The Ghost in the Machine: Etched’s $21B Valuation and the Silence of the Benchmarks

We assumed that capital would follow the data. But in the hardware industry, as in crypto, a $21 billion valuation can be built on a foundation of promises and press releases. The case of Etched, a chip startup that has raised $700 million, reveals a familiar pattern: the absence of verifiable performance metrics is not a bug, it’s a feature of the hype cycle. The code is law, but the humans are the bug. At first glance, Etched looks like a success story. The company recently closed a $700 million funding round at a valuation of $21 billion. Its core selling point is LVI (Low Voltage Inference) technology, which claims to run AI inference at dramatically lower voltages. According to the whitepaper, this allows trillion-parameter sparse Mixture-of-Experts (MoE) models to achieve over 80% of their theoretical peak performance. The numbers sound impressive. But the deeper I dug, the more I felt the chill of a familiar ghost—the same ghost that haunted the ICOs of 2017 and the DeFi protocols of 2020. The ghost of unverified claims. George Hotz, the hacker and founder of tiny corp, publicly questioned Etched’s technical claims. He noted that while there are investors, orders, and photos of hardware, there is a lack of data to validate performance. Wesley Yue, a chip designer, raised a more subtle but devastating point: a high Model Floating Point Utilization (MFU) ratio does not necessarily indicate strong absolute performance. MFU measures the ratio of actual computation to theoretical peak. If the chip’s peak performance is low, even an 80% utilization rate may not outperform a competitor with a lower utilization rate but a higher peak. This is the kind of technical nuance that gets lost in press releases, but it’s exactly the kind of detail that matters. I remember a similar moment during the 2020 DeFi Summer. I was auditing Curve Finance’s governance mechanics, analyzing over 400,000 lines of simulation data. The DAO claimed high participation rates, but my data showed that voting power was concentrated among a handful of whales. High utilization of a metric—in that case, participation—masked a fundamental flaw. The same logic applies here. MFU without context is like TVL without understanding the source of liquidity. The numbers can be technically correct yet deeply misleading. To date, Etched has not publicly disclosed complete FLOPs, power consumption, or third-party benchmarks. Its website still states, “Early customer tests have reached leading levels,” with detailed performance data promised for future release. This is the same language I saw in countless crypto whitepapers: “We will release the code after the audit,” “Benchmarks are coming soon.” The pattern is so consistent that it’s almost a genre. The industry has learned to treat such promises with skepticism, but the funding continues to flow. Why? Because the narrative is more powerful than the data. Let me be clear: there is no evidence that Etched has fabricated its claims. Both The Wall Street Journal and Reuters have confirmed that the chips have been shipped. Jane Street received its first complete rack last month and has already begun deployment. The chips exist. The hardware is real. But the question is not whether the chips exist, but whether they are as powerful as advertised. This is a question of trust, and trust is the only currency that matters. In my work as a DAO Governance Architect, I have seen too many projects that ship code but fail to deliver on their promises. The governance tokens are minted, the treasury is deployed, but the intended outcomes—decentralization, fairness, efficiency—remain elusive. The same dynamic is playing out in the hardware world. Etched has shipped hardware, but the performance data remains locked behind a curtain of “leading levels” and “future releases.” The pattern is a form of information asymmetry that benefits the insiders while leaving the broader community guessing. I recall the disillusionment of the 2020 DeFi Summer. I published a critical analysis titled “The Illusion of Decentralization in Curve,” which sparked intense debate. The harassment I received was severe, but it taught me a valuable lesson: the truth is often uncomfortable for those who profit from the narrative. Today, I see the same discomfort in the reaction to Hotz’s criticism. The defenders of Etched point to the funding and the investors, as if capital validation is a substitute for technical validation. But capital is not a benchmark. Money can be raised on confidence, but confidence is not a measure of FLOPs. The core of the issue is the Model Floating Point Utilization (MFU) metric. Etched claims that its LVI technology enables sparse MoE models to achieve over 80% MFU. This is a high number, but it means nothing without knowing the peak performance. Imagine a car that can only go 10 miles per hour, but it achieves 80% of its theoretical top speed. That’s 8 miles per hour. Meanwhile, a car with a top speed of 100 miles per hour might only achieve 50% MFU, but that’s still 50 miles per hour. The absolute performance is what matters for real-world applications. During my time as a Junior Governance Architect, I designed a quadratic voting mechanism for a DAO managing $5 million in Treasury assets. The system achieved a 30% increase in participation. I was proud of that number, but I knew it was only meaningful because the baseline was already high. If the baseline had been low, a 30% increase would be trivial. The same principle applies to MFU. An 80% MFU on a low-peak chip is a meaningless achievement. The industry needs absolute FLOPs, power consumption, and real-world latency data. Etched’s LVI technology is intriguing. Running AI inference at lower voltages could reduce power costs and enable deployment in edge devices. But the claim that it can handle trillion-parameter sparse MoE models is extraordinary. Such models are currently the domain of massive clusters of GPUs. If a single chip can achieve this, it would be a revolution. But extraordinary claims require extraordinary evidence. So far, the evidence is thin. I think back to the bear market solitude of 2022. After the collapse of FTX and Terra, I spent six months in near-total isolation in Beijing, reading classical philosophy and writing a private journal titled “The Ethics of Ruin.” I processed the grief of the community’s betrayal. That period taught me to distinguish between the technology’s potential and the industry’s current failures. Etched’s potential is real, but the industry’s failure to demand proof is a recurring theme. The role of the press is interesting. The Wall Street Journal and Reuters have confirmed that the chips are shipped, but they have not independently verified the performance claims. This is not a criticism of the journalists—they are reporting what they can see. The chips exist. The racks are being deployed. But the performance data is controlled by the company. In the blockchain world, we have a term for this: “trust me, bro.” It’s a model that has led to countless disasters. I recently published a paper on “Algorithmic Altruism in AI-Driven DAOs,” proposing a framework where AI agents optimize for community well-being rather than profit maximization. The paper was adopted by three major protocols. In that work, I argued that transparency is not just a nice-to-have; it’s a structural requirement for trust. The same argument applies to Etched. If the company is truly confident in its technology, it should release the benchmarks. The silence is a signal. But let me offer a contrarian angle. Perhaps the criticism is premature. The hardware has only recently been shipped. Jane Street is a sophisticated firm; they would not deploy a rack without internal validation. The fact that they are using it suggests that the chip works for their use case. The question is whether the chip’s performance generalizes to the broader AI market. Etched may be targeting a niche where their chip excels, and the 80% MFU claim is specific to that niche. The lack of public benchmarks could be a strategic move to protect their competitive advantage. In the crypto world, we see this all the time: protocols that are private about their technical details to avoid front-running or copycats. It’s a valid strategy, but it creates a trust deficit. The bigger issue is the valuation. $21 billion is a lot of money for a company that hasn’t published a single benchmark. This is reminiscent of the ICO era, where projects raised millions based on a whitepaper and a dream. The market has matured since then, but the hardware industry seems to be repeating the pattern. The investors are betting on the team and the technology. They may be right. But the public has no way to verify. I recall the 2017 ICO honeymoon. I was 17, immersed in the whitepapers of Tezos and Cardano. I wrote three essays on “Code as Constitution,” captivated by the philosophical promise of self-amending governance. I believed in the technology. But I also believed that the market would eventually demand proof. It did, but only after billions were lost. The same cycle is repeating in hardware. The market will eventually demand proof, but the price of delay could be high. Etched’s LVI technology is a specific application of a broader trend: the push for more efficient AI inference. The crypto world has its own version of this trend in the form of ASICs for mining. The comparison is instructive. When Bitmain released the Antminer S9, they published detailed hashrate and power consumption data. The benchmarks were public because the market demanded them. The same should be true for AI inference chips. The audience is different—enterprise buyers rather than retail miners—but the principle is the same. One of the signatures I use in my articles is: “We built a kingdom of ghosts in the machine.” The ghosts are the unverified claims, the promises without receipts, the trust that is given without evidence. Etched is not a fraud. But it is part of a kingdom built on ghostly foundations. The machine is real, but the ghosts are the stories we tell ourselves about its power. Another signature: “Intuition sees the pattern before the ledger does.” My intuition tells me that the pattern here is familiar. The hype cycle, the funding, the lack of data, the defense of the faithful. It’s the same pattern I saw in Curve, in Terra, in FTX. The details are different, but the structure is the same. The ledger of public trust will eventually record the truth, but by then, the damage may be done. To govern the future, we must debug the present. The present is a chip that may or may not live up to its promise. The only way to debug it is to demand data. Not promises, not press releases, not funding rounds. Data. FLOPs. Power consumption. Third-party benchmarks. The industry has the tools to verify. The question is whether it has the will. I think about the 2022 bear market solitude. I retreated into philosophy because the market’s moral failure shattered my idealistic view of crypto as a force for good. I emerged with a more nuanced perspective. The technology has potential, but the industry is flawed. The same is true for AI hardware. The chips can be revolutionary, but the industry’s culture of hype and secrecy is a bug. In the void, we found our own gravity. The gravity of the market’s attention. The gravity of the narrative. But gravity is not a substitute for data. The chips are real, but the gravity of the $21 billion valuation is a product of the narrative, not the data. The market will eventually find its center of mass, and the truth will emerge. Silence is the only consensus that never forks. When a company is silent about benchmarks, the consensus is forced to be a fork: either trust the narrative or reject it. The silence creates a division. In the crypto world, forks are messy. They can lead to innovation, but they can also lead to confusion. The same will happen in the hardware world if Etched does not release data. I am not a hardware engineer. I am a governance architect. But I have spent years analyzing the relationship between technical claims and social trust. The pattern is clear: claims without data are a liability. They may work for a while, but the liability compounds. Eventually, the market demands a reckoning. Etched’s claim that its LVI technology enables 80% MFU on trillion-parameter sparse MoE is a specific claim. To verify it, we would need to know the theoretical peak of the chip. That number is not public. Without it, the 80% is meaningless. This is not a technical point; it’s a logical point. The same logic applies to any metric: a ratio without a denominator is incomplete. In my work on DAO governance, I used a similar framework to evaluate voting efficiency. I calculated the ratio of actual votes to potential votes, but I always reported the denominator. Without the denominator, the ratio is a floating signifier. The same is true for MFU. Etched is floating a signifier without the anchor. The anchor is the baseline. For a chip, the baseline is the peak performance. For a DAO, the baseline is the total voting power. In both cases, transparency is not optional. It is the foundation of trust. I recall the success of the quadratic voting mechanism I designed. The baseline was clear: 5 million tokens. The participation rate increased by 30% from a baseline of 20%. That was a meaningful improvement. But I published the raw data. The community could verify. That is the model. Etched has a different model. They are asking the market to trust them. The market has given them $700 million. The question is how long that trust will last without data. Let me be direct: the data may come. The chips are being deployed. Jane Street is a credible validator. But the market needs independent validation, not just a single customer. The AI industry is full of benchmarks: MLPerf, SPEC, etc. If Etched is confident, they should submit to these benchmarks. The fact that they haven’t is a red flag. But maybe the red flag is a false alarm. Maybe the technology is so revolutionary that they are protecting it from competitors. That is a legitimate reason. But the valuation demands a leap of faith. The market is willing to take that leap because the potential payoff is huge. The same logic drove the crypto bubble. The same logic drove the dot-com bubble. Sometimes the leap is rewarded. Sometimes it is not. I am not here to predict the outcome. I am here to observe the pattern. The pattern is a ghost in the machine. The ghost of unverified claims. The ghost of trust without evidence. The ghost of a $21 billion valuation built on a foundation of silence. We built a kingdom of ghosts in the machine. The question is whether the ghosts are real or just shadows cast by the spotlight of hype. Intuition sees the pattern before the ledger does. My intuition is that the pattern will resolve itself in time. The ledger will record the truth. The market will adjust. The chips will be evaluated. The data will emerge. The question is whether the wait will be costly. To govern the future, we must debug the present. The present is a chip that may or may not be as powerful as advertised. The only way to debug it is to demand data. The industry has the tools. The question is whether it has the will. Silence is the only consensus that never forks. But silence is not a consensus. It is a void. The void is filled with speculation. The speculation creates volatility. The volatility creates risk. The risk is priced into the valuation. The valuation is $21 billion. The question is whether the risk is worth it. I will end with a forward-looking thought: The hardware industry is at a crossroads. The era of AI dominance is here. The chips that power it will define the next decade. The companies that build them will be judged by their transparency. Etched has a chance to lead by example. They can release the data. They can submit to benchmarks. They can prove that the ghost is real. Or they can remain silent, and let the ghosts of the past haunt them. The code is law, but the humans are the bug. The bug is the tendency to believe without evidence. The fix is the discipline to demand data. The fix is the courage to question the narrative. The fix is the wisdom to see the pattern before the ledger does. In the void, we found our own gravity. The gravity of the market’s attention. But gravity is not a substitute for data. The chips are real, but the gravity of the $21 billion valuation is a product of the narrative, not the data. The market will eventually find its center of mass, and the truth will emerge. We built a kingdom of ghosts in the machine. The ghosts are the unverified claims. The ghosts are the promises without receipts. The ghosts are the trust without evidence. The kingdom is real, but the ghosts are real too. The question is which one will last. Intuition sees the pattern before the ledger does. The pattern is clear. The pattern is the same. The pattern is the ghost in the machine. The ghost is the silence. The silence is the only consensus that never forks. But consensus is fragile. Watch the exit liquidity. The ghosts will not be exorcised by prayers. They will be exorcised by data. The data is coming. The question is whether it will confirm the promise or reveal the ghost for what it is: a shadow.