Anthropic raises Claude Code usage limits, credits new deal with SpaceX

Anthropic struck a deal to access the full compute capacity of SpaceX’s Memphis data center, adding more than 300 megawatts of new compute capacity. The company also doubled Claude Code five-hour window limits for Pro and Max subscribers, removed peak-hour reductions for those accounts, and raised API limits for its Opus model. SpaceX said Colossus 1 includes over 220,000 NVIDIA GPUs, while Anthropic said it is interested in future orbital compute capacity.

Analysis

The immediate read-through is not just "more AI demand" but a step-function reduction in the probability of a near-term compute bottleneck for frontier inference. That matters for NVDA because the market has been debating whether AI monetization could decelerate before utilization saturates; this deal signals the opposite, with demand still elastic enough that large buyers will lock up dedicated capacity to preserve product quality and usage tiers. In the near term, that supports incremental GPU procurement and keeps supply tight at the high end of the stack, especially for H-series and the newest generation accelerators.

The second-order winner is the infrastructure layer: a single tenant absorbing an entire data center implies hyperscale-style capacity is still being commoditized by the largest model companies, which favors the few vendors capable of delivering power, networking, and accelerated systems at scale. The risk is that investors over-anchor on headline megawatts and underweight the timing: the commercial benefit to chip demand is immediate, but the monetization benefit to AI app vendors arrives over months as higher usage caps translate into retention and expansion, not overnight revenue acceleration. If utilization rises faster than model pricing, the near-term outcome may be more capex intensity rather than margin expansion.

The contrarian angle is that this may be less about durable surplus demand and more about defensive provisioning against volatility in service quality. If so, the bullish read-through to NVDA is real but more modest than the market may infer, because the incremental demand could be absorbed by efficiency gains elsewhere or by shifting workloads across existing clusters. The bigger medium-term signal is that AI capacity constraints are moving from chip availability to power and site buildout, which could extend NVDA strength even if software sentiment cools.

Tail risk is a sudden reversal in AI capex sentiment if enterprise spend fails to convert into usage or if power-constrained data center expansion slips by 2-4 quarters. Conversely, if orbital compute remains rhetorical rather than executable, it does not change the next 12 months; the practical catalyst window is 1-3 quarters for GPU order revisions and 12-24 months for any meaningful infrastructure scaling narrative.

Key Decisions for Investors

• Stay tactically long NVDA into the next 1-3 quarters on any post-news dip; use weakness to add exposure because dedicated compute lockups suggest utilization remains high and supply remains tight.
• Pair long NVDA / short a lower-quality AI beneficiary basket over 1-2 quarters to express the view that hardware demand is the cleaner near-term winner than software monetization.
• Buy NVDA call spreads 2-4 months out to capture upside from continued capex commentary while limiting premium burn if the market rotates away from AI infrastructure.
• Do not chase the "orbital compute" narrative as a tradable catalyst; treat it as a long-dated option on power-constrained expansion, not a near-term earnings driver.
• Set a hedge trigger if AI capex rhetoric broadens but utilization data softens: trim NVDA exposure on signs of order normalization or if the next quarter shows slower-than-expected cloud and inference demand.