【SpaceX总裁兼首席运营官解释了SpaceX的人工智能战略】
SpaceX总裁兼首席运营官格温·肖特韦尔(Gwynne Shotwell)向CNBC清晰地解释了SpaceX的人工智能战略，而大多数投资者仍然将其视为一家火箭公司。

她首先澄清了一个大家都容易误解的问题。

当CNBC问及轨道计算是否会使地球上的数据中心过时时，肖特韦尔回答说不会，两者都不可或缺。

肖特韦尔直言不讳地指出增长方向：随着时间的推移，太空计算的扩张速度将远远超过地球上的计算扩张速度，因为限制地球上数据中心发展的物理限制——土地、电力、冷却、许可——在轨道上根本不存在。

太空中的辐射冷却是免费的，轨道上的太阳能源源不断且丰富。

如今，所有超大规模数据中心运营商都在争夺数据中心的土地和电力使用权，原因在于地球环境受限，而太空则不然。

与Anthropic的交易揭示了SpaceX在计算市场中的真实立场。

Anthropic公司每月向SpaceX支付12.5亿美元，合同有效期至2029年5月。但合同中包含一项为期六个月的退出条款，Shotwell详细解释了该条款存在的原因。

SpaceX永远不会出售其自身真正需要的计算资源。

SpaceX首先为自身构建计算资源，其次才将剩余资源出售给像Anthropic这样的公司，并保留在内部需求需要时随时收回这些计算资源的权利。

Cursor的收购案则完全是另一回事。

大多数投资者都认为这笔交易已经完成，但Shotwell的解读比收购本身更有意思。

她将其描述为一家拥有优秀训练数据的模型公司与一家拥有庞大计算资源的公司相遇，并表示双方目前正在研究彼此的模型，互相学习，并在做出最终决定前密切合作。

由此展现的垂直整合格局，与当今公开市场上的任何其他交易都截然不同。

SpaceX 正在建设自己的天然气管道、开采自己的推进剂、建造自己的发射基础设施、制造自己的芯片、部署自己的卫星、运营自己的地球和轨道数据中心、运行自己的 AI 模型，并有选择地将剩余容量出售给世界上最先进的 AI 公司。
SpaceX President and COO Gwynne Shotwell gave CNBC the clearest explanation of what SpaceX's AI strategy actually is and most investors are still processing it as a rocket company.

The first thing she clarifies is the question everyone gets wrong.

When CNBC asked whether orbital compute makes Earth-based data centers obsolete, Shotwell said no, you will need both.

Shotwell was direct about where the growth goes, compute expansion in space will far outpace compute expansion on Earth over time, because the physical constraints that limit data center growth on Earth land, power, cooling, permitting, simply do not exist in orbit.

Radiative cooling in space is free, solar energy in orbit is constant and abundant.

The reason every hyperscaler is fighting over land and power rights for data centers today is because Earth is a constrained environment and space is not.

The Anthropic deal is the part that reveals SpaceX's actual posture in the compute market.

Anthropic is paying SpaceX $1.25 billion per month through May 2029, but the contract includes a six-month exit clause and Shotwell explained exactly why that clause is there.

SpaceX will never sell compute it actually needs.

SpaceX is building compute for itself first, selling the overflow to companies like Anthropic second and reserving the right to pull that compute back the moment internal demand requires it.

The Cursor situation adds another layer entirely.

Most investors are treating it as a done deal, but Shotwell's framing is more interesting than the acquisition itself.

She describes it as a model company with excellent training data meeting a company with enormous compute and says they are currently looking at each other's models, learning from each other, and collaborating closely before any final decision is made.

The vertical integration picture that emerges from all of this is unlike anything else trading in the public markets today.

SpaceX is building its own natural gas pipelines, mining its own propellant, constructing its own launch infrastructure, building its own chips, deploying its own satellites, operating its own data centers on Earth and in orbit, running its own AI models and selectively selling overflow capacity to the most sophisticated AI companies in the world. http://t.cn/AXaWOAeo