calls, data processing, and task orchestration; long context and high concurrency expand active KV cache demand, driving HBM capacity and bandwidth upgrades; DDR supports virtual machines, tool processes, and some cache offloading, while enterprise SSDs carry files, long-term memory, and context-tiered storage. When cross-node prefill/decode separation is adopted, KV cache also needs to be moved through high-speed networks, expanding demand for network equipment and corresponding optical interconnects. Effective compute depends on whether the entire system can scale in coordination. This also explains why the investment mainline is spreading to CPUs, storage, and high-speed optical interconnects inside data centers. On September 21, demand expectations driven by the Muse download frenzy pushed AMD up about 10%, Intel Corporation about 12%, and Arm about 17%; South Korea's September data released on October 1 further provided verification on the memory chip manufacturing side (South Korea has two of the world's largest memory chip manufacturersSK Hynix and Samsung Electronics): semiconductor exports were about $60.3 billion, up a frenzied 262.8% year over year; computer exports were $7 billion, up 435.3% year over year; semiconductor equipment imports were $3.44 billion, up 50.8% year over year. Export value is driven by both volume and price, while the growth in equipment imports shows that compute demand is being fully transmitted to semiconductor capacity expansion. Based on this trend, some Wall Street analysts have put forward a judgment running through the industry chain: the unprecedented global corporate frenzy to deploy AI is forming an "AI compute delivery kill line"the upper limit of commercial scale increasingly depends on how many successful tasks an enterprise can continuously deliver at what cost, latency, and reliability. SpaceX's advance power locking, cluster expansion, customer acquisition, and reliability strengthening are precisely competing for supply position on this boundary. For the capital market, the indicators with a basis for revaluation have also become concreteavailable power, capacity brought online on schedule, paid utilization, effective task throughput per megawatt, and ultimately the cash flow formed. It is understood that top Wall Street firm Financial Institutions, Inc.TD Cowen initiated coverage of SpaceX in late September with a "Buy" rating and a $200 target price, implying about 35.1% upside from the October 1 closing price of $148.07; the core logic is that terrestrial AI compute leasing will become the main near-term growth engine, with related revenue expected to reach $66 billion in 2027, about 58% of total revenue, and to contribute more than half of revenue starting in the first quarter of that year. In terms of target price coverage by other Wall Street financial giants, Goldman Sachs Group, Inc. gave a "Buy" rating and a 12-month target price of $220, Morgan Stanley gave "Overweight" and a target price as high as $300, UBS Group AG gave "Buy" and a $210 target price, and Bank of America gave a "Buy" rating and a $235 target price. According to a sample of 37 Wall Street analysts aggregated by S&P Global as of September 30, the consensus rating is "Buy," the most optimistic and positive rating, with an average 12-month target price of $226.04, implying about 52.7% potential upside relative to $148.07; based on a static calculation assuming the current total share count of about 13.57 billion shares remains unchanged, the implied market value is about $3.07 trillion, higher than the current about $2.01 trillion. The core expectation analysts attach to this optimistic valuation scenario is precisely that SpaceX, under Musk's leadership, is increasingly focused on converting power, construction speed, and cluster operating capabilities into sustained revenue serving multiple AI customers.