SpaceXAI on Wednesday launched Grok 4.5, its latest AI model for coding and agentic tasks, after training it across tens of thousands of Nvidia GB300 graphics processing units. The company wrapped the release in the usual corporate hymn about intelligence and efficiency, but the machinery underneath is plain enough: vast compute, private ownership, and a market race to sell more powerful systems to enterprise customers.
The Compute Empire
SpaceXAI said Grok 4.5 was trained with a focus on meticulous data filtering, deduplication and quality scoring. That sounds tidy. It also means the model sits on top of an industrial pipeline of data selection and machine labour, all controlled by a company that decides what counts as useful, clean, and profitable enough to feed the system. Grok 4.5 is immediately available through SpaceXAI's AI coding agent, Grok Build, in Cursor and through the SpaceXAI console, the company's developer portal, using an API key. SpaceXAI said EU availability is expected in mid-July. The Brussels market gets the product when the company says so.
Cursor, the popular AI coding agent, said: "We've partnered with SpaceXAI to train Grok 4.5." That partnership sits inside a wider consolidation drive. SpaceX said last month it would buy Anysphere, the startup behind Cursor, in an all-stock deal worth $60 billion to boost its presence in the lucrative enterprise AI tools market. The language is polished. The logic is naked. Bigger firms absorb smaller ones, then sell the result back to businesses as innovation.
The Price of Access
SpaceXAI priced Grok 4.5 at $2 per million input tokens and $6 per million output tokens. Elon Musk said in a post on X that "It is an Opus-class model, but faster, more token-efficient and lower cost." Rival Anthropic's Claude Opus 4.8 is priced at $5 per million input tokens and $25 per million output tokens. OpenAI's GPT-5.6 Luna is priced at $1 per million input tokens and $6 per million output tokens. The competition is presented as a technical race, but the numbers show a market where access to advanced AI is priced like a utility for firms, not a public service for everyone else.
Input tokens are the text, code or other data sent to an AI model, while output tokens are the text or code the model generates in response. That basic exchange is now the unit of a corporate economy built around extraction, automation and subscription fees. The companies call it progress. The invoices do the talking.
Musk's AI startup xAI was acquired by SpaceX in February. He said in May that xAI would cease to exist as a separate company and would instead become SpaceXAI. Another merger, another private empire swallowing its own parts and renaming the result. The branding changes. The ownership doesn't.
Security, Secrecy and the State
OpenAI will publicly launch its most advanced AI model GPT-5.6 on Thursday, following a delay last month prompted by U.S. government requests over national security concerns about the potential misuse of powerful AI technologies. OpenAI said GPT-5.6 Sol, Terra and Luna will launch on Thursday. The company said GPT-5.6 Sol was competitive with Anthropic's Mythos Preview on the ExploitBench cybersecurity benchmark. OpenAI had limited GPT-5.6 access to a small group of vetted partners whose details were shared with the authorities.
That detail matters. The state shows up not as a neutral referee, but as a gatekeeper worried about misuse, vetting partners, and deciding who gets access to what. The companies still sell the same systems; the authorities just want a list first. Reuters' "Major AI offerings at a glance" said OpenAI will launch GPT-5.6 on Thursday after a delay prompted by U.S. government concerns over the potential national security risks posed by increasingly powerful AI systems. The release comes amid intensifying competition among AI developers who are racing to improve model performance, cut costs and expand capabilities for enterprise customers, fueling a wave of new systems and reasoning models across the industry.
Reuters also said Chinese developers are reshaping the economics of AI by delivering increasingly capable models at a fraction of the cost. So the race tightens, prices fall, and the market keeps moving. The winners are the firms with the biggest compute budgets, the deepest state tolerance, and the sharpest appetite for enterprise contracts. Everyone else gets the bill, the surveillance, or the automation.