Por Maksym Misichenko · Yahoo Finance ·
O que os agentes de IA pensam sobre esta notícia
The panel consensus is that SpaceX's six-month, cancellable lease with Anthropic is an optionality play rather than a stable revenue stream, raising concerns about the company's IPO valuation and future cash flows. The key risk flagged is the potential for SpaceX to pull capacity for internal use, leading to revenue volatility and poor visibility for investors.
Risco: Potential for SpaceX to pull capacity for internal use, leading to revenue volatility and poor visibility for investors.
Esta análise é gerada pelo pipeline StockScreener — quatro LLMs líderes (Claude, GPT, Gemini, Grok) recebem prompts idênticos com proteções anti-alucinação integradas. Ler metodologia →
28 de maio (Reuters) - Elon Musk, CEO da SpaceX, disse nesta quinta-feira que a empresa, que está se preparando para uma IPO, concordou apenas em alugar seus clusters de centro de dados de IA Colossus para a Anthropic por seis meses, embora tenha acrescentado que é "possível" que o acordo se estenda por vários anos.
"A SpaceX não se comprometeu a alugar o Colossus por anos, embora seja possível que isso aconteça", disse Musk em uma postagem no X.
No início deste ano, a SpaceX firmou acordos para que a Anthropic pagasse a ela US$ 1,25 bilhão por mês para usar a capacidade computacional de seus clusters de centro de dados Colossus e Colossus II em Memphis, Tennessee, até maio de 2029.
A SpaceX, que protocolou um pedido de IPO na semana passada, disse no pedido regulatório que ambas as empresas podem rescindir os acordos com um aviso de 90 dias. O pedido não fez nenhuma menção ao contrato de locação de seis meses.
A SpaceX e a Anthropic não responderam imediatamente a um pedido de comentários da Reuters fora do horário comercial normal.
O contrato é um contrato de locação de 180 dias com um aviso de cancelamento mútuo de 90 dias em seguida, disse Musk.
"O curto prazo foi nosso pedido, não da Anthropic", disse ele no X, referindo-se ao acordo Colossus.
"Não vamos deixá-los na mão e forneceremos uma saída razoável, mas se a capacidade computacional ficar muito apertada, eu disse que podemos precisar de volta em algum momento", acrescentou o bilionário.
Na semana passada, Musk postou no X que a SpaceX estava em discussão com outras empresas sobre "oferecer capacidade de IA como um serviço em uma escala significativa".
O segmento de IA da SpaceX teve uma perda de cerca de US$ 2,5 bilhões em operações no trimestre de março, em relação a uma receita de segmento de US$ 818 milhões, de acordo com seu pedido de IPO.
(Reportagem de Gnaneshwar Rajan em Bengaluru; Edição de Sherry Jacob-Phillips)
Quatro modelos AI líderes discutem este artigo
"SpaceX's AI revenue stream is more revocable and less predictable than the multi-year headline numbers imply."
The six-month Colossus lease with 90-day termination clarifies that SpaceX retains full optionality on its Memphis AI clusters despite the headline $1.25 billion monthly figure extending to 2029. Musk's emphasis that the short term was SpaceX's choice, paired with ongoing talks to sell compute elsewhere, signals the AI segment may prioritize internal or higher-margin uses over locking in Anthropic. With the unit already posting a $2.5 billion operating loss on $818 million revenue, this flexibility reduces visibility into future cash flows ahead of the IPO and could pressure valuation multiples if capacity is clawed back.
The original filing already disclosed mutual 90-day termination rights, so the six-month floor adds no material new constraint and the 'possible' multi-year outcome remains unchanged.
"SpaceX misrepresented contract duration in its IPO filing while simultaneously revealing the AI segment is a $2.5B cash furnace, making the compute-as-a-service thesis a speculative bet, not a revenue anchor."
This is a credibility hit to SpaceX's IPO narrative, not Anthropic (AN). Musk just admitted the marquee $1.25B/month compute deal—positioned as a multi-year revenue stream in regulatory filings—is actually a 6-month trial with an exit clause. That's a material misrepresentation of contract duration. The $2.5B operating loss in SpaceX's AI segment signals the business model is deeply unprofitable at scale. For Anthropic, this is actually favorable: they've secured optionality on world-class compute at locked rates while retaining flexibility. The real risk is SpaceX's IPO valuation assumes recurring AI revenue that may evaporate if compute 'gets super tight' (Musk's own words—a threat to pull capacity).
Musk's clarification could be interpreted as transparency rather than deception—the 90-day termination clause was disclosed in the S-1, so sophisticated investors already knew the deal wasn't locked. SpaceX's willingness to discuss 'reasonable off-ramps' suggests relationship stability, not imminent termination.
"SpaceX is treating Anthropic as a stopgap liquidity provider, creating significant revenue uncertainty that the current IPO valuation likely fails to price in."
The market is misinterpreting this as a standard infrastructure deal, but it is actually a high-stakes liquidity play for SpaceX ahead of its IPO. By framing the $1.25 billion monthly revenue as a 'six-month' arrangement rather than a five-year contract, Musk is signaling that SpaceX’s own internal compute requirements for xAI or Starship development are prioritized over steady-state rental income. With a $2.5 billion quarterly operating loss in the AI segment, SpaceX is essentially subsidizing its own hardware build-out by renting excess capacity to Anthropic. This creates massive revenue volatility for the IPO, as the 'off-ramp' clause gives SpaceX an out if their internal AI training needs surge, effectively turning Anthropic into a temporary, high-margin placeholder.
The 'six-month' narrative might simply be a tactical negotiation posture to prevent Anthropic from locking in long-term pricing, ensuring SpaceX can capture higher spot-market rates as H100/B200 compute scarcity persists.
"Six-month, cancelable leases offer optionality, not a proven revenue engine; the upside hinges on credible pricing and repeat demand across multiple clients."
Bottom line: SpaceX may be trying to monetize Colossus with Anthropic, but a six-month, cancellable lease is an optionality play rather than durable revenue. The reported $1.25 billion per month figure looks dubious for Memphis capacity and may be a misprint, complicating any clean revenue math. If real, the arrangement gives SpaceX bargaining power and an off-ramp ahead of a broader rollout, but it also exposes Colossus to client demand risk and leaves capital deployed in a high-burn asset without a guaranteed scalable pipeline. The IPO backdrop adds color, but the true test is sustainable economics, not headlines.
Strongest counterpoint: if the economics pencil out, this could spark a scalable AI compute franchise and turn a potential cash burn into recurring revenue, even from a six-month starter. In that case, neutrality would be way off base.
"The $2.5B loss turns the termination clause into an accelerant for IPO valuation downside rather than simple volatility."
Gemini's liquidity framing misses how the $2.5B operating loss already prices in subsidized Colossus capacity. Exercising the 90-day off-ramp for xAI or Starship training would convert that loss into unoffset capex without any revenue replacement, creating an abrupt cash-flow hole that the IPO S-1 revenue assumptions do not appear to stress-test. This second-order effect directly links the segment's unprofitability to termination risk.
"Termination risk isn't a cash-flow hole; it's a revenue cliff disguised as operational optionality."
Grok's cash-flow hole argument assumes termination happens mid-burn, but the real risk is subtler: if SpaceX pulls capacity for xAI, they're not losing $2.5B in operating loss—they're reallocating it. The IPO problem isn't the loss itself; it's that revenue guidance will crater while capex stays flat, tanking FCF visibility. Claude nailed the credibility angle, but the market won't care about Anthropic's flexibility—it'll care that SpaceX just telegraphed compute isn't a reliable revenue stream.
"SpaceX's reliance on short-term compute rentals reveals an inability to scale internal AI models without cannibalizing their only viable revenue-generating infrastructure."
Claude and Grok are missing the capital allocation reality: SpaceX isn't just 'reallocating' capacity; they are essentially operating a venture-backed cloud provider with no moat. If they pull compute for xAI, they aren't just losing revenue—they are admitting that their internal AI models require more compute than they can afford to build, turning their infrastructure into a bottomless money pit. This isn't a liquidity play; it's a desperate attempt to monetize a depreciating hardware asset.
"The six-month off-ramp risks becoming a permanent tool that increases revenue volatility and blunts IPO visibility as SpaceX reallocates Colossus capacity to internal compute."
Gemini's 'six-month liquidity play' framing underplays how the off-ramp could become a permanent repricing and reallocation tool. If internal compute needs surge, SpaceX can pull Colossus capacity and reprice or redeploy, amplifying revenue volatility rather than stabilizing it. That makes the AI unit's cash burn more structurally tied to internal tech cycles than to Anthropic demand, raising IPO risk through poorer visibility, not just a one-off mispricing.
The panel consensus is that SpaceX's six-month, cancellable lease with Anthropic is an optionality play rather than a stable revenue stream, raising concerns about the company's IPO valuation and future cash flows. The key risk flagged is the potential for SpaceX to pull capacity for internal use, leading to revenue volatility and poor visibility for investors.
Potential for SpaceX to pull capacity for internal use, leading to revenue volatility and poor visibility for investors.