Elon Musk said Tuesday that xAI is training seven models on “SpaceXAI Colossus 2,” and wrote that there is “some catching up to do.”
The April 8 post offered a concise update on xAI’s current position in the frontier AI race. Musk listed the training runs as Imagine V2, two variants of 1T, two variants of 1.5T, 6T, and 10T. He did not provide benchmark data, architectural detail, or release timing.
In current AI usage, labels such as 1T, 1.5T, 6T, and 10T are generally understood as references to trillion-parameter model classes. Parameters are the internal numerical weights learned during training. They are one measure of scale, but not a direct measure of capability. A larger parameter count can signal a more expensive and ambitious training run without establishing that a model matches a rival system on reasoning, coding, or autonomous task performance.
The 10T label nonetheless matters. It signals a company willing to train at extraordinary scale while still describing itself as behind the leaders. Musk’s wording made that point unusually plain.
xAI entered this phase of the race with a larger corporate base behind it than most AI companies can draw on. In March 2025, xAI acquired X in an all-stock transaction. On February 2, 2026, SpaceX acquired xAI, bringing the AI company into a broader Musk structure that already spanned launch, satellites, communications infrastructure, and the X platform.
That consolidation gave xAI more than a training cluster. It gave the company deeper capital backing and a more integrated operating base as it accelerated its buildout. On Jan. 6, xAI said it had raised $20 billion in a Series E round at a valuation of more than $120 billion. The company said the financing would support infrastructure and product development.
Public reporting has reinforced the scale of that infrastructure push. Reuters reported in December 2024 that xAI’s Memphis supercomputer was expected to be the world’s largest and that the first phase had been assembled in 122 days using 100,000 Nvidia Hopper GPUs. Reuters reported in September 2024 that xAI was planning a broader Memphis expansion aimed at supporting as many as one million GPUs.
Those moves have given xAI a larger role in the market conversation. They have also clarified its strategy. The company is investing heavily in compute and training capacity as it works to narrow the distance with firms whose leading models already support broader product and enterprise deployments.
OpenAI has spent recent months expanding around that wider stack. In February, the company introduced Frontier, which it described as a platform for enterprises to build, deploy, and manage AI agents. Later that month, OpenAI announced Frontier Alliances with BCG, McKinsey, Accenture, and Capgemini to support enterprise deployment. On April 6, the company published a new industrial-policy agenda and said it would open an OpenAI Workshop in Washington in May.
Anthropic has moved along a different but equally consequential track. On April 7, the company announced Project Glasswing with partners including Amazon Web Services, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, Microsoft, Nvidia, and Palo Alto Networks. Anthropic said the initiative was built around Claude Mythos Preview, an unreleased model that had reached a level of cyber capability warranting restricted defensive deployment.
Those developments help explain why xAI’s scale claims do not settle the competitive picture. A large training run can demonstrate access to capital, chips, and infrastructure. It does not establish parity with models already shaping coding, enterprise workflows, and high-consequence security applications. Musk’s post reflected that distinction. He described a major training effort and a company still trying to close the gap.
xAI’s public product lineup remains narrower. Grok, Imagine, API access, voice products, and companions give the company visible consumer and developer surfaces. The latest training disclosure indicated that xAI is still concentrating on the model layer underneath those products as it tries to move into the top tier.
The company’s rapid expansion has also been accompanied by a recent wave of departures from senior and technically important staff. In February, xAI cofounders Jimmy Ba and Tony Wu both left the company. More departures followed across research, infrastructure, growth, product, and technical staff. The list included Hang Gao, Vahid Kazemi, Ayush Jaiswal, Shayan Salehian, Simon Zhai, Andrew Ma, Radhakrishnan Venkataramani, Rahul Ravishankar, and founding member Toby Pohlen. Several had worked on Grok, reasoning, reinforcement learning, coding systems, search, recommendation, or the Imagine product. The departures came during a broader internal reorganization that Musk said was intended to improve speed of execution.
That turnover does not resolve the broader competitive question, but it adds context to a period in which xAI is scaling aggressively on both capital and compute. The company is trying to train larger systems while also managing organizational change across its research and product ranks.
Musk’s post did not establish whether xAI is now close to matching the strongest systems from OpenAI or Anthropic. It did establish something narrower and still important. xAI is training at significant scale, spending heavily, and moving with visible urgency because it still sees the leaders as ahead.
