From $2.9 billion to $7,386: the number that reveals Tesla's manufacturing gap
A South Korean supplier's write-down offers forensic evidence of the distance between Battery Day promises and factory floor reality
Seven thousand, three hundred and eighty-six dollars.
That is the revised value of a supply contract that, when announced in February 2023, was worth $2.9 billion. South Korean battery materials supplier L&F Co. filed the adjustment this week, and the figure deserves contemplation. Not $7.4 million. Not $74,000. Seven thousand dollars—roughly the cost of a decent used car.
Someone at L&F had to calculate that number. Someone had to sign a regulatory filing that would crystallise, in four digits, the complete collapse of Tesla's battery promises. In a discourse polluted by bulls and bears spinning every development into tribal narrative, this filing offers something rare: a number that cannot be contested.
The contract was supposed to supply enough high-nickel cathode material for approximately 783,000 electric vehicles. The revised figure suggests capacity for roughly one.
The anatomy of a promise
The L&F deal was built on Tesla's 4680 battery cell—the cylindrical cells unveiled at Battery Day in September 2020 with the ambition characteristic of such occasions. Chief executive Elon Musk and then-senior vice president Drew Baglino promised these cells would hold five times the energy of previous designs, deliver six times the power, and halve production costs. The technology that would make this possible was a process called dry electrode coating, acquired through Tesla's 2019 purchase of a company called Maxwell Technologies.
The pitch was seductive. By eliminating the wet slurry process that conventional battery makers use—a process requiring enormous drying ovens and toxic solvents—Tesla claimed it could shrink factory footprint by 90 per cent and cut thousands of dollars from each vehicle. There was a caveat, mentioned almost in passing: Maxwell had developed this process for supercapacitors, not batteries. Musk considered this a "tiny detail" that would be easy to overcome.
Five years and several billion dollars later, the tiny detail has consumed the programme. Internal figures suggest Tesla loses 70 to 80 per cent of its cathodes during production, against an industry standard below 2 per cent. The cells that survive—currently used only in the slow-selling Cybertruck—deliver less energy than older designs, charge more slowly, and overheat. At the 2025 Investor Day, Musk admitted that pursuing dry electrode was "way harder" than anticipated. He called it a mistake.
Drew Baglino, who stood on stage making those Battery Day promises, resigned in April 2024 after eighteen years with the company. His departure came as Tesla cut more than 10 per cent of its workforce, with the 4680 programme specifically affected. According to Walter Isaacson's biography, after Baglino's first meeting with Musk over battery calculations, he told co-founder JB Straubel: "I never want to be in another meeting with Elon." He stayed anyway, for nearly two decades, before leaving as the programme he championed collapsed around him.
The iron law
Why do Tesla's announcements so consistently diverge from what its factories can deliver? The answer may lie outside the electric vehicle industry entirely.
Bent Flyvbjerg, a professor at Oxford's Saïd Business School, has spent decades studying why large projects fail. His data spans hundreds of megaprojects across infrastructure, technology, and defence. The pattern he documents is relentless: nine out of ten experience cost overruns. Timelines slip. Benefits fall short. This has not improved in seventy years of records. Flyvbjerg calls it the iron law of megaprojects—over budget, over time, under benefits, over and over again.
What makes his work useful here is a distinction he draws between two types of failure. The first is cognitive bias: genuine but misplaced optimism about one's ability to overcome obstacles. The second is what Flyvbjerg calls strategic misrepresentation—forecasts designed not to predict reality but to get projects approved, to attract capital, to move markets.
For small projects with modest stakes, failures tend toward the cognitive. People simply misjudge difficulty. For large projects with high strategic importance and intense scrutiny from senior leadership, Flyvbjerg finds that bias shifts toward the strategic. The announcements become performances.
Battery Day 2020 fits this pattern with uncomfortable precision. Maximum strategic importance. Maximum attention from the chief executive. A presentation designed not merely to inform but to captivate. Tesla's stock moved on the announcement, and the company sold the 4680 as the key to affordable electric vehicles. The question Flyvbjerg's framework raises is whether the promises were cognitive errors—genuine mistakes about what dry electrode could achieve—or strategic ones, forecasts crafted for their market effect rather than their accuracy.
A court has already offered one answer. In September 2024, a lawsuit alleging Musk committed securities fraud through misleading statements about autonomous driving was dismissed after Tesla's lawyers successfully argued that his predictions constituted "corporate puffery"—vague optimism that reasonable investors should not rely upon. The legal system, in other words, has concluded that Musk's announcements should not be taken at face value. Wikipedia now maintains an entire article cataloguing his failed predictions about autonomous Tesla vehicles. A viral video shows him promising full self-driving "next year" every year since 2014.
The pattern is not occasional. It is structural.
The believers
The L&F write-down quantifies what happens to companies that build business plans on Tesla's word.
L&F was founded in 2000 to make LCD backlighting units before pivoting to battery materials. By 2023, it had become one of South Korea's three largest cathode manufacturers. The Tesla contract, announced that February, was worth nearly four times the company's annual revenue. Analyst Kim Hyun-soo at Hana Securities calculated it would supply enough material for 300,000 vehicles annually—783,000 over the contract's two-year term.
L&F's chief executive, Choi Su-an, had been working to reduce dependence on two customers—Tesla and LG Energy Solution—who together accounted for 77 per cent of supplies. The Tesla deal was supposed to anchor growth while diversification proceeded. Instead, Choi is now scrambling to sign deals with Northvolt in Europe and invest in alternative chemistries through a partnership with American startup Mitra Chem.
The broader South Korean supply chain tells a similar story. LG Energy Solution has seen contracts worth 13.5 trillion won cancelled this month. SK On has ended its joint venture with Ford. An entire industrial ecosystem built to serve Tesla's promised production volumes is now searching for customers as those volumes fail to materialise.
These are not abstractions. They are factories that hired workers, purchased equipment, and secured financing based on a $2.9 billion commitment that has become $7,386.
Two companies
The genuinely strange fact is this: as L&F filed its write-down, Tesla's stock was hitting all-time highs.
The 4680 programme has failed. The Cybertruck sells at roughly 20,000 units annually against factory capacity of 250,000. European sales have collapsed. Chinese market share is eroding. Yet Tesla's market capitalisation exceeds that of most global automakers combined. Both things are true simultaneously. The question is how.
The answer appears to be that Tesla has become two companies operating in parallel. The first is a manufacturing business subject to ordinary constraints—production yields, supply chains, customer demand. This company is struggling visibly.
The second is something else: a narrative engine that produces not vehicles but announcements. Each new promise—robotaxi, Optimus robot, artificial general intelligence—generates a fresh cycle of attention. The failure of previous promises gets absorbed and forgotten because discourse has already moved to the next horizon.
The dynamic this creates is peculiar. The 4680 does not need to work, because the Cybercab has already taken its place in investor imagination. The Cybercab does not need to work, because Optimus will take its place. Each announcement serves its purpose—sustaining the narrative—regardless of whether fulfilment ever arrives.
For those who own shares in the narrative company, operational failures are noise. For those who believed they owned shares in a manufacturing company—the one that was supposed to produce revolutionary batteries enabling a $25,000 electric vehicle—the L&F filing is signal.
The case against this interpretation
The strongest argument for Tesla's valuation comes from analysts like Dan Ives at Wedbush, who maintains price targets between $600 and $800 based on potential in autonomous vehicles and robotics. The logic: past failures in one domain do not preclude success in another. SpaceX succeeded spectacularly after early rocket explosions. Tesla itself was written off repeatedly before becoming the world's most valuable automaker.
This argument has merit. Companies can fail at one thing while succeeding at another. The 4680's collapse does not logically entail that robotaxis will fail, or that Optimus cannot work. Samsung Securities analyst Cho Hyun-ryul attributes the L&F situation partly to broader EV demand weakness—Tesla is not alone in facing headwinds.
But the counterargument conflates different kinds of failure. SpaceX's early rockets exploded because rocket engineering is hard; the company iterated and improved. The 4680 programme has not failed because battery engineering is hard. It has failed because Tesla acquired a manufacturing process designed for supercapacitors, assumed adaptation to batteries would be trivial, and discovered it was not. The failure is in the assumption, not the iteration.
The question for current promises is whether similar assumptions underlie them. Musk's confidence in imminent robotaxis rests on the premise that camera-only autonomy can match or exceed lidar-based systems like Waymo's. His confidence in Optimus rests on premises about humanoid robotics that most specialists do not share. Tesla's own disclosures warn that its self-driving software "may do the wrong thing at the worst time."
These caveats, buried in legal filings, tell a different story from the stage presentations.
What remains
The L&F write-down may prove to be a turning point—the moment when the manufacturing gap became impossible to ignore. Or it may simply dissolve into the pattern, forgotten within weeks as attention shifts to the next announcement, the next promise, the next horizon.
What the filing establishes, with the clarity of an audited number, is that Tesla's physical business continues to underperform its narrative business by margins that defy ordinary explanation. A $2.9 billion contract does not shrink to $7,386 through normal volatility. That kind of collapse requires systematic disconnection between what is announced and what can be built.
Investors must decide which company they own. The narrative company offers exposure to futures that may or may not arrive—robotaxis, humanoid robots, artificial intelligence breakthroughs. The manufacturing company offers exposure to factories that struggle to produce what was promised years ago.
Both companies trade under the same ticker. Only one of them can deliver the returns that justify the price.
Seven thousand, three hundred and eighty-six dollars. That is what Battery Day is worth, marked to market.