As with every fraud, you have to connect the dots and collect the pieces of the puzzle in order to get as full a picture as possible, and adjust the data and conclusions accordingly. So now we have a new PowerPoint slide from Firmus, from the failed IPO. So apparently, the range of VR to H200s is operated in Singapore with only 5MW. Let me say it again: 5MW. The other "42MW" is in Australia, located in Melbourne, and they claim that the shipment they received in March is already "Operational". In November 2025, Firmus raised at a 4.2B USD valuation, when they had only a 5MW operational datacenter. Yes, that's right. Now they had a great PowerPoint, like the one below, and they knew that in order to expand, they needed more money and more debt. So in February, Blackstone urgently had to help arrange financing and debt for the GB300 shipments, to protect the next fundraising round, which happened in April, and the plans for the IPO (which failed). Second, how did Firmus, a 5MW micro neocloud that has no experience in operating and deploying datacenters and racks, happen to take a shipment in March and have it Operational by September? Let's take SpaceX: the 400MW of GB300s that they claimed to deploy by the end of June (which also took a little time to deploy) is about to become operational these days, a quarter and a little bit more later. Counting the time before and after to make it fully operational, we may get around 5 to 6 months. How did Firmus do that in that time frame? Maybe Meta, which, according to the reports, is the customer of these 42MW of GB300s, can provide more information?

NVIDIA Series | PART 1 is out. In this series, I will discuss Nvidia GPUs, revenue, quarterly reports, Rubin, data centers, OEMs, hyperscalers, neoclouds, and how they are all connected and served to sustain the most important machine in the markets right now, The Nvidia’s Beat and Raise. PART 1 covers most of Blackwell’s timeline, numbers, and definitions that are important to understand if you follow or invest in Nvidia, along with the revenue associated with Blackwell.

False. As of the end of February 2026, xAI proceeded with the installation of an array of 27 mobile gas turbines at the Southaven site. Assuming these units are operational, they would provide a generating capacity of approximately 495 MW, effectively functioning as a large-scale power plant. Assuming the incomplete data center is occupied and capable of fully utilizing the entire 495 MW of power, that capacity could support approximately 350,000 to 400,000 H100-equivalent GPUs.

Jensen last minute joined Trump’s trip to China to make sure the flow to Malaysia & Co. remains as usual, sorry, I mean, so he can finally clear Nvidia’s leftover H200 inventory. Or maybe move some of the GB200s sitting in warehouses? Blackwell already celebrated its first birthday It is the time to pitch it to China?