Tesorb Signal: Tesla, SpaceX, and the Musk Companies

Six weeks ago, a former Nasdaq chief warned that hundreds of billions of dollars in locked SpaceX stock was about to hit the market. On August sixth, the first hundred billion dollars of it became legal to sell.

Six weeks ago, a former Nasdaq chief warned that hundreds of billions of dollars in locked SpaceX stock was about to hit the market. On August sixth, the first hundred billion dollars of it became legal to sell. The stock went up six percent that day. The flood everyone braced for turned out to be a question mark.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. What actually happened at the first two unlocks. Why shares that are free to sell mostly still can’t be sold. And the calendar ahead, because the biggest gates haven’t opened yet.

Start with the scorecard. When SpaceX went public in June, only about five percent of the company could trade. Everything else sat behind a staggered lockup, a series of scheduled release dates instead of one big cliff. The first gate was tied to earnings. Two days after SpaceX reported its first quarter as a public company, nine hundred eleven million shares came free.

Worth about a hundred billion dollars.

The setup looked ugly. The stock had just touched an all-time low, down more than half from its June peak, sitting well below the hundred thirty-five dollar IPO price. Short sellers had piled in. And then the flood didn’t come. No wave of insider selling showed up. The stock rose about six percent that day and kept climbing, back above the IPO price within a week.

Then came the second gate, on August twentieth. Three hundred nineteen million shares this time, about a third the size of the first. The stock fell roughly five percent and dipped back below the IPO price. A smaller unlock produced a bigger drop. That tells you the market isn’t reacting to share counts. It’s reacting to mood.

Two fairness points. First, eligible does not mean sold. Analysts at Mizuho made this exact point in a client note. A lockup expiring grants the ability to sell, not an obligation. There’s precedent. When Facebook’s biggest lockup expired in twenty twelve, investors braced for a crash, insiders mostly held, and the stock jumped almost thirteen percent in a day. Relief is a market force too.

Second, the staggered design is doing its job. Instead of one cliff where everything hits at once, SpaceX built more than a dozen scheduled gates spread into next year. One bonus tranche never even opened, because it required the stock to trade thirty percent above the IPO price, and it didn’t. Those shares roll forward to December. The structure absorbs shocks by rationing them.

And then there’s the part almost nobody priced in. A detailed analysis in Forbes walked through the securities rules, and the punchline is that most newly eligible shares still can’t legally move. Insiders selling under a trading plan face a mandatory cooling-off period of at least thirty days. Officers and directors wait about ninety.

Company trading windows only open for a few weeks each quarter.

So the vault has a second door. August sixth made shares eligible. The rules decide when they’re actually sellable, and for many holders that’s weeks or months later. Historically, the real volume from lockup expirations shows up well after the expiration date. If you judged this event by one Thursday, you judged it early.

Who’s on each side of the door is worth a look. The eligible sellers are early employees and early money. One of them is Jessie Bates, an NFL safety who put a hundred fifty thousand dollars in four years ago, watched it grow about ten fold, and told CNBC he’s selling all of it.

On the other side, filings this month showed more than fifteen hundred institutions holding positions, with Alphabet the biggest at over five hundred million shares.

Think of it like a stadium with more than a dozen exit gates that open one at a time. Everyone inside is free to leave when their gate opens. So far, most of the crowd is staying for the second half, and season-ticket buyers keep showing up outside. The question was never whether the gates would open. It’s whether anyone actually walks.

Now the calendar, because the gates get bigger. Another three hundred nineteen million shares come free in early September, with more releases through October. Around third quarter earnings in November, a tranche of roughly one point three billion shares arrives, four times the size of the August twentieth batch. The full one hundred eighty day lockup expires December eighth.

By then, about forty percent of the company can trade.

And one date stands alone. Elon Musk’s six point four billion shares, nearly half the company, stay locked until June of next year. The single biggest holder can’t sell a share until then. Every unlock between now and next summer is a preview. That one is the feature.

There’s also new supply arriving through a side door. When SpaceX closed its Cursor acquisition in mid August, it paid in stock, issuing about three hundred ninety million new shares. That’s not a lockup expiring. That’s the share count growing, the same arithmetic we’ve walked through with Tesla’s dilution. Different mechanism, same pressure.

So the story isn’t who can sell. It’s who does.

A hundred billion dollars came unlocked and the stock rose. A third of that came unlocked two weeks later and the stock fell. Same company, same structure, opposite reactions. The supply was never the signal. Watch November and December, when the biggest gates open into the same quarter. If the crowd stays seated through those, the vault was never the story.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

Last Tuesday, a Swedish freight company ordered five hundred Tesla Semis. That’s the largest order the truck has ever received, and it beat the previous record by a hundred and thirty trucks.

Last Tuesday, a Swedish freight company ordered five hundred Tesla Semis. That’s the largest order the truck has ever received, and it beat the previous record by a hundred and thirty trucks. Two days later, Tesla said the Semi is coming to Europe. The order is real. The map is the question.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. What the Einride order actually commits, and who’s paying for it. Why Tesla picked Hannover to announce Europe. And why the truck that took nine years to build now faces a market that didn’t wait.

Start with the buyer. Einride runs electric freight networks for big shippers, with Amazon as its named customer. It went public on the Nasdaq in June. On August eighteenth, alongside its first-half results, it announced the Tesla order, delivered in phases over twenty-four months starting in September, across California, Texas, New Jersey, Illinois, and Georgia.

Einride has about two hundred fifty electric trucks today. This order triples the fleet.

Two fairness points. First, this is a real order from a real operator, not a reservation. Einride’s CEO called it proof the company can execute at the scale customers demand. Tesla’s Semi director, Dan Priestley, said the trucks offer lower cost per mile than diesel through fuel and maintenance savings. Deliveries start next month.

After nine years of Semi announcements, a dated delivery schedule with a public company on the other end is a different kind of news.

Second, the factory behind it exists. Tesla rolled the first high-volume Semi off the line at its dedicated Nevada plant on April twenty-ninth. The building is designed for fifty thousand trucks a year. Pricing is public too, at about three hundred thousand dollars for the long-range version. In May, a California port operator ordered hundreds. The Semi has stopped being a pilot.

Now the financing, because Electrek did the work here and it’s worth passing on. Einride closed June with about seventy-seven million dollars in cash. The order at Tesla’s prices is a commitment well north of a hundred million. Einride’s CFO told analysts the purchase is funded by a four-year asset-backed loan at an effective rate of about fourteen percent. That’s not a red flag on its own.

Truck financing is a mature business, and Einride’s model is built on shippers not carrying the capital. But fourteen percent is expensive money, and the deployment is paced over two years for a reason.

Then, on August twentieth, Tesla’s Semi account posted one line. Specs and launch details for Semi in Europe to be unveiled at IAA Transportation in Hannover. That’s the continent’s biggest commercial vehicle show, running September fifteenth to twentieth. Tesla stock rose about three percent on the post before a Nevada robotaxi permit carried it higher.

Why Hannover? Because Tesla was there two years ago with a single truck shipped from America and almost nothing to say. Priestley told the show in twenty twenty-four that a European launch wouldn’t happen before twenty twenty-six, and only after U.S. production scaled. U.S. production scaled in April. He also mentioned a shorter prototype built for European length rules and trailer couplings.

Hannover is Tesla closing a loop it opened on the same stage.

And that’s where the map gets hard. Europe did not wait. Mercedes-Benz’s electric long-haul truck is on the road in fifteen countries. Volvo is the region’s electric heavy-truck sales leader and showed a new generation in April. MAN, Scania, Renault, and DAF all have battery-electric tractors in production or delivery, with dealer networks, service bays, and certification already done.

Tesla has a business development manager in France and a truck in a convention hall.

The regulation cuts both ways. European rules require truckmakers to cut new heavy-truck emissions fifteen percent from twenty nineteen levels now, rising to forty-five percent by twenty thirty, with penalties for missing. That creates demand for every electric truck, including Tesla’s. It also means the domestic makers have been selling into that demand for years.

Tesla arrives with a cost and range advantage on paper, and no service network on the ground.

Think of it like a strong American college team getting invited to play in a European league. The athletes are real. The stats travel. But the league has its own rules, its own referees, and teams that have been playing each other for a decade. The first season is about learning the field, not winning it.

The Semi has the longest promise ledger of any Tesla product. Unveiled in twenty seventeen. Promised for twenty nineteen. Volume production in twenty twenty-six. We walked that timeline earlier this summer. What’s different now is that the ledger has a customer with a delivery date, and a factory with a ramp.

What’s the same is that the next promise, Europe, comes with a show date and not a sale date.

So the truck is real. The map is the question.

Five hundred trucks is the number that matters this month, and it’s financed at fourteen percent by a company that just tripled its fleet on Tesla’s word. Watch September for two things. Whether the first Einride trucks actually roll in the first week. And whether Hannover produces a price in euros and an order date, or another truck in a hall with a promise attached.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

On Thursday a financial commentator posted a chart of Wall Street’s five-year revenue forecasts for the biggest companies in tech. Nvidia, two hundred eighty-eight percent.

On Thursday a financial commentator posted a chart of Wall Street’s five-year revenue forecasts for the biggest companies in tech. Nvidia, two hundred eighty-eight percent. Apple, fifty-four percent. SpaceX, two thousand ninety percent. Elon Musk replied that this sounds crazy, and that both SpaceX and Tesla will beat it.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. What those percentages actually require when you turn them into dollars. Why the Tesla half of the claim is the harder one. And what Musk’s own track record on revenue targets says about the word crazy.

Start with the arithmetic, because percentages hide the size of the ask. The SpaceX forecast means revenue ends about twenty-two times where it started. To get there, a company has to roughly double every thirteen months, five years running. SpaceX’s revenue last year was in the fifteen to sixteen billion dollar range by Musk’s own estimate.

Twenty-two times that is somewhere north of three hundred billion dollars a year.

For scale, that’s roughly what Apple makes today. Analysts are modeling SpaceX, a company that reported a net loss last quarter, growing into an Apple-sized revenue line in five years. And Musk’s reply is that the estimate is too low.

Two fairness points. First, the SpaceX number isn’t invented. It has a mechanism. Starlink added one point seven million net subscribers last quarter, and the company says it’s on pace for a hundred billion dollars in annualized revenue by the end of this year. If that holds, a big piece of the twenty-two times is already in the bag before twenty twenty-seven starts.

Analysts are extrapolating a curve that’s visible in the filings, not drawing one from scratch.

Second, Musk has hit absurd-sounding targets before. In twenty fourteen Tesla made about three billion dollars in revenue. Ten years later it made ninety-seven billion. SpaceX went from a thirty-six billion dollar valuation in early twenty twenty to a public company worth over a trillion. When he says crazy, he’s pointing at a history where crazy sometimes happened.

Now the harder half. The Tesla forecast on that chart is a hundred nineteen percent. That sounds modest next to the SpaceX line, but look at what it requires. Tesla’s revenue last year was about ninety-five billion dollars. Doubling it means about two hundred billion by the end of the window. Over five years that’s roughly seventeen percent growth a year, every year.

Tesla’s revenue grew one percent in twenty twenty-four. The second quarter of this year was a record on the top line, but earnings per share fell more than forty percent, and the stock is down about twenty-three percent for the year. So the question isn’t whether Tesla can grow seventeen percent a year. It did far more than that for a decade. The question is what grows. Cars aren’t doing it.

The analysts’ number assumes robotaxi, Optimus, and energy storage fill the gap. Musk’s number assumes they overfill it.

Think of it like a runner who used to win every race, then spent two seasons jogging, and now says she’ll beat the course record. It’s not impossible. She has the history. But the claim is about the next five seasons, and the last two are the only recent evidence.

And there’s a cost to the curve that nobody put on the chart. SpaceX lost money in both quarters this year while spending more than fifteen billion dollars on AI infrastructure in the second quarter alone.

Revenue growing twenty-two times is the forecast. Spending growing alongside it is the filing. A company can hit a revenue target and still not make money doing it, and the chart measured only one of those.

Which brings us to the track record on this specific kind of statement. Musk reportedly told Twitter’s backers the platform would reach twenty-six billion dollars in revenue. It didn’t get close. He said Tesla would have a million robotaxis on the road by twenty twenty. We’ve done that ledger. He said Grok 5 would ship last year, and it hasn’t.

Revenue predictions are the category where his record is weakest, because revenue depends on customers, and customers don’t respond to deadlines.

There’s also the reason this post exists. SpaceX stock spent weeks below its IPO price this summer and only climbed back above it on August tenth. Tesla is down for the year. A public reply saying analysts are too conservative costs nothing and moves both tickers the next morning.

Tesla closed up five percent on Friday, helped by a Nevada robotaxi permit, and SpaceX has been rebounding alongside it. Whether the claim is right is a question for the next decade. Whether it worked was settled by lunch.

So do the arithmetic on crazy.

Two thousand ninety percent means SpaceX becomes an Apple-sized business in five years, and Musk says that’s low. A hundred nineteen percent means Tesla finds a second company’s worth of revenue from products that barely exist today, and Musk says that’s low too. One of those has a curve you can see in the filings. The other has a curve you can see in the promises.

When the next earnings report lands, check which one is actually bending.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

Since last summer, Elon Musk has given four different arrival windows for Grok 5, xAI’s next flagship model. End of twenty twenty-five.

Since last summer, Elon Musk has given four different arrival windows for Grok 5, xAI’s next flagship model. End of twenty twenty-five. Early twenty twenty-six. The first quarter. The second quarter. All four have passed. In that same stretch, xAI shipped six smaller Grok releases. The headline says xAI ships faster than anyone. The data says the one model that matters keeps moving.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. The Grok 5 promise ledger, date by date. What xAI shipped instead, and why it’s not nothing. And what the latest date tells you about the merger that now owns Grok.

Here’s the ledger. August twenty twenty-five, Grok 4 is a month old, and Musk says Grok 5 arrives before the end of the year and will be crushingly good. November, he pushes it to early twenty twenty-six and says there’s a ten percent chance it reaches human-level intelligence. January, xAI closes a twenty billion dollar funding round and confirms the model is still training.

The first quarter ends. The company’s own account points to the second quarter. The second quarter ends. On SpaceX’s earnings call on August fourth, Musk puts it at year end.

That’s a year of dates, and each one was stated as a plan, not a hope. The pattern isn’t subtle. Every time a date approaches, it moves about a quarter.

Two fairness points. First, training a frontier model at this scale is slow and hard for everyone. Grok 5 is reported to be a six trillion parameter model training on Colossus 2 in Memphis. No lab has shipped a model that size on a clean schedule. OpenAI and Google have slipped flagship dates too. Musk’s dates are worse than the industry’s, but the industry’s aren’t great.

Second, what xAI shipped instead has real users and real numbers. Grok 4.5 landed July eighth at two dollars per million input tokens, with a five hundred thousand token context window, and placed fourth on an independent intelligence index. Grok 4.6 followed August twelfth at the same price. SpaceX stock rallied on the news.

Developers can use these models today inside Cursor, the code editor SpaceX bought. Shipping a usable product every few weeks is not a consolation prize.

So why does the flagship matter if the point releases are good? Because the point releases are refinements of the same foundation. Grok 4.5 and four point six run on a one and a half trillion parameter base that xAI calls V9. The 4.7 model, promised a few weeks after 4.6, is a bigger base at about two trillion.

Grok 5 is the step change, the model Musk has attached the words human-level intelligence to. That’s the one the funding round, the capital spending, and the stock story are built on.

And capital spending is the part the merger changed. When xAI was private, a slipped model was a story for the AI press. Now Grok is a division of a public company. In SpaceX’s first earnings report, AI capital expenditure came to fifteen point eight billion dollars in one quarter, more than double the quarter before and well above what analysts expected.

That money is buying compute for Grok 5. Every quarter the model slips, that spending shows up in the filings with nothing to point to.

Think of it like a construction company that keeps announcing the skyscraper is topping out next quarter, while it builds a series of very nice four-story buildings on the same lot. The four-story buildings have tenants. The rent is real. But the loan was sized for the skyscraper, and the lender is reading the calendar.

There’s a newer wrinkle in the latest date. On the earnings call, Musk said Grok 5 is being trained on basically all the data SpaceX has ever generated. That’s engineering data, telemetry, internal documents. It’s a pitch for why the merger makes the model better. It’s also a reason the schedule is hard to pin down, because that’s a training run nobody has done before.

The more novel the claim, the softer the date.

And 4.7 is already drifting. Musk said four weeks on July twenty-fourth, then a few weeks, then three to four weeks on August fourth. Late August became early-to-mid September. Same pattern, smaller scale. Watch whether 4.7 ships before the Grok 5 date moves again. If it does, the cadence story holds.

If it doesn’t, the cadence story starts to look like the flagship story.

We’ve walked this ledger format before with the robotaxi promises. The Grok version has the same shape. Many specific dates, a lot of visible activity, and the headline deliverable always one quarter out.

So read the version numbers as a calendar.

Four windows, four misses, six releases in between. Grok 4.7 is the next test, and it’s due in a matter of weeks. Grok 5 is now a year-end promise from a company that has spent fifteen billion dollars in a quarter getting ready for it. The question isn’t whether xAI can ship. It can. The question is whether the model everyone is paying for arrives on the fifth date, or the sixth.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

On Friday, China’s market regulator published the largest recall in the history of the world’s biggest electric car market. Nearly three million Teslas.

On Friday, China’s market regulator published the largest recall in the history of the world’s biggest electric car market. Nearly three million Teslas. The fix is a sticker and a software update. The problem is a door handle that Tesla taught the whole industry to copy.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. What the recall actually fixes, and what it doesn’t. Why eight Chinese automakers got pulled in with Tesla. And what a door handle says about who sets the rules in the EV market now.

Start with the mechanics, because the word recall is doing a lot of work. Tesla’s flush door handles sit level with the body and pop out on an electrical pulse. Inside, the regular handle is electronic too. The mechanical backup release is built into the door panel, and in the Model 3 and Model Y it blends into the trim. In a severe crash, the car’s low-voltage system can fail.

When it does, the electronic handles die, and occupants and rescuers have to find a release most of them have never touched.

China’s regulator said that release is difficult to distinguish and operate. The remedy comes in two parts, starting September twenty-fifth. A warning label at the emergency release point, and an over-the-air update that automatically lowers the windows after a collision. No service appointment for most owners.

The recall covers Model 3, Model Y, Model S, and Model X, built in Shanghai or imported, going back to twenty nineteen.

There’s a second recall in the same announcement that got less attention. Most of the same China-built Model 3s and Model Ys need a software change to the driver-monitoring system. Today it relies on steering-wheel torque to decide if you’re paying attention. The update adds in-cabin camera tracking.

According to CleanTechnica, some drivers had been defeating the old system by mounting a doll’s head in front of the camera.

Two fairness points. First, Tesla didn’t get singled out. The regulator pulled in nine automakers in one action. Xiaomi recalled hundreds of thousands of its SU7 sedans. Leapmotor, Xpeng, Geely’s brands, Chery, and others followed. More than four million vehicles in total. Tesla has the biggest share because Tesla has sold the most cars with this design.

That’s a volume story, not a defect story.

Second, the fix matches the risk. A label and a window-drop update address the actual failure mode, which is people not finding the release in the dark after a crash. This is not a structural recall. It’s cheap for Tesla and fast for owners, and the regulator accepted it. By the standard of recalls, this is about as light as they come.

So why does it matter? Because of where the rule came from, and where it’s going. Last year a driver in China died trapped in a burning car when bystanders couldn’t open the doors. That incident, and others like it, pushed the regulator to finalize a ban on flush, electronic-only handles for new models starting in twenty twenty-seven.

New cars will need a recessed mechanical handle of a specified size, and doors on the non-impact side must open without tools after a crash or a battery fire.

Tesla popularized the flush handle with the Model S more than a decade ago. It became shorthand for a modern EV. Xiaomi copied it. Xpeng copied it. Half the Chinese startups copied it. And now the market that buys more EVs than anywhere else has decided the design is a safety problem and written the replacement spec. Tesla has said it’s working on a redesign. It didn’t have a choice.

Think of it like a fashion house that invented a silhouette everyone imitated, and then watched the biggest retailer in the world ban the silhouette from its shelves. The original designer gets the most attention. But the real change is that someone else now decides what goes on the rack.

That’s the pattern to watch. Chinese regulators have been setting standards for driver assistance, for batteries, and now for doors, at a pace that matches how fast their automakers build cars. Tesla’s China sales are already under pressure. The Model Y sold about twenty-five thousand units there in July, down eighteen percent from a year earlier, with BYD and Xiaomi underneath it on price.

A recall this size lands as a headline Tesla didn’t need in a market it can’t afford to lose.

And it won’t stay in China. American regulators have already opened an inquiry into Tesla’s emergency door releases, and on Friday the U.S. safety agency said the automakers haven’t disclosed plans for a similar recall at home. If the Chinese spec becomes the global spec, the next redesign of every Tesla door starts in Beijing, not Fremont.

So the handle is the story, not the recall.

Nearly three million cars is the number, and the fix is a sticker and a software push. The bigger shift is quieter. The country that Tesla taught to build electric cars just wrote a rule that Tesla has to follow, on a design Tesla invented. Watch the twenty twenty-seven standard. Whatever replaces the flush handle will show up on the next Model Y, and it won’t have been Tesla’s idea.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

Three days ago, a federal hearing that could have shut down the power plant behind the Pentagon’s favorite chatbot was postponed. State records show sixty-nine gas turbines at the site.

Three days ago, a federal hearing that could have shut down the power plant behind the Pentagon’s favorite chatbot was postponed. State records show sixty-nine gas turbines at the site. The number of air permits for them is zero. An agreed order with the state lets them keep running until July of next year. The question is still the question. Can the AI buildout move faster than the permit system, or can’t it.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. How a chatbot ended up with its own unpermitted power plant. Why the Justice Department stepped in on the company’s side, with an argument nobody saw coming. And what the postponement tells you about every company racing to build AI, not just this one.

The site sits in Southaven, Mississippi, just across the state line from Memphis. The turbines there generate roughly five hundred megawatts, and the power feeds Colossus 2, the data center that trains Grok. When the NAACP sued in April, the count was twenty-seven turbines. By May, it had grown to forty-six. By the time the state’s own records were current, sixty-nine.

The company got sued for the turbines and responded by adding turbines.

The legal fight turns on a single word. Mobile. The turbines are mounted on trailers, and the company and Mississippi’s environmental regulator say that makes them mobile sources, exempt from the air permits that a power plant would need. The plaintiffs point out that once installed, each unit is nearly a hundred feet long and weighs over two hundred thousand pounds.

In January, the federal environmental agency ruled that turbines like these do need permits. The lawsuit is where those two readings collide.

Then in June, the case stopped being a normal pollution suit. The Justice Department moved to intervene on the company’s side and asked the court to toss the case entirely. Its argument wasn’t about air. It was national security.

A Pentagon official told the court that Grok is one of only four frontier AI models capable of supporting national security applications, and that it’s woven into military targeting systems.

The declaration went further than anyone expected. It said Grok’s integration helped forces deploy two thousand munitions against two thousand targets within ninety-six hours during recent operations. Set aside how you feel about that sentence. As a legal move, it reframes the whole case. Shutting down the turbines, the government argues, doesn’t just pause a chatbot.

It degrades a defense capability.

Two fairness points. First, the company has a compliance record worth noting, because it cuts in its favor. At the first Colossus site in Memphis, after the same legal groups threatened to sue, the company removed its unpermitted turbines and got permits for the fifteen it kept. There’s a documented path where pressure leads to permits, not shutdowns.

The company says it follows the law, and Mississippi’s own regulator has so far agreed with its reading.

Second, the underlying problem is real and bigger than one company. AI training clusters now need power at a scale utilities cannot deliver on the timelines the industry wants. Every operator faces the same choice. Wait years for grid upgrades, or generate on site and sort the framework out later. This company chose the second path more aggressively than anyone.

But the pressure that created the choice is industry-wide.

In late July, Mississippi’s regulator and the company signed an agreed order. The sixty-nine turbines can keep running, but they have to come out by July 2027. The replacement is a permanent 1.2 gigawatt plant with proper permits, already under construction. Environmental groups called the agreement a capitulation. Congress has opened its own inquiry, demanding records from the company.

The other side of the ledger is why the case exists. The plaintiffs say the turbines can emit more than seventeen hundred tons of smog-forming pollution a year, plus formaldehyde, in an area that already ranks among the nation’s worst for asthma. The plant sits about half a mile from homes and a mile from an elementary school. Residents have filed a separate suit over the constant noise.

Whatever happens in court, those neighbors are the reason a court is involved at all.

And the money makes the stakes concrete. SpaceX disclosed fifteen point eight billion dollars of AI capital spending in a single quarter, and Grok 5 is training on this power right now. If the hearing leads to an injunction, the flagship model everyone is waiting on loses part of its power supply mid-run.

If it doesn’t, the message to every data center builder in America is that trailer-mounted generation is a workable bridge, at least until the appeals run out.

Think of it like a food truck parked in the same spot for a year, running a full restaurant off a generator, while arguing it doesn’t need a restaurant license because the truck has wheels. The city inspector says the wheels are what counts. The neighbors say look at the dining room.

And now the mayor has shown up saying the truck feeds the police department, so the case isn’t really about the truck anymore.

So the ruling isn’t about turbines. It’s about runway.

Sixty-nine turbines, zero permits, one word in a statute, and a national security argument sitting on top of all of it. The hearing that was supposed to answer the question this week got pushed. The turbines keep running. An agreed order says they come out next summer, but the permanent plant going up beside them is twice the size. Watch for the new hearing date. And watch the turbine count. That number has only ever moved in one direction.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

On Thursday, Nevada regulators cleared Tesla to run up to five thousand driverless cars around Las Vegas. Across the entire country, Tesla’s unsupervised robotaxi fleet at any one moment is about twenty cars.

On Thursday, Nevada regulators cleared Tesla to run up to five thousand driverless cars around Las Vegas. Across the entire country, Tesla’s unsupervised robotaxi fleet at any one moment is about twenty cars. Five thousand is the headline. Twenty is the data.

This is Tesorb Signal, and I’m Lena Ruiz. Three things today. What Nevada actually approved, and why the number jumped five hundred fold in a month. What Tesla said to the regulators once the cameras were rolling. And why the gap between the ceiling and the floor is the whole story.

Start with the permit. Back in June, Tesla applied to Nevada’s Transportation Authority to operate a driverless network across Clark County. That’s the county that holds Las Vegas and the airport. In late July, the regulator came back with an interim order. Ten cars. A corridor along the Strip. A speed cap of forty-five miles an hour. No airport pickups.

That interim order read like a rebuke, and Electrek, Fortune, and Axios all covered it that way. Tesla asked for thousands and got ten. Then on Thursday, August twentieth, the full permit came through. The ten became the full request. The corridor became the whole county. The airport ban lifted, pending the airport’s own sign-off. Tesla stock jumped five percent the next day.

Two fairness points before we pull at this. First, the interim order was always interim. The regulator’s spokesperson said as much to Fortune. Ten cars was the proof-of-concept phase, not the verdict. Tesla wasn’t singled out. Waymo got a thousand on the same day, Uber got a thousand, and Zoox already holds a hundred-car permit. Nevada just decided to let everyone in at once.

Second, a permit ceiling is not a deployment plan. No regulator anywhere hands out thousands of slots expecting them filled on day one. The number is room to grow without coming back for another hearing. Read that way, it’s sensible paperwork, not a prediction.

So what did Tesla itself say? This is the part worth your time. During the four-hour session, Tesla’s Cybercab chief engineer, Eric Early, told the commissioners that the big number has always been a ceiling. He said he doesn’t expect to be in a position to deploy that many cars by this time next year. He said the company would be extremely happy to reach twenty-five hundred.

Tesla’s own engineer cut the headline in half while the headline was being written.

Now put that next to the floor. Electrek’s count, which Tesla hasn’t disputed, is that the company’s entire unsupervised fleet in the United States runs a couple dozen cars at the same time. Austin peaked in late April and drifted down. Dallas and Houston run a handful each. Tesla has said it won’t scale the fleet until the next generation of its driving software, version fifteen, ships.

That’s expected late this year or early next.

Think of it like a restaurant that just signed a lease for a thousand-seat dining room. The lease is real. The landlord is happy. But the kitchen has two burners, and the chef has said the third burner arrives next year. The lease tells you about ambition. The burners tell you about dinner.

The burners, in this case, are software and cars. On software, Tesla says roughly forty percent of version fifteen’s core pieces are already running on the robotaxi fleet, with positive early feedback. That’s Tesla’s characterization. On cars, Tesla told the commission it will launch the Vegas service with Model Ys, with Cybercab joining later.

Cybercab production started at Giga Texas this year, but the company itself has said volume is a twenty twenty-seven story.

There’s also a practical sequence before the first paid ride. Vehicle inspections. Proof of insurance. Submitting fares to the regulator. Tesla expects to clear those in about thirty days. So the honest near-term picture is a few cars on the Strip in September, then growth tied to whatever version fifteen actually delivers.

Why does the gap matter to anyone outside Nevada? Because the five percent stock move was priced on the ceiling, not the floor. A ceiling is free. Nevada’s permit costs Tesla nothing if it never fills it. What costs money is the floor. The cars, the compute, the remote operators, the charging hub Tesla just filed permits for in South Austin.

The market rewarded the number that doesn’t require any of that.

And this isn’t Tesla’s first Vegas promise. We recently walked through the Boring Company’s tunnel loop under the Strip, where the airport connector slipped past its target date and still isn’t open. Same city, same pattern. A big number on a permit, and a slower number on the ground.

Count the cars, not the permits.

Five thousand is what Nevada allows. Twenty is what Tesla runs. Twenty-five hundred is what Tesla’s own engineer called a good year. When the next headline says Tesla was approved for thousands of robotaxis in a new state, the question to ask is the same one. How many are driving right now, and what has to ship before that number moves.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

Over the past twelve months, Tesla the company got dramatically more valuable. Tesla the share, the thing you can actually buy, went almost nowhere.

Over the past twelve months, Tesla the company got dramatically more valuable. Tesla the share, the thing you can actually buy, went almost nowhere. Both of those statements are true at the same time. The gap between them is where this episode lives.

This is Tesorb Signal and this is Lena Ruiz. Today, three things. First, the quiet event that separated the company from its shareholders this year. Second, the fair case for why Tesla did it. And third, the arithmetic of what one share of Tesla actually earned, and the number worth watching from now on.

Start with the quiet event. Over the twelve months ending in June, the number of Tesla shares in existence grew by roughly twenty-two percent. Not the stock price. The share count. For every five shares that existed a year ago, there are now about six.

Think of a company as a pizza. The market decides what the whole pizza is worth. Your shares are slices. If the pizza gets more valuable but somebody keeps cutting more slices, your slice can shrink even while the pizza grows. That’s dilution. It doesn’t show up on your statement as a fee. It shows up as gains you never received.

Most big companies work in the opposite direction. They buy their own shares back, shrinking the count, so each remaining slice quietly grows. Tesla spent this year cutting new slices instead. That choice isn’t automatically wrong. But it changes what a rising stock price actually means for the person holding it.

So where did the new slices come from? Two places, and both lead to the same person. First, Elon Musk exercised the mountain of stock options from his two thousand eighteen pay deal. We covered that move, and the voting power that came with it, when it happened.

Second, and bigger. In September, Tesla’s board granted him a brand new award. Four hundred twenty-four million restricted shares. More than a tenth of the entire company, created in a single grant. Shareholders approved it in November by a wide margin.

Now the fair case, and it deserves a full hearing. Musk doesn’t simply keep those shares. They sit behind twelve locked gates. Tesla’s market value has to climb, step by step, toward eight and a half trillion dollars, several times what the company is worth today. Robots have to ship. Robotaxis have to run. Self-driving subscriptions have to multiply.

And even the shares he earns come with strings. Vesting periods stretching toward a decade. A five year holding rule after that. He even has to pay the company back the grant day value of every share before he keeps the upside. On day one, the award was designed to be worth nothing unless the stock rises.

The board’s argument is history. Its letter to shareholders says the two thousand eighteen deal cost Tesla a couple of billion in accounting charges and coincided with more than seven hundred billion dollars of new market value. Whatever you think of the man, they argue, this structure worked once. Pay for miracles, but only if the miracles arrive.

That’s the case. Shareholders heard it and said yes, about three in four votes in favor. Nobody was tricked. The deal is legitimate. But legitimate and free are different words. And this is where the arithmetic comes in.

Over those same twelve months ending in June, Tesla’s total market value rose about sixty-two percent. A monster year for the pizza. But remember the new slices. Measure what a single share gained over that stretch, and the number falls to roughly thirty-two percent. Half the headline, gone before it ever reached you.

The other half didn’t vanish. It went to the new shares. Mostly, it went to one person’s future stake in the company. That’s not an accusation. It’s just where the math points. The company’s growth was real. It was simply shared among more slices than before.

And the story gets tougher after June. Tesla stock has been on a rough ride this year. By the middle of August, a share bought twelve months ago is sitting roughly where it started.

The company posted record revenue this summer, in the quarter we covered as a verdict. Record sales up top, almost nothing left at the operating line. Spread what remains across twenty-two percent more shares, and each slice’s claim on those earnings gets thinner still.

None of this says Tesla is doomed, and none of it says the award was wrong. Plenty of great companies pay their people in stock. What it says is simpler. When a company pays for its future in shares, its shareholders pay for it too. Not with a bill in the mail. With a smaller claim on everything the company builds next.

If the milestones hit, if the robots ship and the market value multiplies, today’s dilution will look like the cheapest hire in corporate history. If they don’t, shareholders gave up real slices of the pizza for a promise. Both outcomes are still on the table. The award runs for a decade. This story is at its beginning.

So the number that matters isn’t what the pie is worth. It’s how many slices it’s cut into. Every quarter, Tesla reports its share count right next to its revenue, and almost nobody reads it. Start reading it. The stock price tells you what the market feels. The share count tells you what you actually own.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

On New Year’s Eve, while most people were drafting resolutions they’d abandon by February, Elon Musk posted one for his brain implant company. In twenty twenty-six, he wrote, Neuralink will start high volume production of brain computer interfaces, with a surgical procedure that’s almost entirely automated.

On New Year’s Eve, while most people were drafting resolutions they’d abandon by February, Elon Musk posted one for his brain implant company. In twenty twenty-six, he wrote, Neuralink will start high volume production of brain computer interfaces, with a surgical procedure that’s almost entirely automated. It’s August. Time to check the ledger.

This is Tesorb Signal and I’m Lena Ruiz. Today, Neuralink’s promise ledger. Three things. What exactly was promised for this year. What the clinical reality looks like right now. And the fight brewing around this company that most headlines skip.

Quick context for anyone new to this story. The device is a coin sized implant that replaces a small piece of skull, with threads finer than human hair reading signals from brain tissue. Musk once described it as a Fitbit in your skull. Its first real job is letting paralyzed people control computers with thought alone.

And the stakes reach well past one company. Millions of people worldwide live with severe paralysis, and every brain interface maker is trying to get from heroic individual surgeries to something a health system can actually deliver. Whoever solves scale first won’t just win a market. They’ll set the rules for it.

The New Year’s Eve post was specific in a way worth pausing on. High volume production of devices. A streamlined, almost entirely automated surgery. Even a technical change, implant threads passing through the brain’s protective membrane without removing it, which Musk called a big deal. This wasn’t a vision statement. It read like a production plan with a deadline.

Now the ledger’s other column. In January, Neuralink announced it had reached twenty-one trial participants, each living with an implant reading their neural activity, most of them people with paralysis. That’s the entire installed base of the product line that’s supposed to enter high volume production before this year ends.

And to be clear, what those participants are doing is remarkable. The very first participant famously played online chess within weeks of his surgery. Since then, they’ve progressed from moving cursors to controlling smartphones and robotic arms. This summer, reports described trial participants steering powered wheelchairs with decoded brain signals, thought becoming motion.

Earlier this year, a paralyzed British student made news controlling a laptop entirely with his mind. For the people in these trials, this isn’t hype. It’s independence.

The pace is accelerating too. The first participant was implanted about two and a half years ago, one person, one implant, one careful experiment.

Now trials run in multiple countries, including a study across two sites in Britain and another in the Middle East. Doctors describe the early results as promising, even while stressing the technology remains experimental. As clinical programs go, this one moves fast.

But hold the promise against the pace. High volume production usually describes thousands of units. The clinical reality is a couple dozen closely monitored participants in regulated trials, where every implant is a scheduled hospital surgery.

That’s not a criticism of the science. It’s a measure of the distance between the December post and the August reality, with four months left on the calendar.

Now, in fairness, Neuralink could well be building devices faster than it implants them. Factories can outrun trials, and stockpiling hardware ahead of approval is a legitimate strategy. But a brain implant only counts when it’s in a brain, approved, monitored, and supported. And those gates belong to regulators and hospitals, not production lines.

Even the automated surgery promise is only half new. A robot already places the implant’s threads today, because they’re far too fine for human hands. What the promise really means is removing the humans around the robot, the neurosurgeon on standby, the operating room, the recovery ward, the follow up care. Automating a machine is one problem. Automating a hospital is another one entirely.

Which brings us to the fight most headlines skip. Late last year, Neuralink hired away a senior official from the FDA office that regulates it, a move that reportedly surprised and infuriated rivals in the field.

Competitors have gone public with a sharper complaint too, that the company’s talk of merging humans with AI and one day implanting healthy people is undermining trust in an industry trying to build medical devices.

That tension runs through everything Neuralink says about itself. Is it a medical device company restoring independence to paralyzed patients? The trial results say yes. Or is it building toward consumer implants and a symbiosis with artificial intelligence? The founder’s own words say that too.

Investors pricing the company certainly seem to be betting on the bigger story. We ran that valuation math earlier this summer.

In medicine, though, the ledger that matters is counted in patients, not posts. Twenty-one is a genuinely historic number, more simultaneous brain implant participants than this field has ever supported.

It is also not high volume, and no reading of the calendar makes it so by December. When the promise and the patient count disagree, believe the patient count. It’s the one with regulators attached.

So hold the promise and the patient count side by side and let them argue. If volume manufacturing truly begins before New Year’s, it will be one of the fastest medical device scale ups ever attempted, and we’ll cover it. If it doesn’t, the promise will roll forward the way these promises tend to do. Either way, this ledger stays open.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.

In October of twenty sixteen, Tesla published a sentence that would quietly become one of the most expensive promises in the car business. Every Tesla leaving the factory, it said, now ships with all the hardware it needs to fully drive itself.

In October of twenty sixteen, Tesla published a sentence that would quietly become one of the most expensive promises in the car business. Every Tesla leaving the factory, it said, now ships with all the hardware it needs to fully drive itself. People believed it. People paid for it. And this summer, almost ten years later, those people finally found out what they’re getting.

This is Tesorb Signal and I’m Lena Ruiz. Today, the hardware three story. Three things. The promise Tesla made and kept remaking. What actually arrived on those cars this summer. And what the answer says about every promise the company is making right now.

Some history, because the details matter. The self driving hardware in that twenty sixteen promise didn’t last. Tesla replaced it in twenty nineteen with a computer of its own design, called hardware three, and upgraded earlier buyers who had paid for full self driving.

The salesmanship around it was total. Back in twenty sixteen, Tesla had posted a demo of a car driving itself with a caption saying the person in the seat was only there for legal reasons. Hardware three inherited that swagger. On stage, Musk described a coming robotaxi network and famously called Teslas appreciating assets.

For years, hardware three cars carried that future on their window stickers. The full self driving package sold for as much as fifteen thousand dollars, on the promise the car would one day drive itself. And this isn’t a niche fleet. The computer shipped in Teslas for roughly four years, which puts millions of these cars on roads today. Every one of them is a line in this ledger.

Then the ground shifted. Newer Teslas got a far more powerful computer. The software teams built for that one first. And hardware three quietly became legacy hardware, parked on year old software while newer cars ran two generations ahead.

The candor arrived in stages. Early last year, Musk admitted Tesla would need to upgrade the computers in hardware three cars whose owners bought full self driving, and called that reality painful.

Then this spring, on an earnings call, he went further. Hardware three cars, he said, can’t achieve unsupervised full self driving. The promise wasn’t cancelled. It was reassigned to different hardware. For the people who paid a decade ago, that distinction matters enormously.

So what did arrive? At the end of June, Tesla began rolling out FSD version fourteen Lite, a slimmed down build of its newest driving software, distilled to fit the old computer’s limits.

And those limits are real. Hardware three has a fraction of the memory bandwidth of its successor. The Lite build essentially lets the old computer study the new one’s homework, learning from the bigger model’s behavior.

Let’s be fair about the engineering, because Lite is genuinely impressive. Squeezing a modern driving model onto a seven year old computer is genuinely hard engineering, and it worked well enough that owners posted drives handling rain, complex merges, and dense city traffic. For drivers who had written off ever seeing new software, it felt like a resurrection.

That part of the story is a clear win.

And Tesla didn’t treat it as a one time gift. Within weeks, Lite went from a cautious limited release to a wide release, reaching roughly a tenth of the North American fleet within days, and the company unified its software line so hardware three and the newer cars now update together. Whatever else is true, the oldest cars are back inside the development loop instead of frozen outside it.

But the wide release also surfaced the costs. Owners reported regressions, hesitation and phantom braking among them. And a cluster of reports tied the update to overheating warnings on the self driving computer, with a handful of computers failing outright.

A caution flag on that last one, because the evidence cuts both ways. The overheating fault predates Lite by years. Tesla hasn’t confirmed any connection. And prominent supporters point out the computers are liquid cooled, calling the claims speculation running ahead of data. What’s undisputed is the timing, a pattern of reports arriving just as the rollout went wide.

The bigger issue isn’t a bug, though. It’s the ceiling. Lite is supervised software, hands ready, eyes up. And on this hardware, supervised is where the road ends.

The unsupervised future Tesla is targeting for later this year runs only on the newer computer. For hardware three owners, that leaves three outcomes. Live with the ceiling. Wait for an upgrade Tesla has promised but not detailed. Or watch the sticker promise expire quietly.

Pull back, and this stops being a story about one chip. Tesla is selling new promises right now, robotaxis in more cities, cars earning money while their owners sleep, and we keep a running ledger on those.

All of it rests on the same voice that made the twenty sixteen claim. Almost ten years is long enough to grade the original. The grade isn’t zero. But it isn’t what was sold, and the gap was paid by the customers who believed earliest.

So price the newest promise by how the oldest one is being paid off. Watch what Tesla offers hardware three owners next. A real upgrade path with a date on it, or a Lite label doing heavy lifting. However that lands, it will tell you more about the robotaxi timeline than any demo ever will. And when it lands, you’ll hear about it here.

That’s the signal. I’m Lena Ruiz with the Tesorb Signal podcast. For more news about Tesla, SpaceX, and Elon Musk’s companies, visit our website at tesorb.com.

This podcast was developed with the help of using AI assistants, including the voice, and undergoes a detailed review during production. Corrections are posted to the episode page.