16 October, 2024
Back Market lays out its plan to make refurbished phones go mainstream
Back Market held a press conference on Thursday morning in Paris to talk about upcoming product launches and give an update on the company’s current situation. If you’re not familiar with the French startup, it operates a marketplace of refurbished electronics devices — mostly smartphones. It has attracted a lot of investor cash in recent years but has also been through tougher times.
In 2021, just like many large tech companies, Back Market rode the wave of zero-interest rate policies around the world and raised an enormous amount of money: a $335 million Series D round was followed by another $510 million Series E round mere months later.
After reaching a valuation of $5.7 billion, Back Market realized that the economy was slowing down. It conducted a small round of layoffs in late 2022, telling French newspaper Les Échos it was “the best way to achieve profitability in the coming years.”
Fast-forward to Thursday’s press conference and the company was keen to demonstrate its focus is back on product launches and new projects. Back Market said it wants to find new distribution channels and go premium so that more people think about buying a refurbished device instead of a new one.
Finding customers where they are already
Over the past 10 years, Back Market hasn’t just captured a decent chunk of the secondhand electronics market, it has expanded the market for refurbished smartphones. The pitch is simple: A refurbished device is cheaper than a new one and it’s also better for the planet. Moreover, when it comes to smartphones, it has become much harder to define why this year’s model is better than last year’s — so why shell out lots of money buying new to get only an incremental upgrade?
The company doesn’t handle smartphones and other electronic devices directly. Instead, it partners with 1,800 companies that repair and resell old devices. So it’s essentially a specialized services marketplace. Since its inception, it has sold 30 million refurbished devices to 15 million customers.
Most Back Market customers buy devices on its website or through its mobile app. But the company has recognized it’s sometimes constrained by its partners’ inventory. This is why it wants to expand supply and demand with strategic partnerships.
For instance, it’s partnering with Sony for PlayStation consoles. “A lot of people are coming to Back Market to try and purchase their PlayStation,” said co-founder and CEO Thibaud Hug de Larauze. But the issue is that Back Market is constrained when it comes to supplies for this type of device.
While many people think about smartphone trade-ins, most people don’t think about selling their old consoles. “With this partnership with PlayStation by Sony, we are the only partner to trade in every PlayStation within Sony’s website, within the Sony PlayStation store,” he noted.
As a result, people buying a new PlayStation get a discount with trade-ins at checkout and Back Market is no longer out of stock for old PlayStation consoles. This is a good example of what Back Market has in mind for future partnerships.
“This is one of the first [partnerships of this kind] but we really want to bring it everywhere where customers are actually shopping new. We want to get them where they are, in order to get their old tech — in order to serve it to people who want access to refurbished tech,” Hug de Larauze added.
On the smartphone front, trade-ins are already quite popular. However, customers visiting a phone store usually end up buying a new device along with a long-term plan.
Back Market is going to partner with telecom companies so that customers can also get a discount on refurbished devices in exchange for a long-term plan. The first two partners for this are Bouygues Telecom in France and Visible, a subsidiary of Verizon Wireless in the U.S.
A new premium tier with official parts
Quality remains the main concern when it comes to buying refurbished devices. In addition to allowing returns, the company is constantly tracking the rate of faulty devices on its platform and trying to bring that number down. Back Market now has a defective rate of 4%, meaning that one in every 25 phones doesn’t work as expected in one way or another.
When customers buy a smartphone on Back Market, they can choose between a device in “fair,” “good,” or “excellent” condition. The company has now rolled out a new top tier — called “premium.”
The main difference between smartphones with no signs of use and premium refurbished devices is that Back Market certifies that premium devices have been repaired with official parts exclusively.
In addition to this new premium tier, Back Market is working on an app update to turn it into a smartphone companion. You can register your smartphone with your Back Market account to receive tips to keep your device in good shape for longer. They are also working on gamification features, including badges and rewards.
Similarly, Back Market will make it easier to check the value of your current phone. “You open the Back Market app, you shake your phone and you’ll find out,” chief product officer, Amandine Durr, explained. This feature will launch around Black Friday.
Finally, Back Market is going to use generative AI to make it easier to browse the catalog. It can be hard to compare two smartphone models to understand which one is better for you. In a few months, you’ll be able to select two phones and get an AI-generated summary of how the two models compare.
Profitability in Europe this year
When thinking about growth potential, instead of focusing on the smartphone industry, Back Market said it draws inspiration from the car industry.
“Nine people out of 10 are purchasing a pre-owned car today,” said Hug de Larauze. “Everything has been created and lined up for that — the availability of spare parts for everyone, you’re not forced to repair your car where you purchased it.”
Similarly, repairability is changing for smartphones and spare parts, starting with the European Union. By June 2025, manufacturers will be forced to sell their spare parts to people and companies who want to fix devices themselves.
The shift to refurbished devices is also already well underway in Europe. “Back Market is going to be profitable for the first time in Europe in 2024,” said Hug de Larauze. “This is a big milestone for us because when we created the company and until very recently… we had that label that said: ‘OK, this is an impact company.’ Impact means good feelings, but the money is not there.
“Well it’s not the case, it’s actually making money,” he added. Now, let’s see if Back Market can become the go-to destination for refurbished devices in more countries, starting with the U.S.
How VanMoof’s new owners plan to win over its old customers
When VanMoof declared bankruptcy last year, it left around 5,000 customers who had preordered e-bikes in the lurch. Now VanMoof is up and running under new management, and the company’s current owners are courting those same customers by offering them a €1,000 discount off a new bike.
It’s an audacious strategy, one that bets on jilted customers loving VanMoof’s bikes so much that they’ll shell out several thousand more euros for them.
Before it went bust, VanMoof had asked customers to pay close to the full amount when they preordered, a move designed to give the startup working capital that also resulted in long wait times for delivery. The down payment cost anywhere from €2,300 to €2,500, depending on the model and year, money many customers never got back.
Today’s models — the full-sized S5 with 27.5-inch wheels and a straight frame, as well as the smaller A5 with 24-inch wheels and a step-through frame — cost €3,298. Which means customers who want to take advantage of this discount will have to put down another €2,298 (€3,298 cost of bike minus €1,000 discount) on top of what they already paid for their undelivered e-bike. Simply put, they’d be spending close to €4,600 all together for one VanMoof bike.
“Obviously it’s not a full resolution. We’re very much aware of that,” Eliott Wertheimer, VanMoof’s co-CEO, told TechCrunch. “The way we see it is this is a gesture to help people get back on the road who still believe in [VanMoof].”
Before going bankrupt in July 2023, VanMoof had raised close to $200 million in venture capital and gained a cult following on the vision of its sleek, trendy, uncluttered e-bikes designed end-to-end and controlled by an integrated app. The style was there, but the startup lacked execution. Using bespoke parts meant the bikes often broke, and it was difficult to replace those parts in a timely manner, especially without a robust servicing network in place. The company also used its VC money to artificially lower prices in a way that quickly became unsustainable, according to Wertheimer.
Lavoie, a division of McLaren Applied that was formed in 2022 to build e-scooters, acquired VanMoof in August 2023. Since then, Lavoie has worked to re-establish VanMoof’s supply chain and set up a wide service network throughout Europe and parts of the U.S.; reinvigorate VanMoof’s technical ecosystem, including its apps and website; and re-engineer VanMoof’s core products. In other words, today VanMoof claims to offer more reliable, repairable e-bikes that have gone through McLaren’s testing and design iteration process.
“We’re past restructuring, we’re past restarting. We’re getting into how we re-establish the brand and relaunch,” said Wertheimer. “An ongoing consideration throughout this whole journey was what can we do for people who didn’t get their bikes?”
Apparently, the answer to that question is to try and hook customers with discounts instead of giving them their money back because that money is tied up in bankruptcy proceedings. Wertheimer told TechCrunch the money customers used to pay for their bikes, as well as the bikes themselves, are part of the bankruptcy estate, which is being managed by the estate’s administrators in the Netherlands. That means Lavoie doesn’t have access to those funds.
“So anything we could do to support people who didn’t get their bikes from the old company will effectively have to come out of our own pocket,” said Wertheimer, noting that €1,000 is the most Lavoie could afford “without threatening our existence.”
Wertheimer also noted that the bankruptcy process is ongoing, and customers still stand to get partial refunds through that once it’s resolved. Although, given what is likely a long line of secured creditors and priority unsecured creditors ahead of those customers (not to mention legal fees associated with the bankruptcy process), customers probably shouldn’t hold their breath.
For those who do want to sign up for the discount, they can apply here — but get ready for a somewhat convoluted process.
When Lavoie took over VanMoof, it wasn’t able to access the company’s customer orders due to a combination of a chaotic back-end and data-sharing constraints from Europe’s GDPR regulation. That means customers who want to cash in their discount will need to reach out to VanMoof directly and show documentation to prove they made an order.
They’ll also need to go through the rigmarole of trying to get a refund from their bank via a chargeback, if they haven’t already. VanMoof will only provide discounts to people who can prove that they tried and failed to get their money back this way.
For those who are happy to follow all those steps and ante up, they have until December 31, 2027 to apply their discount.
It’s unclear if VanMoof’s strategy will pay off. One thing is certain: The startup’s future hinges on its ability to regain customer trust and deliver on its promises. Customers will have to decide on whether the allure of a sexy, re-engineered e-bike is worth the price and the effort, or if past failures will keep them away for good.
Updated 7/15/24 9:30 a.m. PT to clarify costs for customers who placed preorders (undelivered) before VanMoof declared bankruptcy.
Tesla Dojo: Elon Musk's big plan to build an AI supercomputer, explained
For years, Elon Musk has talked about Dojo — the AI supercomputer that will be the cornerstone of Tesla’s AI ambitions. It’s important enough to Musk that he recently said the company’s AI team is going to “double down” on Dojo as Tesla gears up to reveal its robotaxi in October.
But what exactly is Dojo? And why is it so critical to Tesla’s long-term strategy?
In short: Dojo is Tesla’s custom-built supercomputer that’s designed to train its “Full Self-Driving” neural networks. Beefing up Dojo goes hand-in-hand with Tesla’s goal to reach full self-driving and bring a robotaxi to market. FSD, which is on almost 2 million Tesla vehicles today, can perform some automated driving tasks but still requires a human to be attentive behind the wheel.
Tesla delayed the reveal of its robotaxi, which was slated for August, to October, but both Musk’s public rhetoric and information from sources inside Tesla tell us that the goal of autonomy isn’t going away.
And Tesla appears poised to spend big on AI and Dojo to reach that feat.
Tesla’s Dojo backstory
Musk doesn’t want Tesla to be just an automaker, or even a purveyor of solar panels and energy storage systems. Instead, he wants Tesla to be an AI company, one that has cracked the code to self-driving cars by mimicking human perception.
Most other companies building autonomous vehicle technology rely on a combination of sensors to perceive the world — like lidar, radar and cameras — as well as high-definition maps to localize the vehicle. Tesla believes it can achieve fully autonomous driving by relying on cameras alone to capture visual data and then use advanced neural networks to process that data and make quick decisions about how the car should behave.
As Tesla’s former head of AI, Andrej Karpathy, said at the automaker’s first AI Day in 2021, the company is basically trying to build “a synthetic animal from the ground up.” (Musk had been teasing Dojo since 2019, but Tesla officially announced it at AI Day.)
Companies like Alphabet’s Waymo have commercialized Level 4 autonomous vehicles — which the SAE defines as a system that can drive itself without the need for human intervention under certain conditions — through a more traditional sensor and machine learning approach. Tesla has still yet to produce an autonomous system that doesn’t require a human behind the wheel.
About 1.8 million people have paid the hefty subscription price for Tesla’s FSD, which currently costs $8,000 and has been priced as high as $15,000. The pitch is that Dojo-trained AI software will eventually be pushed out to Tesla customers via over-the-air updates. The scale of FSD also means Tesla has been able to rake in millions of miles worth of video footage that it uses to train FSD. The idea there is that the more data Tesla can collect, the closer the automaker can get to actually achieving full self-driving.
However, some industry experts say there might be a limit to the brute force approach of throwing more data at a model and expecting it to get smarter.
“First of all, there’s an economic constraint, and soon it will just get too expensive to do that,” Anand Raghunathan, Purdue University’s Silicon Valley professor of electrical and computer engineering, told TechCrunch. Further, he said, “Some people claim that we might actually run out of meaningful data to train the models on. More data doesn’t necessarily mean more information, so it depends on whether that data has information that is useful to create a better model, and if the training process is able to actually distill that information into a better model.”
Raghunathan said despite these doubts, the trend of more data appears to be here for the short-term at least. And more data means more compute power needed to store and process it all to train Tesla’s AI models. That is where Dojo, the supercomputer, comes in.
What is a supercomputer?
Dojo is Tesla’s supercomputer system that’s designed to function as a training ground for AI, specifically FSD. The name is a nod to the space where martial arts are practiced.
A supercomputer is made up of thousands of smaller computers called nodes. Each of those nodes has its own CPU (central processing unit) and GPU (graphics processing unit). The former handles overall management of the node, and the latter does the complex stuff, like splitting tasks into multiple parts and working on them simultaneously. GPUs are essential for machine learning operations like those that power FSD training in simulation. They also power large language models, which is why the rise of generative AI has made Nvidia the most valuable company on the planet.
Even Tesla buys Nvidia GPUs to train its AI (more on that later).
Why does Tesla need a supercomputer?
Tesla’s vision-only approach is the main reason Tesla needs a supercomputer. The neural networks behind FSD are trained on vast amounts of driving data to recognize and classify objects around the vehicle and then make driving decisions. That means that when FSD is engaged, the neural nets have to collect and process visual data continuously at speeds that match the depth and velocity recognition capabilities of a human.
In other words, Tesla means to create a digital duplicate of the human visual cortex and brain function.
To get there, Tesla needs to store and process all the video data collected from its cars around the world and run millions of simulations to train its model on the data.
Dojo pics pic.twitter.com/Lu8YiZXo8c
— Elon Musk (@elonmusk) July 23, 2024
Tesla appears to rely on Nvidia to power its current Dojo training computer, but it doesn’t want to have all its eggs in one basket — not least because Nvidia chips are expensive. Tesla also hopes to make something better that increases bandwidth and decreases latencies. That’s why the automaker’s AI division decided to come up with its own custom hardware program that aims to train AI models more efficiently than traditional systems.
At that program’s core is Tesla’s proprietary D1 chips, which the company says are optimized for AI workloads.
Tell me more about these chips
Tesla is of a similar opinion to Apple in that it believes hardware and software should be designed to work together. That’s why Tesla is working to move away from the standard GPU hardware and design its own chips to power Dojo.
Tesla unveiled its D1 chip, a silicon square the size of a palm, on AI Day in 2021. The D1 chip entered into production as of at least May this year. The Taiwan Semiconductor Manufacturing Company (TSMC) is manufacturing the chips using 7 nanometer semiconductor nodes. The D1 has 50 billion transistors and a large die size of 645 millimeters squared, according to Tesla. This is all to say that the D1 promises to be extremely powerful and efficient and to handle complex tasks quickly.
“We can do compute and data transfers simultaneously, and our custom ISA, which is the instruction set architecture, is fully optimized for machine learning workloads,” said Ganesh Venkataramanan, former senior director of Autopilot hardware, at Tesla’s 2021 AI Day. “This is a pure machine learning.”
The D1 is still not as powerful as Nvidia’s A100 chip, though, which is also manufactured by TSMC using a 7 nanometer process. The A100 contains 54 billion transistors and has a die size of 826 square millimeters, so it performs slightly better than Tesla’s D1.
To get a higher bandwidth and higher compute power, Tesla’s AI team fused 25 D1 chips together into one tile to function as a unified computer system. Each tile has a compute power of 9 petaflops and 36 terabytes per second of bandwidth, and contains all the hardware necessary for power, cooling and data transfer. You can think of the tile as a self-sufficient computer made up of 25 smaller computers. Six of those tiles make up one rack, and two racks make up a cabinet. Ten cabinets make up an ExaPOD. At AI Day 2022, Tesla said Dojo would scale by deploying multiple ExaPODs. All of this together makes up the supercomputer.
Tesla is also working on a next-gen D2 chip that aims to solve information flow bottlenecks. Instead of connecting the individual chips, the D2 would put the entire Dojo tile onto a single wafer of silicon.
Tesla hasn’t confirmed how many D1 chips it has ordered or expects to receive. The company also hasn’t provided a timeline for how long it will take to get Dojo supercomputers running on D1 chips.
In response to a June post on X that said: “Elon is building a giant GPU cooler in Texas,” Musk replied that Tesla was aiming for “half Tesla AI hardware, half Nvidia/other” over the next 18 months or so. The “other” could be AMD chips, per Musk’s comment in January.
What does Dojo mean for Tesla?
Taking control of its own chip production means that Tesla might one day be able to quickly add large amounts of compute power to AI training programs at a low cost, particularly as Tesla and TSMC scale up chip production.
It also means that Tesla may not have to rely on Nvidia’s chips in the future, which are increasingly expensive and hard to secure.
During Tesla’s second-quarter earnings call, Musk said that demand for Nvidia hardware is “so high that it’s often difficult to get the GPUs.” He said he was “quite concerned about actually being able to get steady GPUs when we want them, and I think this therefore requires that we put a lot more effort on Dojo in order to ensure that we’ve got the training capability that we need.”
That said, Tesla is still buying Nvidia chips today to train its AI. In June, Musk posted on X:
Of the roughly $10B in AI-related expenditures I said Tesla would make this year, about half is internal, primarily the Tesla-designed AI inference computer and sensors present in all of our cars, plus Dojo. For building the AI training superclusters, Nvidia hardware is about 2/3 of the cost. My current best guess for Nvidia purchases by Tesla are $3B to $4B this year.
“Inference compute” refers to the AI computations performed by Tesla cars in real time and is separate from the training compute that Dojo is responsible for.
Dojo is a risky bet, one that Musk has hedged several times by saying that Tesla might not succeed.
In the long run, Tesla could theoretically create a new business model based on its AI division. Musk has said that the first version of Dojo will be tailored for Tesla computer vision labeling and training, which is great for FSD and for training Optimus, Tesla’s humanoid robot. But it wouldn’t be useful for much else.
Musk has said that future versions of Dojo will be more tailored to general-purpose AI training. One potential problem with that is almost all AI software out there has been written to work with GPUs. Using Dojo to train general-purpose AI models would require rewriting the software.
That is, unless Tesla rents out its compute, similar to how AWS and Azure rent out cloud computing capabilities. Musk also noted during Q2 earnings that he sees “a path to being competitive with Nvidia with Dojo.”
A September 2023 report from Morgan Stanley predicted that Dojo could add $500 billion to Tesla’s market value by unlocking new revenue streams in the form of robotaxis and software services.
In short, Dojo’s chips are an insurance policy for the automaker, but one that could pay dividends.
How far along is Dojo?
Reuters reported last year that Tesla began production on Dojo in July 2023, but a June 2023 post from Musk suggested that Dojo had been “online and running useful tasks for a few months.”
Around the same time, Tesla said it expected Dojo to be one of the top five most powerful supercomputers by February 2024 — a feat that has yet to be publicly disclosed, leaving us doubtful that it has occurred.
The company also said it expects Dojo’s total compute to reach 100 exaflops in October 2024. (One exaflops is equal to 1 quintillion computer operations per second. To reach 100 exaflops, and assuming that one D1 can achieve 362 teraflops, Tesla would need more than 276,000 D1s, or around 320,500 Nvidia A100 GPUs.)
Tesla also pledged in January 2024 to spend $500 million to build a Dojo supercomputer at its gigafactory in Buffalo, New York.
In May 2024, Musk noted that the rear portion of Tesla’s Austin gigafactory will be reserved for a “super dense, water-cooled supercomputer cluster.”
Just after Tesla’s second-quarter earnings call, Musk posted on X that the automaker’s AI team is using Tesla HW4 AI computer (renamed AI4), which is the hardware that lives on Tesla vehicles, in the training loop with Nvidia GPUs. He noted that the breakdown is roughly 90,000 Nvidia H100s plus 40,000 AI4 computers.
“And Dojo 1 will have roughly 8k H100-equivalent of training online by end of year,” he continued. “Not massive, but not trivial either.”
Tesla Dojo: Elon Musk's big plan to build an AI supercomputer, explained
For years, Elon Musk has talked about Dojo — the AI supercomputer that will be the cornerstone of Tesla’s AI ambitions. It’s important enough to Musk that he recently said the company’s AI team is going to “double down” on Dojo as Tesla gears up to reveal its robotaxi in October.
But what exactly is Dojo? And why is it so critical to Tesla’s long-term strategy?
In short: Dojo is Tesla’s custom-built supercomputer that’s designed to train its “Full Self-Driving” neural networks. Beefing up Dojo goes hand-in-hand with Tesla’s goal to reach full self-driving and bring a robotaxi to market. FSD, which is on about 2 million Tesla vehicles today, can perform some automated driving tasks, but still requires a human to be attentive behind the wheel.
Tesla delayed the reveal of its robotaxi, which was slated for August, to October, but both Musk’s public rhetoric and information from sources inside Tesla tell us that the goal of autonomy isn’t going away.
And Tesla appears poised to spend big on AI and Dojo to reach that feat.
Tesla’s Dojo backstory
Musk doesn’t want Tesla to be just an automaker, or even a purveyor of solar panels and energy storage systems. Instead, he wants Tesla to be an AI company, one that has cracked the code to self-driving cars by mimicking human perception.
Most other companies building autonomous vehicle technology rely on a combination of sensors to perceive the world – like lidar, radar and cameras – as well as high-definition maps to localize the vehicle. Tesla believes it can achieve fully autonomous driving by relying on cameras alone to capture visual data and then use advanced neural networks to process that data and make quick decisions about how the car should behave.
As Tesla’s former head of AI, Andrej Karpathy, said at the automaker’s first AI Day in 2021, the company is basically trying to build “a synthetic animal from the ground up.” (Musk had been teasing Dojo since 2019, but Tesla officially announced it at AI Day.)
Companies like Alphabet’s Waymo have commercialized Level 4 autonomous vehicles – which the SAE defines as a system that can drive itself without the need for human intervention under certain conditions — through a more traditional sensor and machine learning approach. Tesla has still yet to produce an autonomous system that doesn’t require a human behind the wheel.
About 1.8 million people have paid the hefty subscription price for Tesla’s FSD, which currently costs $8,000 and has been priced as high as $15,000. The pitch is that Dojo-trained AI software will eventually be pushed out to Tesla customers via over-the-air updates. The scale of FSD also means Tesla has been able to rake in millions of miles worth of video footage that it uses to train FSD. The idea there is that the more data Tesla can collect, the closer the automaker can get to actually achieving full self-driving.
However, some industry experts say there might be a limit to the brute force approach of throwing more data at a model and expecting it to get smarter.
“First of all, there’s an economic constraint, and soon it will just get too expensive to do that,” Anand Raghunathan, Purdue University’s Silicon Valley professor of electrical and computer engineering, told TechCrunch. “Some people claim that we might actually run out of meaningful data to train the models on. More data doesn’t necessarily mean more information, so it depends on whether that data has information that is useful to create a better model, and if the training process is able to actually distill that information into a better model.”
Raghunathan says despite these doubts, the trend of more data appears to be here for the short-term at least. And more data means more compute power needed to store and process it all to train Tesla’s AI models. That is where Dojo, the supercomputer, comes in.
What is a supercomputer?
Dojo is Tesla’s supercomputer system that’s designed to function as a training ground for AI, specifically FSD. The name is a nod to the space where martial arts are practiced.
A supercomputer is made up of thousands of smaller computers called nodes. Each of those nodes has its own CPU (central processing unit) and GPU (graphics processing unit). The former handles overall management of the node, and the latter does the complex stuff, like splitting tasks into multiple parts and working on them simultaneously. GPUs are essential for machine learning operations like those that power FSD training in simulation. They also power large language models, which is why the rise of generative AI has made Nvidia the most valuable company on the planet.
Even Tesla buys Nvidia GPUs to train its AI (more on that later).
Why does Tesla need a supercomputer?
Tesla’s vision-only approach is the main reason. The neural networks behind FSD are trained on vast amounts of driving data to recognize and classify objects around the vehicle and then make driving decisions. That means, when FSD is engaged, the neural nets have to collect and process visual data continuously at speeds that match the depth and velocity recognition capabilities of a human.
In other words, Tesla means to create a digital duplicate of the human visual cortex and brain function.
To get there, Tesla needs to store and process all the video data collected from its cars around the world and run millions of simulations to train its model on the data.
Dojo pics pic.twitter.com/Lu8YiZXo8c
— Elon Musk (@elonmusk) July 23, 2024
Tesla appears to rely on Nvidia to power its current Dojo training computer, but it doesn’t want to have all its eggs in one basket — not least because Nvidia chips are expensive. Tesla also hopes to make something better that increases bandwidth and decreases latencies. That’s why the automaker’s AI division decided to come up with its own custom hardware program that aims to train AI models more efficiently than traditional systems.
At that program’s core is Tesla’s proprietary D1 chips, which the company says are optimized for AI workloads.
Tell me more about these chips
Tesla is of a similar opinion to Apple, in that it believes hardware and software should be designed to work together. That’s why Tesla is working to move away from the standard GPU hardware and design its own chips to power Dojo.
Tesla unveiled its D1 chip, a silicon square the size of a palm, on AI Day in 2021. The D1 chip entered into production as of at least May this year. The Taiwan Semiconductor Manufacturing Company (TSMC) is manufacturing the chips using 7 nanometer semiconductor nodes. The D1 has 50 billion transistors and a large die size of 645 millimeters squared, according to Tesla. This is all to say that the D1 promises to be extremely powerful and efficient, and handle complex tasks quickly.
“We can do compute and data transfers simultaneously, and our custom ISA, which is the instruction set architecture, is fully optimized for machine learning workloads,” said Ganesh Venkataramanan, former senior director of Autopilot hardware, at Tesla’s 2021 AI Day. “This is a pure machine learning machine.”
The D1 is still not as powerful as Nvidia’s A100 chip, though, which is also manufactured by TSMC using a 7 nanometer process. The A100 contains 54 billion transistors and has a die size of 826 square millimeters, so it performs slightly better than Tesla’s D1.
To get a higher bandwidth and higher compute power, Tesla’s AI team fused 25 D1 chips together into one tile to function as a unified computer system. Each tile has a compute power of 9 petaflops and 36 terabytes per second of bandwidth, and contains all the hardware necessary for power, cooling and data transfer. You can think of the tile as a self-sufficient computer made up of 25 smaller computers. Six of those tiles make up one rack, and two racks make up a cabinet. Ten cabinets make up an ExaPOD. At AI Day 2022, Tesla said Dojo would scale by deploying multiple ExaPODs. All of this together makes up the supercomputer.
Tesla is also working on a next-gen D2 chip that aims to solve information flow bottlenecks. Instead of connecting the individual chips, the D2 would put the entire Dojo tile onto a single wafer of silicon.
Tesla hasn’t confirmed how many D1 chips it has ordered or expects to receive. The company also hasn’t provided a timeline for how long it will take to get Dojo supercomputers running on D1 chips.
In response to a June post on X that said: “Elon is building a giant GPU cooler in Texas,” Musk replied that Tesla was aiming for “half Tesla AI hardware, half Nvidia/other” over the next 18 months or so. The “other” could be AMD chips, per Musk’s comment in January.
What does Dojo mean for Tesla?
Taking control of its own chip production means that Tesla might one day be able to quickly add large amounts of compute power to AI training programs at a low cost. Particularly as Tesla and TSMC scale up chip production, making the chips more affordable.
It also means that Tesla may not have to rely on Nvidia’s chips in the future, which are increasingly expensive and hard to secure.
During Tesla’s second-quarter earnings call, Musk said that demand for Nvidia hardware is “so high that it’s often difficult to get the GPUs.” He said he was “quite concerned about actually being able to get steady GPUs when we want them, and I think this therefore requires that we put a lot more effort on Dojo in order to ensure that we’ve got the training capability that we need.”
That said, Tesla is still buying Nvidia chips today to train its AI. In June, Musk posted on X:
“Of the roughly $10B in AI-related expenditures I said Tesla would make this year, about half is internal, primarily the Tesla-designed AI inference computer and sensors present in all of our cars, plus Dojo. For building the AI training superclusters, Nvidia hardware is about 2/3 of the cost. My current best guess for Nvidia purchases by Tesla are $3B to $4B this year.”
Inference compute refers to the AI computations performed by Tesla cars in real time, and is separate from the training compute that Dojo is responsible for.
Dojo is a risky bet, one that Musk has hedged several times by saying that Tesla might not succeed.
In the long run, Tesla could theoretically create a new business model based on its AI division. Musk has said that the first version of Dojo will be tailored for Tesla computer vision labeling and training, which is great for FSD and training Optimus, Tesla’s humanoid robot. But it wouldn’t be useful for much else.
Musk has said that future versions of Dojo will be more tailored to general purpose AI training. One potential problem with that is that almost all AI software out there has been written to work with GPUs. Using Dojo to train general purpose AI models would require rewriting the software.
That is, unless Tesla rents out its compute, similar to how AWS and Azure rent out cloud computing capabilities. Musk also noted during Q2 earnings that he sees “a path to being competitive with Nvidia with Dojo.”
A September 2023 report from Morgan Stanley predicted that Dojo could add $500 billion to Tesla’s market value by unlocking new revenue streams in the form of robotaxis and software services.
In short, Dojo’s chips are an insurance policy for the automaker, but one that could pay dividends.
How far along is Dojo?
Reuters reported last year that Tesla began production on Dojo in July 2023, but a June 2023 post from Musk suggested that Dojo had been “online and running useful tasks for a few months.”
Around the same time, Tesla said it expected Dojo to be one of the top five most powerful supercomputers by February 2024 — a feat that has yet to be publicly disclosed, leaving us doubtful that it has occurred. The company also said it expects Dojo’s total compute to reach 100 exaflops in October 2024.
(1 exaflop is equal to 1 quintillion computer operations per second. To reach 100 exaflops and assuming that one D1 can achieve 362 teraflops, Tesla would need more than 276,000 D1s, or around 320,500 Nvidia A100 GPUs.)
Tesla also pledged in January 2024 to spend $500 million to build a Dojo supercomputer at its gigafactory in Buffalo, New York.
In May 2024, Musk noted that the rear portion of Tesla’s Austin gigafactory will be reserved for a “super dense, water-cooled supercomputer cluster.”
Just after Tesla’s second-quarter earnings call, Musk posted on X that the automaker’s AI team is using Tesla HW4 AI computer (renamed AI4), which is the hardware that lives on Tesla vehicles, in the training loop with Nvidia GPUs. He noted that the breakdown is roughly 90,000 Nvidia H100s plus 40,000 AI4 computers.
“And Dojo 1 will have roughly 8k H100-equivalent of training online by end of year,” he continued. “Not massive, but not trivial either.”
How VanMoof’s new owners plan to win over its old customers
When VanMoof declared bankruptcy last year, it left around 5,000 customers who had preordered e-bikes in the lurch. Now VanMoof is up and running under new management, and the company’s current owners are courting those same customers by offering them a €1,000 discount off a new bike.
It’s an audacious strategy, one that bets on jilted customers loving VanMoof’s bikes so much that they’ll shell out several thousand more euros for them.
Before it went bust, VanMoof had asked customers to pay close to the full amount when they preordered, a move designed to give the startup working capital that also resulted in long wait times for delivery. The down payment cost anywhere from €2,300 to €2,500, depending on the model and year, money many customers never got back.
Today’s models — the full-sized S5 with 27.5-inch wheels and a straight frame, as well as the smaller A5 with 24-inch wheels and a step-through frame — cost €3,298. Which means customers who want to take advantage of this discount will have to put down another €2,298 (€3,298 cost of bike minus €1,000 discount) on top of what they already paid for their undelivered e-bike. Simply put, they’d be spending close to €4,600 all together for one VanMoof bike.
“Obviously it’s not a full resolution. We’re very much aware of that,” Eliott Wertheimer, VanMoof’s co-CEO, told TechCrunch. “The way we see it is this is a gesture to help people get back on the road who still believe in [VanMoof].”
Before going bankrupt in July 2023, VanMoof had raised close to $200 million in venture capital and gained a cult following on the vision of its sleek, trendy, uncluttered e-bikes designed end-to-end and controlled by an integrated app. The style was there, but the startup lacked execution. Using bespoke parts meant the bikes often broke, and it was difficult to replace those parts in a timely manner, especially without a robust servicing network in place. The company also used its VC money to artificially lower prices in a way that quickly became unsustainable, according to Wertheimer.
Lavoie, a division of McLaren Applied that was formed in 2022 to build e-scooters, acquired VanMoof in August 2023. Since then, Lavoie has worked to re-establish VanMoof’s supply chain and set up a wide service network throughout Europe and parts of the U.S.; reinvigorate VanMoof’s technical ecosystem, including its apps and website; and re-engineer VanMoof’s core products. In other words, today VanMoof claims to offer more reliable, repairable e-bikes that have gone through McLaren’s testing and design iteration process.
“We’re past restructuring, we’re past restarting. We’re getting into how we re-establish the brand and relaunch,” said Wertheimer. “An ongoing consideration throughout this whole journey was what can we do for people who didn’t get their bikes?”
Apparently, the answer to that question is to try and hook customers with discounts instead of giving them their money back because that money is tied up in bankruptcy proceedings. Wertheimer told TechCrunch the money customers used to pay for their bikes, as well as the bikes themselves, are part of the bankruptcy estate, which is being managed by the estate’s administrators in the Netherlands. That means Lavoie doesn’t have access to those funds.
“So anything we could do to support people who didn’t get their bikes from the old company will effectively have to come out of our own pocket,” said Wertheimer, noting that €1,000 is the most Lavoie could afford “without threatening our existence.”
Wertheimer also noted that the bankruptcy process is ongoing, and customers still stand to get partial refunds through that once it’s resolved. Although, given what is likely a long line of secured creditors and priority unsecured creditors ahead of those customers (not to mention legal fees associated with the bankruptcy process), customers probably shouldn’t hold their breath.
For those who do want to sign up for the discount, they can apply here — but get ready for a somewhat convoluted process.
When Lavoie took over VanMoof, it wasn’t able to access the company’s customer orders due to a combination of a chaotic back-end and data-sharing constraints from Europe’s GDPR regulation. That means customers who want to cash in their discount will need to reach out to VanMoof directly and show documentation to prove they made an order.
They’ll also need to go through the rigmarole of trying to get a refund from their bank via a chargeback, if they haven’t already. VanMoof will only provide discounts to people who can prove that they tried and failed to get their money back this way.
For those who are happy to follow all those steps and ante up, they have until December 31, 2027 to apply their discount.
It’s unclear if VanMoof’s strategy will pay off. One thing is certain: The startup’s future hinges on its ability to regain customer trust and deliver on its promises. Customers will have to decide on whether the allure of a sexy, re-engineered e-bike is worth the price and the effort, or if past failures will keep them away for good.
Updated 7/15/24 9:30 a.m. PT to clarify costs for customers who placed preorders (undelivered) before VanMoof declared bankruptcy.
Micromobility startups Tier and Dott plan to merge to find a path to profitability
After years of exploding growth and massive funding rounds, it’s time for consolidation in the micromobility industry. Tier and Dott, two leading European companies in the space, have announced that they plan to merge. The transaction is expected to close within two months of today’s news.
This shouldn’t come as a surprise, as many free-floating scooter and e-bike companies have been struggling over the last few months. Just a few days ago, Superpedestrian announced that it would shut down. Bird also filed for bankruptcy at the end of December.
Tier has had issues of its own, as the company recently laid off 22% of its workforce, representing 180 staff members. It also started offloading some of its activities to better focus on its core business. For instance, it sold Spin to Bird for $19 million — that was before Bird filed for bankruptcy.
That’s why Dott and Tier are going to combine their teams and operations. With razor-thin margins, micromobility is all about scaling to the next level to eventually reach profitability.
Details are still thin as Tier and Dott announced the merger to their respective teams just a few minutes ago. Both companies will now sit down over the coming weeks to figure out how they plan to operate as a company going forward.
For now, nothing is changing — both applications will remain available in the App Store and Google Play Store. Dott and Tier will still be available in more than 20 countries — mostly in Europe. Some of the companies’ main markets are Berlin, Brussels, Dubai, Helsinki, London, Madrid, Paris, Rome, Tel Aviv and Warsaw.
Another €60 million in funding
What’s changing is the leadership team. Lawrence Leuschner, Tier’s co-founder and CEO, will become chairman. It’s still unclear whether he will have an operational role in the company going forward.
Henri Moissinac, Dott’s co-founder and CEO, will be the CEO of the new company. Maxim Romain, who was the COO of Dott, will remain as COO of the combined entity. And finally, Tier’s CFO Alex Gayer will keep his position as CFO.
Some of Tier’s and Dott’s existing investors are putting more money in the new entity. Mubadala Capital and Sofina are leading this new funding round. They were both investors in Tier and in Dott, respectively. It also acquired Spin to expand to North America.
But that’s not all. Estari, M&G, Prosus Ventures, Novator and White Star Capital are also participating in this transaction. It’s worth noting that SoftBank’s Vision Fund 2, one of Tier’s backers, is not participating in this new round.
While terms of the deal remain undisclosed, the total valuation of Dott and Tier is most likely down compared to the big funding rounds of 2021 and 2022.
“Valuations had reached very high levels. And inevitably, as for the tech industry itself, there have been some adjustments,” Dott’s director of policy and communications Matthieu Faure told me.
Two different approaches
While Tier and Dott both operate scooter-sharing and bike-sharing services in major European cities and both work with Ninebot and Okai as their hardware suppliers, they have had different approaches over the years.
Dott is currently active in seven countries representing 40 cities. It has 40,000 scooters and 10,000 bikes and a staff of 600 employees. Dott has raised a total of €210 million in equity and debt ($230 million at today’s exchange rate).
From the very beginning, Dott chose to internalize its operations teams as much as possible and has focused exclusively on free-floating micromobility services. As for Dott’s software stack, everything has been developed in-house.
Tier has expanded to more markets and more cities, as it is currently available in Germany, Austria and Poland, but also Qatar, Saudi Arabia and the United Arab Emirates.
It has also tried a bunch of different things. For instance, it acquired Coup, an electric moped service that mostly operated in Berlin. When the company raised its Series C round, it claimed that it would build a network of user-swappable batteries in European cities.
Tier also acquired Nextbike, one of the leading bike-sharing companies in Europe with a more traditional dock-based system. Overall, Tier now has 2,000 employees when you include all frontline and Nextbike workers.
Now that VC funding has dried up and it’s time to focus on profitability, Tier and Dott may have to refocus slightly to make sure that they can generate profits in their main markets.
“Today, our operating model is pretty good. We manage to be profitable in most of our cities. We’re just lagging behind when it comes to scale,” Dott’s Matthieu Faure told me. “Now, we have the best of both worlds with enough vehicles and a good operating and financial model, so that it’s sustainable over the long term.”
Tier currently generates more revenue than Dott. After the merger, the new micromobility company expects to generate more than €200 million in revenue ($220 million). And, hopefully, they’ll be able to turn a profit.
What the demise of Superpedestrian means for the e-scooter industry
Electric scooter company Bird files for bankruptcy
Scooters in Paris are at a crossroads
Micromobility startups Tier and Dott plan to merge to find a path to profitability
After years of exploding growth and massive funding rounds, it’s time for consolidation in the micromobility industry. Tier and Dott, two leading European companies in the space, have announced that they plan to merge. The transaction is expected to close within two months of today’s news.
This shouldn’t come as a surprise, as many free-floating scooter and e-bike companies have been struggling over the last few months. Just a few days ago, Superpedestrian announced that it would shut down. Bird also filed for bankruptcy at the end of December.
Tier has had issues of its own, as the company recently laid off 22% of its workforce, representing 180 staff members. It also started offloading some of its activities to better focus on its core business. For instance, it sold Spin to Bird for $19 million — that was before Bird filed for bankruptcy.
That’s why Dott and Tier are going to combine their teams and operations. With razor-thin margins, micromobility is all about scaling to the next level to eventually reach profitability.
Details are still thin as Tier and Dott announced the merger to their respective teams just a few minutes ago. Both companies will now sit down over the coming weeks to figure out how they plan to operate as a company going forward.
For now, nothing is changing — both applications will remain available in the App Store and Google Play Store. Dott and Tier will still be available in more than 20 countries — mostly in Europe. Some of the companies’ main markets are Berlin, Brussels, Dubai, Helsinki, London, Madrid, Paris, Rome, Tel Aviv and Warsaw.
Another €60 million in funding
What’s changing is the leadership team. Lawrence Leuschner, Tier’s co-founder and CEO, will become chairman. It’s still unclear whether he will have an operational role in the company going forward.
Henri Moissinac, Dott’s co-founder and CEO, will be the CEO of the new company. Maxim Romain, who was the COO of Dott, will remain as COO of the combined entity. And finally, Tier’s CFO Alex Gayer will keep his position as CFO.
Some of Tier’s and Dott’s existing investors are putting more money in the new entity. Mubadala Capital and Sofina are leading this new funding round. They were both investors in Tier and in Dott, respectively. It also acquired Spin to expand to North America.
But that’s not all. Estari, M&G, Prosus Ventures, Novator and White Star Capital are also participating in this transaction. It’s worth noting that SoftBank’s Vision Fund 2, one of Tier’s backers, is not participating in this new round.
While terms of the deal remain undisclosed, the total valuation of Dott and Tier is most likely down compared to the big funding rounds of 2021 and 2022.
“Valuations had reached very high levels. And inevitably, as for the tech industry itself, there have been some adjustments,” Dott’s director of policy and communications Matthieu Faure told me.
Two different approaches
While Tier and Dott both operate scooter-sharing and bike-sharing services in major European cities and both work with Ninebot and Okai as their hardware suppliers, they have had different approaches over the years.
Dott is currently active in seven countries representing 40 cities. It has 40,000 scooters and 10,000 bikes and a staff of 600 employees. Dott has raised a total of €210 million in equity and debt ($230 million at today’s exchange rate).
From the very beginning, Dott chose to internalize its operations teams as much as possible and has focused exclusively on free-floating micromobility services. As for Dott’s software stack, everything has been developed in-house.
Tier has expanded to more markets and more cities, as it is currently available in Germany, Austria and Poland, but also Qatar, Saudi Arabia and the United Arab Emirates.
It has also tried a bunch of different things. For instance, it acquired Coup, an electric moped service that mostly operated in Berlin. When the company raised its Series C round, it claimed that it would build a network of user-swappable batteries in European cities.
Tier also acquired Nextbike, one of the leading bike-sharing companies in Europe with a more traditional dock-based system. Overall, Tier now has 2,000 employees when you include all frontline and Nextbike workers.
Now that VC funding has dried up and it’s time to focus on profitability, Tier and Dott may have to refocus slightly to make sure that they can generate profits in their main markets.
“Today, our operating model is pretty good. We manage to be profitable in most of our cities. We’re just lagging behind when it comes to scale,” Dott’s Matthieu Faure told me. “Now, we have the best of both worlds with enough vehicles and a good operating and financial model, so that it’s sustainable over the long term.”
Tier currently generates more revenue than Dott. After the merger, the new micromobility company expects to generate more than €200 million in revenue ($220 million). And, hopefully, they’ll be able to turn a profit.
What the demise of Superpedestrian means for the e-scooter industry
Electric scooter company Bird files for bankruptcy
Scooters in Paris are at a crossroads
How to plan for general partner succession
Alan Feld
Contributor
Alan Feld founded Vintage in 2002 and has since grown it into a global ~$4B investment fund investing in funds and companies across the U.S., Europe, Israel, and Canada.
More posts from Alan Feld
How to plan for general partner succession
Nearly 10 years ago, in April 2015, I published a blog called “Confronting the ‘S’ word: Dealing with general partner succession.” As the founder and managing partner of Vintage, I wanted to ensure that Vintage would survive after I retire. Ensuring the survival of Vintage was a responsibility that I owed to our investors, portfolio funds, companies, and employees.
Venture funds take a long time to build and realize their investments — in many cases, more than a decade and a half. An engaged, energetic, committed, and hungry venture management team is as vital at the end of the fund as it is at the beginning. This is true for funds ending their lives and for the two to three additional funds raised along the way. Management team longevity is vital in challenged exit markets (as we are currently all experiencing). Succession management is more critical now than ever before.
Unfortunately, very few VC managers have managed succession well. So, in 2015, I decided to research the best practices in succession management and interviewed the managing partners of several of the world’s leading VCs to see what worked (and did not) in managing their succession processes.
At the time, I identified “six rules of succession”:
1. GPs must proactively manage and time succession: The worst thing a fund manager can do is deal with this issue during fundraising for a new fund. The process needs to be triggered by the GP’s recognition that a long-term team development plan is required, not due to LP questions during fundraising.
2. Implementing the succession process early: A fund management team needs to start the process and implement the mechanisms at least five to seven years before the current leadership team transitions out. It is common for the founding or the current managing partner to start phasing out in their late 50s or early 60s.
3. Gradually devolving management responsibilities to the younger partners: Fundraising and other firm management responsibilities should gradually be transferred to the junior team before the final transition date.
4. A true and full management transition: A successful transition requires just that: a full transition. The founders and managing partners must step aside and allow the new team to run the firm. It means the older partners do not serve on the investment committee of new funds and leave the management and all the new investment decisions to the younger partners.
5. A true economic transition as well: There also has to be a transition in economics. In our research, we found that relatively small, residual economics (carried, but rarely management fee) is given to the older team for the following few funds following retirement.
6. A visible and clear succession process: The process must have both visibility and certainty. It requires an open and genuine dialogue between the senior retiring and incoming management teams.
How did we do against these six rules?
Today, my partners reported the close of a Vintage Growth Fund IV at $200 million. The fund exceeded the $175 million target and the $171 million Vintage Growth Fund III. Almost all our large investors re-upped, and new investors joined. I also became an LP. I am very proud of my partners and the great work they have done so far, and I am excited to watch them take the firm to a new level in the future.
Here’s how we did it:
Founder triggered and timed succession process
My firm started the process at my initiative seven years ago, when I was 55. We did not wait for the LPs to tell us to do this. We put the succession provisions into an agreement among the general partners to apply to me and all general partners at the firm in the future. This ensured that every general partner knew the succession rules in advance.
Duration of the succession process
I gradually started devolving responsibility to my partners. One partner, Abe Finkelstein (now one of the managing partners), took day-to-day responsibility for the investment team three years ago. Another partner, Keren Terner, was brought in a few years ago to gradually take over the operational management of the business. We also agreed that at age 62, you cease to be an investing partner in new funds so that you do not cross age 65 with an active investment period.
Devolving management responsibilities, including fundraising
My partners also started to become more active in fundraising a few years ago. It was essential to pass on as many of the invitations I got to speak at conferences and events to them so they could build their brands. We put them at the center of events for our LPs, particularly by speaking at and leading our annual meetings.
A full management transition
It was more than my ceasing to be an investing partner in new funds at age 62; for a full transition to be completed, I told my partners that it would be a mistake for me to sit on the investment committee of the new funds raised. When a founder is at the table, it undermines the newer leadership because there is a tendency to turn to the founder for their views and give them outsized influence in a fund that they should not be managing at that point.
A fair and unburdening economic transition
I provided that my partners would not have to buy me out. I have a relatively small tail for a few funds, but I felt it essential that my economics in future funds not get in the way of the new managing partners’ ability to build the future partner team. Just as there needed to be a strong team to succeed me, it was no less crucial to have a strong, incentivized team to succeed them, too.
A very open and even public process
The whole succession process was open and visible. We have been openly talking about this for the last seven years. Several years ago, we described the process in detail to our LPs, GPs, and all our employees.
Everyone knew the timetable, “the how,” and “the what.”
Finally, I noted in my 2015 blog that “effective succession requires an overriding element that goes well beyond the mechanics of the process. That element is establishing a firm culture and embedding that culture in the team that takes the reins of the firm.”
As venture capitalists, we are very focused on the culture of the organizations in which we invest. Unfortunately, we devote too little time and attention to the culture of our firms. This can only work if there is a warm, nurturing culture at our firms as well.