Elon Musk's "Mega Soft Plan" new actions revealed! Building a computing power cluster from scratch, completing in 6 months the work of OpenAI & Oracle in 15 months

Elon Musk's "Macrohard" plan has new developments:

In just 6 months, a computing cluster has been built from scratch, with a completed power supply capacity of 200MW, sufficient to support 110,000 NVIDIA GB200 GPUs NVL72.

In only 6 months, it accomplished what took OpenAI and Oracle 15 months to complete, setting a new record.

In response to a question from netizens, Musk revealed that the Colossus II computing cluster is indeed related to the Macrohard plan.

Although the name carries a mocking connotation towards Microsoft, Musk is serious about this matter, and the idea has been in place since 2021.

The core logic is: since software companies do not produce physical hardware, the entire process from coding, design, testing to management can theoretically be replicated by AI.

"Macrohard" will build a multi-agent system based on xAI's large language model Grok. Musk revealed that the project will deploy hundreds of specialized agents, some focused on coding, others on image and video generation, and some on software testing. They will work collaboratively to simulate a complete software development team.

The system will also simulate human users interacting with the software being developed in a virtual machine, refining the product through continuous iteration and feedback. The entire software development lifecycle, from initial requirements analysis, product design, coding implementation, to quality assurance and user testing, will be automated by AI agents.

To enable hundreds of complex AI agents to work simultaneously and conduct large-scale software simulations, supercomputing power is essential.

Colossus II exists for this purpose.

An Unprecedented Computing Behemoth

Everyone is already familiar with xAI's Colossus I, which built a computing cluster of about 200,000 H200 GPUs in just 122 days, and then doubled the scale to 200,000 GPUs within the following 92 days.

Colossus I remains the largest AI training computing cluster to date.

Now Musk is applying "first principles" to transfer successful experiences to Colossus II, expanding the scale by dozens of times, this time for AI inference.

The Colossus II project was launched on March 7, 2025, when xAI acquired a 1 million square foot warehouse and two adjacent plots in Memphis, totaling 100 acres Elon Musk stated in July that some racks have already begun installation.

By August 22, 119 air-cooled chiller units had been installed on site, providing approximately 200MW of cooling capacity, sufficient to support around 110,000 GB200 NVL72 GPUs.

According to the plan, the first phase of Colossus II will deploy 110,000 NVIDIA GB200 GPUs, with a final goal of exceeding 550,000 GPUs, and peak power demand is expected to exceed 1.1 gigawatts.

A longer-term roadmap even plans to expand the total number of GPUs to 1 million.

To address the enormous power demand, xAI has adopted a cross-regional energy strategy.

Due to resistance in obtaining gigawatt-level power locally in Memphis, xAI has acquired a former Duke Energy power plant across the state line in Mississippi. Mississippi regulators have temporarily approved xAI to operate gas turbines on the site for up to 12 months without formal licensing. Currently, the power plant has 7 turbines of 35MW each in operation.

To outpace competitors in deployment speed, xAI relies on leased gas turbines. Supplier Solaris Energy Infrastructure (SEI) has 600MW of gas turbines, of which approximately 400MW currently serves xAI, accounting for 67%. A newly established joint venture (with Solaris holding 50.1% and xAI holding 49.9%) has committed $112 million in capital expenditures for the second quarter of 2025.

Additionally, to avoid impacting the local power grid, xAI has deployed 168 Tesla Megapack battery storage systems at the Colossus II site to provide power support during peak electricity usage, ensuring local residents do not experience power outages.

Musk personally supervises, project enters sprint phase

Just yesterday, Musk shared a crazy work schedule:

• Worked overnight with the Optimus engineering team on Friday night, took a red-eye flight to Austin, arrived at 5 AM, and had lunch with his kids after waking up.

• Spent the entire Saturday afternoon conducting a deep technical review of Tesla's AI5 chip design.

• Flew to Colossus II on Monday, toured the entire data center floors, reviewed transformers and power production (progress is excellent), and left at midnight.

• Followed by a 12-hour meeting with various departments at Tesla, focusing on AI/autonomous driving, robot production, and vehicle production/delivery.

It is evident that Giant has become a key part of Musk's business landscape.

Tesla has positioned itself as an "AI robotics company," with 80% of its future value in robotics. The AI software developed by Giant can be used to optimize Tesla's autonomous driving algorithms, factory automation, and the functionality of the humanoid robot Optimus. In turn, Tesla's vast amounts of real-world data will provide valuable training data for Giant Risk Warning and Disclaimer

The market carries risks, and investment should be approached with caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investment based on this is at one's own risk