Elon Musk's xAI division is preparing to train its next-generation Grok 5 model on a quarter-century of SpaceX engineering data, a move that could transform how engineers approach complex technical problems. Deepwater Asset Management analyst Gene Munster predicted the model could become "the most useful engineering tool in history" as the company scales its AI infrastructure to support increasingly ambitious training efforts.

What Makes Grok 5 Different From Earlier Versions?

The key differentiator for Grok 5 lies in its training data. Unlike previous Grok models trained on general internet text, Grok 5 will incorporate SpaceX's entire historical corpus of engineering documentation, design files, and technical knowledge accumulated over 25 years of rocket development and space operations. This specialized dataset could give the model unprecedented capability in solving real-world engineering challenges.

SpaceX is already building the infrastructure to support this ambitious training effort. The company's Colossus II AI complex in Memphis is currently training Grok 5 using Nvidia GB200 and GB300 processors deployed at scale. SpaceX expects to have more than two gigawatts of computing capacity this year and close to 10 gigawatts by the end of 2027, providing the raw computational power needed for increasingly sophisticated models.

How Does Grok 4.7 Set the Stage for What's Coming?

SpaceXAI released Grok 4.7 on September 21, roughly three weeks after Musk's initial September 1 target date. Despite the delay, Munster noted that shipping a major model update within three weeks of the original deadline represents solid execution by Musk's standards. The release signals that xAI's development pipeline is functioning reliably, which increases confidence that Grok 5 will arrive on schedule later this year.

Grok 4.7 represents a meaningful step forward in capability. The model features a 500,000-token context window, meaning it can process roughly 400,000 words at once, and uses a larger base model than its predecessor with extended reinforcement learning training optimized for multi-hour tasks. On internal benchmarks, Grok 4.7 scored 46.3% on CursorBench 4.0, up from 40.4% for Grok 4.6, and achieved 64% on EEBench. The model became available through the Grok API, Cursor, and Grok Build on launch day.

"What's going to make Grok 5 special is it's incorporating the entire corpus of SpaceX data. Basically, all the data that SpaceX has ever produced, which is a tremendous amount over the course of a quarter century. Translation: Grok 5 will be the most useful engineering tool in history," said Gene Munster.

Gene Munster, Deepwater Asset Management

How to Evaluate xAI's Progress in the AI Market

• Revenue Growth: SpaceX's AI revenue jumped approximately 250% year over year in the second quarter, demonstrating rapid commercial traction for Grok and related services.
• Infrastructure Investment: AI capital spending reached $15.83 billion, with plans to scale compute capacity from two gigawatts currently to nearly 10 gigawatts by end of 2027, showing serious long-term commitment.
• Model Performance Trajectory: Each Grok release shows measurable improvements on coding and knowledge benchmarks, with Grok 4.7 demonstrating gains over 4.6 and setting expectations for Grok 5's capabilities.

What Competition Is xAI Facing?

xAI's momentum is not going unnoticed by competitors. OpenAI is actively developing features to counter SpaceX's Grok Bot, which offers always-on AI teammates with access to email, calendar, Notion, and Figma integration starting at $30 per month. OpenAI is also considering a personal assistant response to Meta's Muse product, leveraging existing agentic technology already integrated with email and calendar systems.

The competitive landscape reflects a broader shift toward AI agents that can handle multi-day tasks and integrate with everyday work tools. OpenAI's approach would repurpose existing Codex and ChatGPT agentic technology, though awareness of these features remains low while SpaceX and Meta have seized the narrative around always-on AI assistants. The timing of OpenAI's launches remains unclear amid ongoing safety considerations.

Munster's confidence in Grok 5's potential rests on a simple but powerful premise: domain-specific training data can unlock capabilities that general-purpose models cannot match. By feeding 25 years of SpaceX's engineering knowledge into Grok 5, xAI is betting that specialized expertise will prove more valuable than raw scale alone. If that bet pays off, the model could reshape how engineers approach problem-solving across aerospace, manufacturing, and other technical fields.