News
SpaceX’s Mr Steven spotted in high-speed test at sea with upgraded net
SpaceX’s newly-outfitted recovery vessel Mr Steven was recently captured conducting aggressive maneuvers off the coast of Port of Los Angeles, just days after the vessel’s massive new arms and net were installed for the first time. The intense pace of upgrades and acceptance testing confirm beyond any reasonable doubt that SpaceX does not intend to waste its next Falcon 9 fairing recovery attempt, set to accompany the July 25th launch of Iridium-7.
The iconic fairing recovery vessel has – for the past three or four weeks – been undergoing major upgrades to its arms or claws, as well as a massive, new net spanning nearly 0.9 acres (3700 m²). With what appears to be a genuine fourfold increase in usable area for fairing recoveries, SpaceX likely has a very strong chance of actually pulling off its first successful catches and reuses of Falcon 9 payload farings, valued at roughly 5% of the rocket’s cost ($3 million per a $60 million base price) per half. Manufacturing cost and price to the customer are difficult to compare, but it at least offers a hint of the full cost of each ~800 kg segment of carbon fiber and aluminum honeycomb.
Mr Steven seen just after a day spent conducting sea-trials a few miles offshore, July 14. (Pauline Acalin)
Based on photos and video captured between July 12 and 15, Mr Steven’s crew and recovery technicians appeared to waste no time at all leaping from arm and net installation to sea-trials of the new hardware at least as extreme as anything previously observed from the SpaceX-leased vessel. Less than half an hour after leaving the harbor for the first time since his massive new arms arrived, Marinetraffic tracking data showed that Mr Steven was already performing aggressive turns and sprints at speeds up to 20 knots (~25 mph), fairly impressive given the vessel’s 200 foot (62 meter) length and gross weight of nearly 200,000 pounds (82,000 kg).
While this may seem impressive, Mr Steven is a class of ship known as a Fast Supply Vessel (FSV) designed to routinely transport a full 400 metric tons of cargo on its deck at cruising speeds of 23 knots (27 mph), which means that the only thing Mr Steven’s wildly expansive arms likely challenge is the vessel’s center of gravity (balance), hence the follow-up tests with hard turns at high speed.
Also of interest, an extraordinary video of some of that testing – unofficially captured, somehow, by drone – showed the ship aggressively maneuvering in reverse, an ability that could come in useful during recovery attempts if the expanded net’s coincidental protection of Mr Steven’s cockpit means that it can become a less fixed element, actively seeking out falling fairings to help close the gap on each parasailing half’s 50 meter error margin.
-
- Mr Steven makes some serious waves, using his pod thrusters to strafe backwards at 5-10 knots. (anonymous)
-
- It’s subtle, but a small plus sign appears to ‘mark the spot’ on Mr Steven’s new net, stretching roughly 60×60 meters. (anonymous)
-
- Mr Steven shows off the fancy new rigging of that upgraded net. (Pauline Acalin)
Another opportunity fast approaches
Previously scheduled for July 20, Iridium’s NEXT 7 multi-satellite launch was pushed back a handful of days to July 25 to give SpaceX engineers and technicians additional time to prepare what is the company’s third Block 5 Falcon 9 to roll off its Hawthorne, CA assembly line. While suboptimal for the customer and for SpaceX’s manifest, that slight delay very likely padded slim schedule margins for Mr Steven’s major arm upgrades, meaning that the vessel will now be able to participate in the imminent launch’s recovery operations. After the first flightworthy vehicle’s debut in May 2018, SpaceX’s rocket production has ramped up in quite an extreme fashion, jumping from four first stages produced in six months to another three or four boosters completed and tested in Texas in just two months.
While the transportation of Falcon fairings and upper stages is far harder to keep track of, production of those critical components of the rocket have also reached throughput levels that are new territory for SpaceX, including an impressive statistic of an average of one full Merlin 1D rocket engine manufactured daily according to an individual with experience on the factory floor.
The Block 5 iteration of the workhorse SpaceX vehicle is in many ways a wholly new rocket, featuring an array of upgrades that include new heat shielding at the rocket’s base, interstage, and legs; retractable landing legs, upgraded Merlin 1D engines, and a clean-sweep refresh of the vehicle’s avionics, to name just a handful of the major changes included.
SpaceX technicians wrench on a trio of varied Merlin 1Ds in McGregor, Texas, where every single engine is test-fired before being attached to a Falcon 9. (SpaceX)
Follow us for live updates, peeks behind the scenes, and photos from Teslarati’s East and West Coast photographers.
Teslarati – Instagram – Twitter
Tom Cross – Twitter
Pauline Acalin – Twitter
Eric Ralph – Twitter
Elon Musk
Elon Musk’s xAI bets $20B on Mississippi with 2GW AI data center project
The project is expected to create hundreds of permanent jobs, dramatically expand xAI’s computing capacity, and further cement the Mid-South as a growing hub for AI infrastructure.
Elon Musk’s xAI plans to pour more than $20 billion into a massive new data center campus in Southaven, Mississippi, marking the largest single economic development project in the state’s history.
The project is expected to create hundreds of permanent jobs, dramatically expand xAI’s computing capacity, and further cement the Mid-South as a growing hub for AI infrastructure.
xAI goes MACROHARDRR in Mississippi
xAI has acquired and is retrofitting an existing facility in Southaven to serve as a new data center, which will be known as “MACROHARDRR.” The site sits near a recently acquired power plant and close to one of xAI’s existing data centers in Tennessee, creating a regional cluster designed to support large-scale AI training and inference.
Once completed, the Southaven facility is expected to push the company’s total computing capacity to nearly 2 GW, placing it among the most powerful AI compute installations globally. The data center is scheduled to begin operations in February 2026.
Gov. Tate Reeves shared his optimism about the project in a press release. “This record-shattering $20 billion investment is an amazing start to what is sure to be another incredible year for economic development in Mississippi. Today, Elon Musk is bringing xAI to DeSoto County, a project that will transform the region and bring amazing opportunities to its residents for generations. This is the largest economic development project in Mississippi’s history,” he said.
xAI’s broader AI ambitions
To secure the investment, the Mississippi Development Authority approved xAI for its Data Center Incentive program, which provides sales and use tax exemptions on eligible computing hardware and software. The City of Southaven and DeSoto County are also supporting the project through fee-in-lieu agreements aimed at accelerating development timelines and reducing upfront costs.
Founded in 2023 by Elon Musk, xAI develops advanced artificial intelligence systems focused on large-scale reasoning and generative applications. Its flagship product, Grok, is integrated with the social media platform X, alongside a growing suite of APIs for image generation, voice, and autonomous agents, including offerings tailored for government use.
Elon Musk highlighted xAi’s growth and momentum in a comment about the matter. “xAI is scaling at an immeasurable pace — we are building our third massive data center in the greater Memphis area. MACROHARDRR pushes our Colossus training compute to ~2GW – by far the most powerful AI system on Earth. This is insane execution speed by xAI and the state of Mississippi. We are grateful to Governor Reeves for his support of building xAI at warp speed,” Musk said.
Tesla’s Head of AI, Ashok Elluswamy, says that something that was expected with version 14.3 of the company’s Full Self-Driving platform has already partially shipped with the current build of version 14.2.
Tesla and CEO Elon Musk have teased on several occasions that reasoning will be a big piece of future Full Self-Driving builds, helping bring forth the “sentient” narrative that the company has pushed for these more advanced FSD versions.
Back in October on the Q3 Earnings Call, Musk said:
“With reasoning, it’s literally going to think about which parking spot to pick. It’ll drop you off at the entrance of the store, then go find a parking spot. It’s going to spot empty spots much better than a human. It’s going to use reasoning to solve things.”
Musk said in the same month:
“By v14.3, your car will feel like it is sentient.”
Amazingly, Tesla Full Self-Driving v14.2.2.2, which is the most recent iteration released, is very close to this sentient feeling. However, there are more things that need to be improved, and logic appears to be in the future plans to help with decision-making in general, alongside other refinements and features.
On Thursday evening, Elluswamy revealed that some of the reasoning features have already been rolled out, confirming that it has been added to navigation route changes during construction, as well as with parking options.
He added that “more and more reasoning will ship in Q1.”
🚨 Tesla’s Ashok Elluswamy reveals Nav decisions when encountering construction and parking options contain “some elements of reasoning”
More uses of reasoning will be shipped later this quarter, a big tidbit of info as we wait v14.3 https://t.co/jty8llgsKM
— TESLARATI (@Teslarati) January 9, 2026
Interestingly, parking improvements were hinted at being added in the initial rollout of v14.2 several months ago. These had not rolled out to vehicles quite yet, as they were listed under the future improvements portion of the release notes, but it appears things have already started to make their way to cars in a limited fashion.
Tesla Full Self-Driving v14.2 – Full Review, the Good and the Bad
As reasoning is more involved in more of the Full Self-Driving suite, it is likely we will see cars make better decisions in terms of routing and navigation, which is a big complaint of many owners (including me).
Additionally, the operation as a whole should be smoother and more comfortable to owners, which is hard to believe considering how good it is already. Nevertheless, there are absolutely improvements that need to be made before Tesla can introduce completely unsupervised FSD.
Elon Musk
Tesla’s Elon Musk: 10 billion miles needed for safe Unsupervised FSD
As per the CEO, roughly 10 billion miles of training data are required due to reality’s “super long tail of complexity.”
Tesla CEO Elon Musk has provided an updated estimate for the training data needed to achieve truly safe unsupervised Full Self-Driving (FSD).
As per the CEO, roughly 10 billion miles of training data are required due to reality’s “super long tail of complexity.”
10 billion miles of training data
Musk comment came as a reply to Apple and Rivian alum Paul Beisel, who posted an analysis on X about the gap between tech demonstrations and real-world products. In his post, Beisel highlighted Tesla’s data-driven lead in autonomy, and he also argued that it would not be easy for rivals to become a legitimate competitor to FSD quickly.
“The notion that someone can ‘catch up’ to this problem primarily through simulation and limited on-road exposure strikes me as deeply naive. This is not a demo problem. It is a scale, data, and iteration problem— and Tesla is already far, far down that road while others are just getting started,” Beisel wrote.
Musk responded to Beisel’s post, stating that “Roughly 10 billion miles of training data is needed to achieve safe unsupervised self-driving. Reality has a super long tail of complexity.” This is quite interesting considering that in his Master Plan Part Deux, Elon Musk estimated that worldwide regulatory approval for autonomous driving would require around 6 billion miles.
FSD’s total training miles
As 2025 came to a close, Tesla community members observed that FSD was already nearing 7 billion miles driven, with over 2.5 billion miles being from inner city roads. The 7-billion-mile mark was passed just a few days later. This suggests that Tesla is likely the company today with the most training data for its autonomous driving program.
The difficulties of achieving autonomy were referenced by Elon Musk recently, when he commented on Nvidia’s Alpamayo program. As per Musk, “they will find that it’s easy to get to 99% and then super hard to solve the long tail of the distribution.” These sentiments were echoed by Tesla VP for AI software Ashok Elluswamy, who also noted on X that “the long tail is sooo long, that most people can’t grasp it.”
Elon Musk’s xAI bets $20B on Mississippi with 2GW AI data center project
Tesla AI Head says future FSD feature has already partially shipped
