News
SpaceX seeks approval for Starlink internet tests on high-performance govt. planes
According to updated regulatory documents and recent Aviation Week interviews with the US Air Force Research Laboratory, it can be all but guaranteed that the USAF has begun working with SpaceX to test the feasibility of using the company’s planned Starlink satellite internet constellation for military communications purposes.
In early August, SpaceX updated regulatory documents required by the Federal Communications Commission (FCC) for the company to be permitted to experimental test its two prototype Starlink internet satellites, named Tintin A and B. Launched roughly six months ago as a copassenger on one of SpaceX’s own Falcon 9 rockets, the satellite duo has been quietly performing a broad range of tests on orbit, particularly focused on general satellite operations, orbital maneuvering with SpaceX’s own custom-built electric propulsion, and – most importantly – the experimental satellites’ cutting-edge communications capabilities.
The orbit histories of @SpaceX's Tintin A/B Starlink prototype satellites, launched in February! Some thoroughly intriguing differences in behavior over the six months they've spent on-orbit. Data and visualizations generated by the lovely https://t.co/xKOdbP89tz. pic.twitter.com/a8CfQaZJep
— Eric Ralph (@13ericralph31) August 9, 2018
Per a public summary of the application update, SpaceX is awaiting FCC permission to begin a new series of tests of its prototype satellite internet network, this time highlighting an intriguing interest in applying Starlink connectivity to moving aircraft in order to provide them an exceptionally flexible and powerful suite of communications capabilities.
“SpaceX seeks to modify its experimental authorization to allow testing of two different antennas, both of which will operate on the ground and one of which will also operate from a moving aircraft.”
In those same documents, the company states that it is “working with a manufacturer of conformal antennas for tactical aircraft” to design and build “a custom installation kit consisting of mechanical plates for the low-profile antennas and fairings reducing wind drag”, seemingly indicating that SpaceX itself intends to supply the phased array antennae itself. Normally, this sort of testing would be fairly mundane and expected for any global satellite network, as one of the largest markets for satellite internet connectivity happens to be commercial aviation, particularly airlines and passenger entertainment.
-
- SpaceX’s first Starlink prototypes launched in late February aboard a flight-proven Falcon 9 booster. (Pauline Acalin)
-
- (SpaceX)
-
- One of the first two prototype Starlink satellites separates from Falcon 9’s upper stage in February 2018. (SpaceX)
However, the plot thickens beyond that extent thanks to a few select phrases. Most tellingly, the company writes that it “will perform a series of tests with the integrated airborne prototype terminal … varying motion for representative roll and pitch rates of a high-performance aircraft“, later also describing the program the update as a request for permission for “additional test activities undertaken with the federal government.”
It just so happens that the US Air Force’s Research Laboratory (AFRL) spoke with Aviation Week earlier this year (just weeks after SpaceX’s first prototype satellites had launched, in fact) about a nascent program exploring the potential utility of a spate of commercial Low Earth Orbit satellite internet constellations proposed for launch in recent years.
“To explore the art of the possible, AFRL is planning to contract with at least one commercial internet provider for a set of antennas that can be mounted onto Air Force test aircraft, Beal says. The team will then fly the aircraft … directly under the associated satellites and establish a communications path.” – Lara Seligman/AviationWeek
As of publishing then (March 2018) and now, SpaceX is the only company in the world to have launched a pair of functional demonstration satellites as a part of its proposed megaconstellation, meaning that it’s the only company that has a (technically) operational network with which they can test inter-satellite connectivity, connection hand-offs between different satellites, and multi-satellite operations.
While it’s currently unclear how that series of tests transpired and if they were or are officially connected to the AFRL’s own program, the briefest thought of the USAF (and thus the Department of Defence) as a prospective anchor customer for SpaceX’s Starlink constellation is extraordinarily exciting, especially given the apparent difficulties and costs associated with actually deploying even the first wave (~900 satellites) of such a massive constellation (~4500 satellites total).
Elon Musk dismissed reports claiming SpaceX had placed a massive order for NVIDIA GPUs worth $52 billion. The denial came hours after Taiwanese media, citing unnamed industry sources, reported that SpaceX planned to acquire approximately 13,000 AI server racks, equating to roughly 1 million GB300 GPUs, from Foxconn.
Each rack was estimated at around $4 million, with deliveries potentially starting in late 2025.
The story suggested this would mark SpaceX’s first major foray into Foxconn-manufactured NVIDIA hardware, breaking from suppliers like Supermicro and Dell. Musk responded bluntly on X:
This is fake news
— Elon Musk (@elonmusk) July 20, 2026
Despite the denial, the rumored scale aligns with SpaceX’s explosive growth in AI infrastructure. NVIDIA’s GB300 (successor to the GB200 NVL) racks deliver unprecedented performance for large-scale training and inference. A $52 billion commitment would dwarf most corporate AI budgets and provide the compute muscle needed for frontier models.
SpaceX already operates gigawatt-scale terrestrial clusters like Colossus in Memphis, Tennessee, and has monetized them aggressively through leasing deals.
SpaceX’s newest Starmind will make earth data centers obsolete
Major customers include Anthropic (paying ~$1.25 billion monthly for 220,000+ GPUs), Google (~$920 million monthly for 110,000 GPUs), and Reflection AI. These arrangements are projected to generate tens of billions in annual revenue, far outpacing traditional SpaceX businesses.
Such an investment would fuel internal AI efforts, particularly Grok models under the integrated SpaceXAI division, while supporting ambitious orbital data center plans. SpaceX envisions launching thousands of AI-optimized satellites powered by solar energy and cooled in space, bypassing terrestrial power and land constraints.
This “Starmind” constellation could position the company as a leader in space-based computing.
SpaceX as an Emerging AI Powerhouse
Once primarily known for reusable rockets and Starlink satellite internet, SpaceX has transformed into a multifaceted AI player.
The 2026 acquisition of xAI integrated Grok development directly into the company. Starlink’s low-latency global network complements massive compute clusters, enabling efficient data flow for training and serving AI models.
Musk has long argued that AI scaling demands solutions beyond Earth, citing things like real estate and electricity limits on the ground.
While the Foxconn deal may not be in the cards, SpaceX’s trajectory is continuing on the path of blending aerospace engineering with hyperscale AI to dominate both launches and intelligence infrastructure.
Elon Musk shed more details on the upcoming improvements to Tesla’s Full Self-Driving suite, specifically one that the CEO mentioned last week, which should help owners see fewer interventions.
Last week, Musk hinted that one major improvement that Tesla planned to roll out to Full Self-Driving users was the car’s ability “to remember your specific interventions and match each person’s individual preferences.”
Elon Musk says your Tesla will start to learn your individual preferences
This small bit of detail was linked to a post from Tesla community member Whole Mars, who said that FSD’s tendency to exit the carpool lane, a feature that owners can turn on but at times the car will disregard.
It sounds like, based on Musk’s two responses since that original post, it is safe to say the things FSD will start to remember are wide-ranging. However, it seems the biggest differences will be noticed with parking performance, which Musk continues to mention.
Highway Lane Preferences
The initial post Musk mentioned, with these new remembered preferences soon to arrive for Tesla owners everywhere, was the Carpool/Express Lane.
Tesla has a setting in the FSD menu that lets drivers enable HOV Lane travel. However, the car won’t always stay in that suggested or preferred lane.
The car will start to remember your specific interventions and match each person’s individual preferences
— Elon Musk (@elonmusk) July 18, 2026
Some owners have also complained of left lane camping, an illegal maneuver in at least some states. Cruising in the passing lane has resulted in tickets for some, as it is illegal in over 30 states in the U.S.
Tesla did not confirm if these preferences would also be included in new FSD behaviors, but it would certainly help move the company toward fewer interventions.
Parking Preferences
This seems to be the real focus of the entire operation, as Musk stated several weeks ago that parking was overwhelmingly the most frequent reason for interventions.
The major issue with parking is not necessarily the parking “performance,” as FSD is generally good at parking. It definitely has its issues; we’ve recorded plenty of them, including this one as recent as last week:
Yeah it seems like FSD v14.3.5 is having some issues with parking early on https://t.co/Bw5ULfVmDq pic.twitter.com/RHdpjOEpIo
— TESLARATI (@Teslarati) July 13, 2026
However, the changes coming are more about preferences, meaning where you park and how your car enters the spot, either pulling in or backing in. Owners have also reported that pulling into the correct driveway is a relatively rare thing for FSD, something else that needs to be confronted.
Musk basically confirmed that all of these things would be part of Tesla’s plan to address driver preferences with FSD:
Yes
— Elon Musk (@elonmusk) July 20, 2026
It’s obvious there is something big coming with FSD, and the company’s focus seems to be eliminating any intervention that would be related to preferences. This is probably the biggest bottleneck between Tesla and being fully autonomous. Critical interventions do occur, but they are much less frequent.
The only time a driver should be taking over is because of a critical intervention; this seems to be the goal of Tesla right now.
This all seems to be a priority as Tesla continues to move closer to the prospect of unsupervised driving.
Elon Musk said today on X that Teslas will start to learn your individual preferences. This is something that he seemed to hint toward earlier this month when he said parking was by far the biggest reason drivers intervene with Full Self-Driving.
Musk made the comment in response to notable Tesla influencer Whole Mars, who said that his vehicle will sometimes disobey the settings he has enabled for his car. He responded to the post, stating that “The car will start to remember your specific interventions and match each person’s individual preferences.”
The car will start to remember your specific interventions and match each person’s individual preferences
— Elon Musk (@elonmusk) July 18, 2026
This is something that could be perhaps one of the biggest ways Tesla could minimize or even work closer toward eliminating interventions altogether. While FSD does a lot of things really well, many people intervene a vast majority of the time not due to major or critical safety errors.
Instead, many take over because the car is doing something that they do not like as a preference; it might park in a parking spot that is not preferred by the driver, it might linger too long in the left lane on the highway (a personal favorite), or it could even take a route that the driver does not like.
These all lead to interventions, but they are not triggered by a major safety issue. Instead, it’s just preference.
READ OUR REVIEW OF TESLA’S LATEST FSD VERSION:
Tesla Full Self-Driving v14.3.5 Early Impressions: new features and early performance
If Teslas could start to learn the personal preferences of the person who owns them, interventions will truly begin to be less frequent. Some of this is already pretty evident, in my opinion. Teslas use a neural network to learn behaviors and accumulate data to improve performance.
For months now, we’ve tracked FSD’s performance at “Except Right Turn” stop signs, something that is very common in Pennsylvania, but many of our readers located in other parts of the U.S. have never heard of. FSD handles one Except Right Turn stop sign very well, one that I travel past frequently. Others that I do not navigate through as often do not have as confident a performance. It seems like the cars might already be doing this to an extent.
🚨 Tesla Full Self-Driving v14.3 proceeds through an Except Right Turn Stop Sign pic.twitter.com/YemRSlens7
— TESLARATI (@Teslarati) April 8, 2026
That example is also for something that is a street sign and not necessarily a driver preference; however, I still feel it is worth mentioning because it only handles that commonly passed Except Right Turn stop sign with true confidence. Others it still seems to struggle with.
This could be one of Tesla’s big moves toward full autonomy, and it could be a pathway to truly unsupervised driving. Every day, millions of cars on the road travel at a human driver’s personal preferences with no incident. Why can’t autonomous vehicles still cater to a passenger’s preferences while being autonomous? Tesla seems to have the idea that it would be possible.
Elon Musk dispels $52 billion SpaceX-NVIDIA GPU deal: ‘Fake news’
Elon Musk sheds details on Tesla FSD’s upcoming improvements
