News
SpaceX to put custom Starship propellant storage tanks through first trial
In the latest twist in the saga of SpaceX’s custom-built Starship launch pad propellant storage tanks, the company appears to have retroactively decided to build small prototype meant solely for testing.
Known as a ‘test tank,’ the relatively small steel structure was fairly rapidly assembled from parts of an older Ground Support Equipment (GSE) tank scrapped in July over the last week or so. SpaceX completed the first Starship-derived propellant storage tank in April 2021 and rapidly rolled that tank (GSE1) and a second (GSE2) from the build site to the orbital launch pad just a few weeks apart. Less than a month after that, SpaceX also completed GSE tank #3, though things seemingly devolved into chaos immediately thereafter.
Only three months later would GSE3 finally be transported to – and installed on a concrete mount at – Starship’s first orbital launch site, and only after a number of structural modifications and in the footsteps of GSE tanks #5 and #6. Little is known about why SpaceX’s custom GSE tank production faltered so soon after it began, why none of the five Starship-sized tanks installed at the orbital pad have been fully plumbed or subjected to any kind of testing, or why structural modifications were seemingly required after the fact. However, it’s safe to say that SpaceX’s brand new GSE ‘test tank’ is now at the center of the mystery.
Thankfully, at minimum, the rapid appearance of SpaceX’s first GSE test tank returns some level of familiarity to the brief but chaotic history of Starship’s orbital launch pad propellant tanks. Test tanks are nothing new and have been an integral part of Starship development since Test Tank 1 first headed to SpaceX’s suborbital launch (and test) facilities in January 2020. In the 20 months since, SpaceX has built and tested seven small test tanks, several of which didn’t survive.
Whether intentionally destroyed or not, each test tank explicitly helped SpaceX qualify new manufacturing techniques, different materials, and different skin thickness and generally gather data more quickly and cheaply than full-scale prototypes would allow. Most recently, for example, SpaceX seemingly successfully tested a Super Heavy booster test tank, subjecting the prototype to cryogenic liquid nitrogen and using hydraulic rams to simulate the thrust of nine Raptor engines on an unproven disk-like thrust structure.
Now, almost as if SpaceX snapped out of a trance and remembered the utility of test tanks, the company has assembled a subscale GSE prototype presumably meant to verify that its custom-built propellant storage tanks can handle a set of conditions significantly different from the Starships they’re derived from. In this case, that GSE tank was quite literally built from scrapped sections of GSE tank #4. In fact, the top half (forward dome section) was quite literally cut off of GSE4 after the tank was scrapped last month for unknown reasons.
Over the last several months, while GSE tank production and installation took an unexpected hiatus, SpaceX workers slowly but surely welded steel rings (stiffeners) to the exterior of GSE1, GSE2, and GSE3. When GSE5 and GSE6 eventually headed to the pad, they left with those stiffeners already installed, implying that whatever tripped SpaceX up was likely structural. The GSE4 test tank also includes external stiffeners along each circumferential weld (where rings were stacked or domes joined).
At the same time as SpaceX was (or wasn’t, for several months) building its own GSE tanks, contractors normally tasked with assembling water towers and storage tanks in situ built eight massive 12m (~40 ft) wide tanks of their own. Deemed “cryo shells,” much like their name suggests, those tanks are meant to fully enclose SpaceX’s GSE tanks. SpaceX will use those shells to insulate their thin, single-walled steel propellant tanks, thus keeping their cryogenic contents cryogenic for as long as possible. How they’ll be insulated is unclear, though.
Based on the seemingly retroactive decision to strengthen the exterior of those GSE tanks, the general consensus as of late is that SpaceX wants to pull at least a partial vacuum in the gap between shell and tank. It’s also possible that SpaceX will do the opposite and pressurize that gap (as much as possible) with an insulative gas like nitrogen. Extra confusion comes from the fact that Starship tanks are technically designed to support a literal spacecraft (operating in a near-total vacuum) without the need for external stiffeners.
Additionally, it’s fairly clear that SpaceX hasn’t built a custom subscale cryoshell or concrete installation pad for its GSE4 test tank, meaning that it will really only be useful for testing some of the loads operational GSE tanks will experience inside their sleeves. Additionally, given that SpaceX has already completed six of the orbital pad’s seven GSE tanks and all seven of their cryosleeves, the discovery of any significant issues during GSE4 testing could easily trigger months of rework and delays. With any luck, though, GSE4 will sail through an imminent test campaign, clearing the way for SpaceX to finish plumbing, sleeving, and activating Starship’s first orbital launch site tank farm.
Elon Musk
Tesla confirms that work on Dojo 3 has officially resumed
“Now that the AI5 chip design is in good shape, Tesla will restart work on Dojo 3,” Elon Musk wrote in a post on X.
Tesla has restarted work on its Dojo 3 initiative, its in-house AI training supercomputer, now that its AI5 chip design has reached a stable stage.
Tesla CEO Elon Musk confirmed the update in a recent post on X.
Tesla’s Dojo 3 initiative restarted
In a post on X, Musk said that with the AI5 chip design now “in good shape,” Tesla will resume work on Dojo 3. He added that Tesla is hiring engineers interested in working on what he expects will become the highest-volume AI chips in the world.
“Now that the AI5 chip design is in good shape, Tesla will restart work on Dojo3. If you’re interested in working on what will be the highest volume chips in the world, send a note to AI_Chips@Tesla.com with 3 bullet points on the toughest technical problems you’ve solved,” Musk wrote in his post on X.
Musk’s comment followed a series of recent posts outlining Tesla’s broader AI chip roadmap. In another update, he stated that Tesla’s AI4 chip alone would achieve self-driving safety levels well above human drivers, AI5 would make vehicles “almost perfect” while significantly enhancing Optimus, and AI6 would be focused on Optimus and data center applications.
Musk then highlighted that AI7/Dojo 3 will be designed to support space-based AI compute.
Tesla’s AI roadmap
Musk’s latest comments helped resolve some confusion that emerged last year about Project Dojo’s future. At the time, Musk stated on X that Tesla was stepping back from Dojo because it did not make sense to split resources across multiple AI chip architectures.
He suggested that clustering large numbers of Tesla AI5 and AI6 chips for training could effectively serve the same purpose as a dedicated Dojo successor. “In a supercomputer cluster, it would make sense to put many AI5/AI6 chips on a board, whether for inference or training, simply to reduce network cabling complexity & cost by a few orders of magnitude,” Musk wrote at the time.
Musk later reinforced that idea by responding positively to an X post stating that Tesla’s AI6 chip would effectively be the new Dojo. Considering his recent updates on X, however, it appears that Tesla will be using AI7, not AI6, as its dedicated Dojo successor. The CEO did state that Tesla’s AI7, AI8, and AI9 chips will be developed in short, nine-month cycles, so Dojo’s deployment might actually be sooner than expected.
Elon Musk
Elon Musk’s xAI brings 1GW Colossus 2 AI training cluster online
Elon Musk shared his update in a recent post on social media platform X.
xAI has brought its Colossus 2 supercomputer online, making it the first gigawatt-scale AI training cluster in the world, and it’s about to get even bigger in a few months.
Elon Musk shared his update in a recent post on social media platform X.
Colossus 2 goes live
The Colossus 2 supercomputer, together with its predecessor, Colossus 1, are used by xAI to primarily train and refine the company’s Grok large language model. In a post on X, Musk stated that Colossus 2 is already operational, making it the first gigawatt training cluster in the world.
But what’s even more remarkable is that it would be upgraded to 1.5 GW of power in April. Even in its current iteration, however, the Colossus 2 supercomputer already exceeds the peak demand of San Francisco.
Commentary from users of the social media platform highlighted the speed of execution behind the project. Colossus 1 went from site preparation to full operation in 122 days, while Colossus 2 went live by crossing the 1-GW barrier and is targeting a total capacity of roughly 2 GW. This far exceeds the speed of xAI’s primary rivals.
Funding fuels rapid expansion
xAI’s Colossus 2 launch follows xAI’s recently closed, upsized $20 billion Series E funding round, which exceeded its initial $15 billion target. The company said the capital will be used to accelerate infrastructure scaling and AI product development.
The round attracted a broad group of investors, including Valor Equity Partners, Stepstone Group, Fidelity Management & Research Company, Qatar Investment Authority, MGX, and Baron Capital Group. Strategic partners NVIDIA and Cisco also continued their support, helping xAI build what it describes as the world’s largest GPU clusters.
xAI said the funding will accelerate its infrastructure buildout, enable rapid deployment of AI products to billions of users, and support research tied to its mission of understanding the universe. The company noted that its Colossus 1 and 2 systems now represent more than one million H100 GPU equivalents, alongside recent releases including the Grok 4 series, Grok Voice, and Grok Imagine. Training is also already underway for its next flagship model, Grok 5.
Elon Musk
Tesla AI5 chip nears completion, Elon Musk teases 9-month development cadence
The Tesla CEO shared his recent insights in a post on social media platform X.
Tesla’s next-generation AI5 chip is nearly complete, and work on its successor is already underway, as per a recent update from Elon Musk.
The Tesla CEO shared his recent insights in a post on social media platform X.
Musk details AI chip roadmap
In his post, Elon Musk stated that Tesla’s AI5 chip design is “almost done,” while AI6 has already entered early development. Musk added that Tesla plans to continue iterating rapidly, with AI7, AI8, AI9, and future generations targeting a nine-month design cycle.
He also noted that Tesla’s in-house chips could become the highest-volume AI processors in the world. Musk framed his update as a recruiting message, encouraging engineers to join Tesla’s AI and chip development teams.
Tesla community member Herbert Ong highlighted the strategic importance of the timeline, noting that faster chip cycles enable quicker learning, faster iteration, and a compounding advantage in AI and autonomy that becomes increasingly difficult for competitors to close.
AI5 manufacturing takes shape
Musk’s comments align with earlier reporting on AI5’s production plans. In December, it was reported that Samsung is preparing to manufacture Tesla’s AI5 chip, accelerating hiring for experienced engineers to support U.S. production and address complex foundry challenges.
Samsung is one of two suppliers selected for AI5, alongside TSMC. The companies are expected to produce different versions of the AI5 chip, with TSMC reportedly using a 3nm process and Samsung using a 2nm process.
Musk has previously stated that while different foundries translate chip designs into physical silicon in different ways, the goal is for both versions of the Tesla AI5 chip to operate identically. AI5 will succeed Tesla’s current AI4 hardware, formerly known as Hardware 4, and is expected to support the company’s Full Self-Driving system as well as other AI-driven efforts, including Optimus.
Tesla confirms that work on Dojo 3 has officially resumed
Elon Musk’s xAI brings 1GW Colossus 2 AI training cluster online
