Teortaxes argues OpenAI and Anthropic's custom silicon projects are primarily fundraising pitches to show Nvidia independence

I can assure u Nvidia has better AI/ML systems for chip design than Anthropic or OpenAI Ant/OpenAI lack the decades of proprietary data for it (Nvidia has lots of that) Also, Clive is hired cuz Anthropic & Broadcom are working on a custom TPU and likely other ASICs Ant/OAI need a partnership with Cadence or Synopsys if they want data access

@zephyr_z9 yeah I don't think Nvidia will lose a lot of market share but there is the hope of frontier labs that they'll get the full stack somehow

@zephyr_z9 ant and oai don’t have ic design, and it’s not even a data problem.

@zephyr_z9 Do you think Google is a bigger long term issue for Nvidia? They’ve been co-designing AI software with TPU over a decade.

Would think they have the better AI long term. And thus maybe the better chip, or at least a shot at it, unless Nvidia’s AI matches it.

@zephyr_z9 Agree that Nvidia has a deep moat but these companies gotta try. Don’t think it will affect Nvidia’s dominant position but since these companies want to try building their XPUs and ASICs, it will at least be beneficial to companies like $AVGO $MRVL $mediatek $SNPS and $CDNS

@zephyr_z9 Huawei runs the same logic: 20+ years of Kirin silicon process data going into their EDA stack. that's why their modem designs keep surprising people. proprietary process data is the actual moat. the model is just the interface to it.

@zephyr_z9 They will lose market share if margins stay that high Chip design are currently being assisted by AI ,so shorter learning curves

@zephyr_z9 Sure. ANTHOPIC or OPENAI needs pay off the learning curve…

@zephyr_z9 Id agree with that. Another interesting point is whether the Labs eventually insource compute (i.e can they run more efficient training/inference workloads vs hyperscalers).

@zephyr_z9 Whose data does Cadence and Synopsys have that can be trained on?

@zephyr_z9 Wait, what? Cerebras used to partner with Cadence for developing chip design tooling

@zephyr_z9 사람들이 잊고 있는 사실 CPU 생산업체는 많이 있었다 GPU 생산업체도 많이 있었다 음...

@zephyr_z9 Jen Hsun might hire and pay a hitman with RTX5090 to anyone who dares to undermine him hmm

@zephyr_z9 Bro, this is the classic “data moat vs frontier lab” debate where everyone is confidently wrong in slightly different directions.

@zephyr_z9 The AI advantage increasingly boils down to who controls the silicon and the training data. ⚡️

@zephyr_z9 a bunch of chip talk where nvidia ends up being the underdog narrative feels off

they literally set the floor for this stuff

@zephyr_z9 hmm i wonder if the anti lag software is a multi-install thing

also genuinely never found it tbh

@zephyr_z9 yeah the data moat here is huge

nvidia's been collecting design data since forever, you can't just prompt engineer your way around that

@zephyr_z9 It feels like we're still in the very very early innings of everything. Mindblown .... !