George Hotz just called bluff. The tiny corp founder publicly questioned Etched’s LVI performance claims. Orders exist. Hardware exists. Benchmarks? Missing.
This is a familiar pattern. The crypto hardware cycle taught us: when a chip company waves a $700M raise and a $21B valuation, but refuses to release third-party benchmarks, you are not buying performance—you are buying narrative. Etched’s core pitch is LVI (Low Voltage Inference), a technique that claims to let AI inference run at significantly lower voltages while maintaining accuracy. Their marketing says it enables trillion-parameter sparse MoE models to hit over 80% Model Floating Utilization (MFU). That sounds impressive. But MFU is a ratio, not an absolute measure. If the chip’s theoretical peak is low, 80% of a low number is still low.
From my experience auditing early rollup hardware in 2017, I learned one hard rule: never trust a utilization claim without the raw FLOPS and power envelope. Etched has not published any of these. Their website still says “Early customer tests have reached leading levels,” with detailed data promised for a future date. That is not a signal. That is a delay tactic.
Wesley Yue, a chip designer, raised the same point: high utilization does not equal strong absolute performance. Imagine a Bitcoin miner with 80% efficiency on a 10 TH/s chip. It still loses to a miner with 50% efficiency on a 100 TH/s chip. The ratio is irrelevant without the denominator. Etched’s denominator is unknown.
The contrarian angle is not about fraud. Both The Wall Street Journal and Reuters have confirmed chips have shipped. Jane Street received a full rack last month and is deploying it. That is real. But deployment does not mean the chips outperform existing solutions. Jane Street could be testing for edge cases or running a proof-of-concept. The fact that they are deploying does not validate the 80% MFU claim. It only validates that the hardware exists.
George Hotz’s skepticism is warranted. He runs tinygrad, an open-source deep learning framework. He knows the gap between a paper claim and a real-world benchmark. He is calling out the lack of data. That is the same signal I used when I predicted the Terra collapse: when a project refuses to show its peg mechanism under stress, you assume the worst.
The real story here is the hype cycle repeating. AI hardware is now where crypto ASICs were in 2018. Investors are pouring money into visions, not verifiable data. Etched’s $700M raise is a bet on the narrative that LVI will dominate inference. But without benchmarks, it is a bet on faith, not math.
Signal confirms. Action required: wait.
Floor holding? No. The floor is unconfirmed. Until Etched releases complete FLOPS, power consumption, and third-party benchmarks, this is a pass. The market is sideways. Chop is for positioning. Do not position on unverified claims.
Gas spike imminent. Wait. When the data drops, if it confirms the 80% MFU, the valuation will be justified. If not, the $700M will be a cautionary tale. The prudent move is to wait for the evidence. No data, no allocation.
Arb window closing. Execute. But the arb here is not on Etched. It is on the companies that can provide verifiable benchmarks. Watch for the first independent test. That is the real signal.
Takeaway: The biggest question is not whether Etched’s chips exist. It is whether they are as powerful as advertised. The answer is not in the press release. It is in the benchmarks. Until they are published, treat this as a narrative trade, not a fundamental one.