News
SpaceX to mature Starship Moon landing and orbital refueling tech with NASA’s help
NASA has announced 19 technology partnerships between the agency’s many spaceflight centers and 13 companies, including SpaceX, Blue Origin, and more. This round of Space Act Agreements (SAAs) shows a heavy focus on technologies and concepts that could benefit exploration of the Moon and deep space more generally, including lunar landers, food production, reusable rockets, and more.
Put simply, all 19 awards are great and will hopefully result in tangible products and benefits, but SpaceX has a track record of achievement on the cutting edge of aerospace that simply has not been touched over the last decade. As such, the company’s two SAAs are some of the most interesting and telling, both ultimately focused on enabling Starship launches to and landings on the Moon and any number of other destinations in the solar system. Perhaps most importantly, it signals a small but growing sect within NASA that is willing and eager to acknowledge Starship’s existence and actively work with SpaceX to both bring it to life and further spaceflight technology in general.
One agreement focuses specifically on “vertically land[ing] large rockets on the Moon”, while the other more generally seeks to “advance technology needed to transfer propellant in orbit”, a feature that Starship’s utility would be crippled without. In this particular round of SAAs, they will be “non-reimbursable” – bureaucratic-speak for a collaboration where both sides pay their own way and no money is exchanged. SpaceX’s wins ultimately show that, although NASA proper all but refuses to acknowledge Starship, the many internal centers it is nothing without are increasingly happy to extend olive branches towards the company and its ambitious next-generation rocket.
“SpaceX of Hawthorne, California, will work with NASA’s Kennedy Space Center in Florida to advance their technology to vertically land large rockets on the Moon. This includes advancing models to assess engine plume interaction with lunar regolith.”
“SpaceX will work with Glenn and Marshall to advance technology needed to transfer propellant in orbit, an important step in the development of the company’s Starship space vehicle.”
NASA, July 30th, 2019
Giant rockets on the Moon
SpaceX’s first SAA centers around studying the task of landing Starship – a “large rocket” – on the Moon and attempting to understand just how the Moon’s powdery regolith (i.e. inorganic topsoil) will respond when subjected to the plume of a Raptor engine. Put simply, the task of landing a spacecraft as massive as Starship has never been attempted on the Moon, and the process itself – irrespective of any potential surprises from plume-regolith interaction – poses some obvious challenges.
In the most basic sense, Starship is massive. According to the vehicle’s circa. 2018 dimensions, it will stretch 55m (180 ft) from nose to tail, be 9m (30 ft) in diameter, and weigh (per 2017 specs) ~85 tons (190,000 lb) empty and upwards of ~1350 tons (2.95 million lbs) fully fueled. For reference, that is almost 80% as tall and more than 2.5 times as heavy as an entire Falcon 9 rocket. In the history of lunar exploration, Apollo’s Lunar Module (LM) – including landing and ascent stages – is the heaviest vehicle to have ever landed on the Moon, weighing a maximum of 5500 kg (12,100 lb) at landing (Apollo 17).
As such, an expendable Starship landing on the Moon with zero propellant for a possible return to Earth would easily break the record for landed mass by a factor of 10-20, while a Starship landing with enough delta V to simply return to lunar orbit – let alone land back on Earth – could easily up that to 30-50x.
Aside from the mass of Starship, there is also the question of how to gently land the spacecraft in the first place. Lunar gravity is roughly 1/6th of Earth’s, meaning that, say, 200 tons (i.e. Raptor’s thrust) would equate to more than 1200 tons of effective thrust on the Moon, a more than 10:1 thrust-to-weight ratio. For reference, the Apollo Lunar Module descent stage was powered by an engine with ~10,000 lbf (4.5 tons) of thrust that could throttle as low as ~1000 lbf (0.45 tons), meaning that even in lunar gravity conditions, the LM could have a thrust-to-weight ratio less than 1. For the purpose of safely landing on the Moon and ensuring a gentle landing, that is an extremely desirable thing to have.
Much like Falcon 9’s upper stage features cold-gas nitrogen thrusters to settle its propellant before MVac ignition, Starship will likely need a similar system, and it’s possible that that system could be used to gently land Starship and tweak its velocity in the final stages of a Moon landing. This study will likely be used in part to figure out what exactly the optimal method of landing Starship is.
How to Refuel Your Starship
Finally, SpaceX’s second NASA SAA focuses on developing the immature technology of in-orbit propellant transfer, an absolute necessity for Starship to simultaneously be fully reusable and capable of landing significant payloads on other planets (or moons). Ever since SpaceX CEO Elon Musk first revealed the company’s Mars-bound launch vehicle in 2016, it has incorporated in-orbit refueling as a foundational feature.
Due to the additions required for full reusability, Starship will essentially need to be launched into Earth orbit and then quickly refueled anywhere from 1 to 10+ times depending on the ultimate destination and the mass of the cargo being delivered. This is not to say that Starship will be useless without refueling – according to SpaceX VP of Sales Jonathan Hofeller, Starship will be capable of launching more than 100 tons (220,000 lb) to low Earth orbit and 20 tons (44,000 lb) to geostationary transfer orbit (GTO), more than enough to satisfy every commercial demand currently in existence.
However, with one or several refueling missions, Starship should be able to turn 100 tons to LEO into 100 tons to the surface of Mars or dozens of tons to the surface of the Moon. Put simply, with reliable and fast refueling, Starship goes from being a major step forward in reusable spaceflight to the key to the solar system and to radically affordable deep spaceflight.
Check out Teslarati’s Marketplace! We offer Tesla accessories, including for the Tesla Cybertruck and Tesla Model 3.
News
Tesla Full Self-Driving v14.2.2.5 might be the most confusing release ever
With each Full Self-Driving release, I am realistic. I know some things are going to get better, and I know some things will regress slightly. However, these instances of improvements are relatively mild, as are the regressions. Yet, this version has shown me that it contains extremes of both.
Tesla Full Self-Driving v14.2.2.5 hit my car back on Valentine’s Day, February 14, and since I’ve had it, it has become, in my opinion, the most confusing release I’ve ever had.
With each Full Self-Driving release, I am realistic. I know some things are going to get better, and I know some things will regress slightly. However, these instances of improvements are relatively mild, as are the regressions. Yet, this version has shown me that it contains extremes of both.
It has been about three weeks of driving on v14.2.2.5; I’ve used it for nearly every mile traveled since it hit my car. I’ve taken short trips of 10 minutes or less, I’ve taken medium trips of an hour or less, and I’ve taken longer trips that are over 100 miles per leg and are over two hours of driving time one way.
These are my thoughts on it thus far:
Speed Profiles Are a Mixed Bag
Speed Profiles are something Tesla seems to tinker with quite frequently, and each version tends to show a drastic difference in how each one behaves compared to the previous version.
I do a vast majority of my FSD travel using Standard and Hurry modes, although in bad weather, I will scale it back to Chill, and when it’s a congested city on a weekend or during rush hour, I’ll throw it into Mad Max so it takes what it needs.
Early on, Speed Profiles really felt great. This is one of those really subjective parts of the FSD where someone might think one mode travels too quickly, whereas another person might see the identical performance as too slow or just right.
To me, I would like to see more consistency from release to release on them, but overall, things are pretty good. There are no real complaints on my end, as I had with previous releases.
In a past release, Mad Max traveled under the speed limit quite frequently, and I only had that experience because Hurry was acting the same way. I’ve had no instances of that with v14.2.2.5.
Strange Turn Signal Behavior
This is the first Full Self-Driving version where I’ve had so many weird things happen with the turn signals.
Two things come to mind: Using a turn signal on a sharp turn, and ignoring the navigation while putting the wrong turn signal on. I’ve encountered both things on v14.2.2.5.
On my way to the Supercharger, I take a road that has one semi-sharp right-hand turn with a driveway entrance right at the beginning of the turn.
Only recently, with the introduction of v14.2.2.5, have I had FSD put on the right turn signal when going around this turn. It’s obviously a minor issue, but it still happens, and it’s not standard practice:
How can we get Full Self-Driving to stop these turn signals?
There’s no need to use one here; the straight path is a driveway, not a public road. The right turn signal here is unnecessary pic.twitter.com/7uLDHnqCfv
— TESLARATI (@Teslarati) February 28, 2026
When sharing this on X, I had Tesla fans (the ones who refuse to acknowledge that the company can make mistakes) tell me that it’s a “valid” behavior that would be taught to anyone who has been “professionally trained” to drive.
Apparently, if you complain about this turn signal, you are also claiming you know more than Tesla engineers…okay.
Nobody in their right mind has ever gone around a sharp turn when driving their car and put on a signal when continuing on the same road. You would put a left turn signal on to indicate you were turning into that driveway if that’s what your intention was.
Like I said, it’s a totally minor issue. However, it’s not really needed, and nor is it normal. If I were in the car with someone who was taking a simple turn on a road they were traveling, and they signaled because the turn was sharp, I’d be scratching my head.
I’ve also had three separate instances of the car completely ignoring the navigation and putting on a signal that is opposite to what the routing says. Really quite strange.
Parking Performance is Still Underwhelming
Parking has been a complaint of mine with FSD for a long time, so much so that it is pretty rare that I allow the vehicle to park itself. More often than not, it is because I want to pick a spot that is relatively isolated.
However, in the times I allow it to pull into a spot, it still does some pretty head-scratching things.
Recently, it tried to back into a spot that was ~60% covered in plowed snow. The snow was piled about six feet high in a Target parking lot.
A few days later, it tried backing into a spot where someone failed the universal litmus test of returning their shopping cart. Both choices were baffling and required me to manually move the car to a different portion of the lot.
I used Autopark on both occasions, and it did a great job of getting into the spot. I notice that the parking performance when I manually choose the spot is much better than when the car does the entire parking process, meaning choosing the spot and parking in it.
It’s Doing Things (For Me) It’s Never Done Before
Two things that FSD has never done before, at least for me, are slow down in School Zones and avoid deer. The first is something I usually take over manually, and the second I surprisingly have not had to deal with yet.
I had my Tesla slow down at a school zone yesterday for the first time, traveling at 20 MPH and not 15 MPH as the sign suggested, but at the speed of other cars in the School Zone. This was impressive and the first time I experienced it.
I would like to see this more consistently, and I think School Zones should be one of those areas where, no matter what, FSD will only travel the speed limit.
Last night, FSD v14.2.2.5 recognized a deer in a roadside field and slowed down for it:
🚨 Cruising home on a rainy, foggy evening and my Tesla on Full Self-Driving begins to slow down suddenly
FSD just wanted Mr. Deer to make it home to his deer family ❤️ pic.twitter.com/cAeqVDgXo5
— TESLARATI (@Teslarati) March 4, 2026
Navigation Still SUCKS
Navigation will be a complaint until Tesla proves it can fix it. For now, it’s just terrible.
It still has not figured out how to leave my neighborhood. I give it the opportunity to prove me wrong each time I leave my house, and it just can’t do it.
It always tries to go out of the primary entrance/exit of the neighborhood when the route needs to take me left, even though that exit is a right turn only. I always leave a voice prompt for Tesla about it.
It still picks incredibly baffling routes for simple navigation. It’s the one thing I still really want Tesla to fix.
Investor's Corner
Tesla gets tip of the hat from major Wall Street firm on self-driving prowess
“Tesla is at the forefront of autonomous driving, supported by a camera-only approach that is technically harder but much cheaper than the multi-sensor systems widely used in the industry. This strategy should allow Tesla to scale more profitably compared to Robotaxi competitors, helped by a growing data engine from its existing fleet,” BoA wrote.
Tesla received a tip of the hat from major Wall Street firm Bank of America on Wednesday, as it reinitiated coverage on Tesla shares with a bullish stance that comes with a ‘Buy’ rating and a $460 price target.
In a new note that marks a sharp reversal from its neutral position earlier in 2025, the bank declared Tesla’s Full Self-Driving (FSD) technology the “leading consumer autonomy solution.”
Analysts highlighted Tesla’s camera-only architecture, known as Tesla Vision, as a strategic masterstroke. While technically more challenging than the multi-sensor setups favored by rivals, the vision-based approach is dramatically cheaper to produce and maintain.
This cost edge, combined with Tesla’s rapidly expanding real-world data engine, positions the company to scale robotaxis far more profitably than competitors, BofA argues in the new note:
“Tesla is at the forefront of autonomous driving, supported by a camera-only approach that is technically harder but much cheaper than the multi-sensor systems widely used in the industry. This strategy should allow Tesla to scale more profitably compared to Robotaxi competitors, helped by a growing data engine from its existing fleet.”
The bank now attributes roughly 52% of Tesla’s total valuation to its Robotaxi ambitions. It also flagged meaningful upside from the Optimus humanoid robot program and the fast-growing energy storage business, suggesting the auto segment’s recent headwinds, including expired incentives, are being eclipsed by these higher-margin opportunities.
Tesla’s own data underscores exactly why Wall Street is waking up to FSD’s potential. According to Tesla’s official safety reporting page, the FSD Supervised fleet has now surpassed 8.4 billion cumulative miles driven.
Tesla FSD (Supervised) fleet passes 8.4 billion cumulative miles
That total ballooned from just 6 million miles in 2021 to 80 million in 2022, 670 million in 2023, 2.25 billion in 2024, and a staggering 4.25 billion in 2025 alone. In the first 50 days of 2026, owners added another 1 billion miles — averaging more than 20 million miles per day.
This avalanche of real-world, camera-captured footage, much of it on complex city streets, gives Tesla an unmatched training dataset. Every mile feeds its neural networks, accelerating improvement cycles that lidar-dependent rivals simply cannot match at scale.
Tesla owners themselves will tell you the suite gets better with every release, bringing new features and improvements to its self-driving project.
The $460 target implies roughly 15 percent upside from recent trading levels around $400. While regulatory and safety hurdles remain, BofA’s endorsement signals growing institutional conviction that Tesla’s data advantage is not hype; it’s a tangible moat already delivering billions of miles of proof.
News
Tesla to discuss expansion of Samsung AI6 production plans: report
Tesla has reportedly requested an additional 24,000 wafers per month, which would bring total production capacity to around 40,000 wafers if finalized.
Tesla is reportedly discussing an expansion of its next-generation AI chip supply deal with Samsung Electronics.
As per a report from Korean industry outlet The Elec, Tesla purchasing executives are reportedly scheduled to meet Samsung officials this week to negotiate additional production volume for the company’s upcoming AI6 chip.
Industry sources cited in the report stated that Tesla is pushing to increase the production volume of its AI6 chip, which will be manufactured using Samsung’s 2-nanometer process.
Tesla previously signed a long-term foundry agreement with Samsung covering AI6 production through December 31, 2033. The deal was reportedly valued at about 22.8 trillion won (roughly $16–17 billion).
Under the existing agreement, Tesla secured approximately 16,000 wafers per month from the facility. The company has reportedly requested an additional 24,000 wafers per month, which would bring total production capacity to around 40,000 wafers if finalized.
Tesla purchasing executives are expected to discuss detailed supply terms during their visit to Samsung this week.
The AI6 chip is expected to support several Tesla technologies. Industry sources stated that the chip could be used for the company’s Full Self-Driving system, the Optimus humanoid robot, and Tesla’s internal AI data centers.
The report also indicated that AI6 clusters could replace the role previously planned for Tesla’s Dojo AI supercomputer. Instead of a single system, multiple AI6 chips would be combined into server-level clusters.
Tesla’s semiconductor collaboration with Samsung dates back several years. Samsung participated in the design of Tesla’s HW3 (AI3) chip and manufactured it using a 14-nanometer process. The HW4 chip currently used in Tesla vehicles was also produced by Samsung using a 5-nanometer node.
Tesla previously planned to split production of its AI5 chip between Samsung and TSMC. However, the company reportedly chose Samsung as the primary partner for the newer AI6 chip.
Tesla Full Self-Driving v14.2.2.5 might be the most confusing release ever
Tesla gets tip of the hat from major Wall Street firm on self-driving prowess
