Burry flags Etched as Nvidia's most credible chip challenger yet
Michael Burry, the investor who famously shorted the housing market before the 2008 crash, is now betting against Nvidia's dominance in AI chips — and he has a specific name in mind. Burry highlighted Etched, a Harvard-dropout-founded startup, as presenting "serious competition" for Nvidia in the AI inference hardware market, sending the signal that Wall Street is starting to take challengers seriously.
Etched isn't vaporware. The company has raised $800 million across four financings at a $21 billion valuation, led by Jane Street, and its A0 silicon has already come back from TSMC's N4P process. With a team of over 400 engineers drawn from Nvidia, Google's TPU division, Broadcom, SK Hynix, and TSMC, Etched claims it has $1 billion in demand and is validating its first rack-scale product with customers. The startup's pitch centers on two technical breakthroughs: Low Voltage Inference, which lets its chips run at under half the voltage of typical AI silicon to avoid thermal throttling, and Cluster Scale Memory, a hybrid HBM/SRAM architecture that promises SRAM-level latency without sacrificing capacity. If Burry is right, Nvidia's moat may not be as deep as the market assumes — the inference market, where most AI workloads will eventually land, is wide open for purpose-built hardware that can undercut GPU economics.
US startups are switching to Chinese LLMs to slash costs by up to 90 percent. CNBC, the Financial Times, and NPR all reported this week that American companies are increasingly routing workloads to models from Alibaba, ByteDance, and DeepSeek instead of paying for OpenAI or Anthropic APIs. The Financial Times framed it as an outright price war, with Chinese providers undercutting US frontier labs on both per-token pricing and fine-tuning costs. For startups operating on tight budgets, the math is simple: a Chinese model that delivers 85 to 90 percent of GPT-4-level performance at a tenth of the cost is good enough for most production use cases. This isn't just a China-domestic phenomenon — it's a structural shift in how the global AI market prices intelligence, and it pressures US labs to justify their premium on the basis of safety, compliance, and marginal quality gains rather than raw capability.
Does cheaper inference from Chinese models change your calculus on which provider to build with? Tell us in the comments.
