Elon Musk announced the transfer of SpaceX engineering data to train the next generation of his AI model, Grok. On July 21, on platform X, he stated that this move will directly enhance the model's engineering capabilities. Key clarification: data subject to US export controls (ITAR) will not be included in the training set, keeping SpaceX within legal bounds while still granting the model access to critical information on rocket and satellite development.
New model with 2 trillion parameters
Training will take place on a new architecture with 2 trillion parameters — nearly double the 1.5 trillion in the V9 version, which underpins the public release of Grok 4.5. SpaceX has accumulated this vast dataset over years of rapid growth: this year, the company successfully conducted a record public offering of SPCX shares and raised fresh capital for engineering and AI projects.
Developer opinions diverge
Musk responded to a developer's post who was simultaneously testing Grok Build and OpenAI's Codex. According to the tester's assessment, Grok Build already outperforms Claude on complex programming tasks, although Codex holds its ground. However, earlier reviewers pointed to inconsistent results from Grok 4.5, despite Musk's claims comparing the model to Anthropic's Opus class.
SpaceX ramps up AI projects
The data transfer coincides with $60 billion in investments in developer tools — including the acquisition of Cursor's parent company just weeks after its IPO. Musk also set a goal to boost SpaceX's revenue to a trillion dollars by 2030. Meanwhile, xAI is expanding Grok's capabilities: the chatbot's official page announced Grok for Excel, allowing users to build financial models, analyze market data, and work with charts based on Grok 4.5.
This launch shows that xAI is building new products on the same model that will soon be trained on SpaceX engineering data. Exclusive information about rockets and satellites is being directly integrated into Grok's development plans. Additionally, Grok gains access to an engineering dataset that competitors like OpenAI and Anthropic lack. Whether the new 2-trillion-parameter model will catch up to Claude and Codex will be determined by real-world tests, not Elon Musk's statements.
My expert conclusion: The integration of SpaceX data is not just a marketing move but a strategic advantage. The unique dataset, tied to real-world engineering tasks, could give Grok a qualitative leap in solving applied problems, especially in physics and modeling. However, without public benchmarks, it is too early to trust the founder's ambitious claims.