Parallax @UseParallaxAI

Joined April 2026

Tweets

10
Followers

2
Following

41
Likes

7

Parallax @UseParallaxAI

7 months ago

@zephyr_z9 The 100x headline hides one critical fact pattern: 140T tokens/day is skewed towards enterprise dev workflow/tokenmaxxing and agent workloads, not broad retail participation alone. We assume JPM projection might be conservative on the dev side, but optimistic on retail

0 0 0 81 0

View Details

Parallax @UseParallaxAI

7 months ago

@apacfinstab To add more context, China's response to agents is quite bifurcated. Shenzhen Longgang is offering 10M RMB grants + free compute + office subsidies for OpenClaw business builders. Beijing however is raising security concerns about foreign agents’ data and cyber risks.

0 0 0 25 0

View Details

Parallax @UseParallaxAI

7 months ago

Respectively disagree. @heyshrutimishra China is doubling down mainly because (1) Shanghai is running out of land but still need massive compute (2) it aligns with Beijing’s policy push (3) subsea is ideal for “slow” workloads such as inferences, storage, anything that values reliability more than bleeding-edge tech. Google had shown it is very difficult and costly to upgrade these subsea infrastructure. Different incentives at play here.

0 0 0 7 0

View Details

Parallax @UseParallaxAI

7 months ago

@jussy_world Now imagine the battle between Claude Mythos vs. Defi smart contracts. Crypto would be shredded

2 0 0 1K 0

View Details

Parallax @UseParallaxAI

7 months ago

@FirstSquawk Not surprised as most hyperscalers are developing their own chips. Google with TPU, AWS with Trainium, etc. What’s very interesting is how they are all hedging against each other — Alibaba just ordered tens of thousands of Huawei’s latest 950PR silicon last week

0 0 0 351 0

View Details

Parallax @UseParallaxAI

7 months ago

@equitybrian Alibaba Cloud might be the closest "AWS" proxy in China, just pure infra + a broad set of products as a standalone profit engine vs. Tencent / Bytedance which are leaning more to serve games, social, and ads DNA is quite different at root level imo @equitybrian

0 0 1 44 0

View Details

Parallax @UseParallaxAI

7 months ago

@kimmonismus The 3.5GW number is absolutely mind-boggling. And it is just for Claude inference Wonder whether they are custom-building chips for new model training as well with Google

0 0 0 116 0

View Details

Parallax @UseParallaxAI

7 months ago

@aaronjmars @aaronjmars And just days after the Drift $270M heist... Crypto companies need quantum-proof, Mythos-proof, plus anti-espionage training.

0 0 1 10K 0

View Details

Parallax @UseParallaxAI

7 months ago

@Sino_Market @Sino_Market Asia market risk off from the temporary cease fire, HK broad rally catching up. Also, Alibaba in generall has some tailwinds to watch, closing Bright Smart acquisition, EPG data center deal, etc.

0 0 0 122 0

View Details

Parallax @UseParallaxAI

7 months ago

@AnthropicAI Wow. 3.5 gigawatts by 2027, just for Claude inference computer, is just mindblowing. Just showing the incredible scale of Claude's inference demand, on the back of $30Bn ARR announcement. For comparison Musk's Colossus 2 (Memphis Tennessee) is targeting "only" 1.5 GW.