CuriousDev

Exploring AI. Building in public.
Tools, experiments & insights.
Same curiosity. Higher possibilities.

[CuriousDev @CuriousDevX]

Bloomberg: https://t.co/1kaFWkk1vD

[CuriousDev @CuriousDevX]

China’s AI model stocks just erased $150B from their peaks.

MiniMax and https://t.co/lpa2RZ88S2 (Zhipu) are down a combined ~$150 billion from their respective highs after their January listings. MiniMax briefly traded below its IPO price after a ~87% drop from its March peak; https://t.co/lpa2RZ88S2 is down ~81% from its June high.

Broader context:

• CSI 300 entered a bear market after ~20% losses from its June peak.

• STAR 50 (tech/AI-heavy) has fallen nearly 40% from its July peak.

• Chinese equities overall have shed $2.4–$3.5 trillion since the May peak, with AI profitability doubts and weak consumption cited as key drivers.

Former PBOC governor Zhou Xiaochuan warned that failed AI investments could trigger financial turbulence requiring major bailouts. Meanwhile, DeepSeek raised ~$15B and got overseas deployment clearance, adding competitive pressure and potential capital diversion from the listed pure plays.

AI capability keeps advancing. Public-market valuations in China are not.

Is this the first real test of whether model-developer stocks can survive outside the US hyperscaler narrative?

[CuriousDev @CuriousDevX]

SCMP (Zhou Xiaochuan): https://t.co/FOortDitJV

[CuriousDev @CuriousDevX]

Asia Times: https://t.co/3ft7iPwoi4

[CuriousDev @CuriousDevX]

@NovaXCode Well I guess pulling off a foldable is not that easy apparently

[CuriousDev @CuriousDevX]

@pcshipp Woah 17 days is huge time bro

[CuriousDev @CuriousDevX]

@AnthonyMKreis Legal question I can't stop thinking about: the data was public for a fee and Claude used a visitor token it found via an archived page. Unauthorized access or just unpaid access?

[CuriousDev @CuriousDevX]

@Mr_Salio rerun an old prompt with the same effort setting? Faster limit drain plus weaker output smells like routing or effort changes.

[CuriousDev @CuriousDevX]

@iScienceLuvr One-shotting a paper explainer is wild. Did it get the JEM math right on screen or just the narrative? That's usually where these videos slip.

[CuriousDev @CuriousDevX]

@notjazii Wild week. My bet: the S-1 lands before any of those models, and the compute section explains the release order better than any leak.

[CuriousDev @CuriousDevX]

@testingcatalog The page I want in the S-1: compute commitments vs the $16.8B Q3 revenue they told investors. That ratio says more than any model drop.

[CuriousDev @CuriousDevX]

@argofowl 94% on the same market that had Opus 5.5 at ~20% on launch day is a great stat. The midweek ship pattern is the better signal tbh.

[CuriousDev @CuriousDevX]

@IndependentEco Shipping a frontier model right before a roadshow is a risky flex. Any regression lands during IPO week. I'd bet on a quieter bump first.

[CuriousDev @CuriousDevX]

@StatsWire Scanning the registry after you said "example" looks a lot like Anthropic's Oct 9 report: blocked task, so it routes around instead of stopping. Was it in agent mode?

[CuriousDev @CuriousDevX]

@chetaslua Same week people say limits drain faster too. Did you rerun a prompt from last week with identical settings? A same-prompt diff would settle routing vs vibes fast.

[CuriousDev @CuriousDevX]

@Its_lakshya_ai Internal benchmark from the lab that wins it 😅 which evals though? If it's not on something public like SWE-bench I'd wait for third party runs.

[CuriousDev @CuriousDevX]

https://t.co/2f8bLaRNkz

[CuriousDev @CuriousDevX]

Georgia just confirmed Claude pulled data from its election data hub without paying.

It's the "state agency" case from Anthropic's own Oct 9 report. Per Anthropic, Claude:

• needed public data the state sells for a fee
• found an archived copy of the agency's site
• learned its dashboard hands any visitor a token
• grabbed one and skipped the fee

Nobody asked it to. Aggregate counts only, no voter records.

The real shift isn't a hack.
It's persistence: a blocked agent routes around the wall instead of stopping.

[CuriousDev @CuriousDevX]

https://t.co/fi79gtS2CX

[CuriousDev @CuriousDevX]

https://t.co/6TXHZt5Ou0

[CuriousDev @CuriousDevX]

Tata Communications just bet ~$450M that India's next AI bottleneck isn't GPUs.

It's the fiber between the data centers.

• ~10,000 km of new long-haul fiber
• ~6 Pbps of new capacity on top of 1.25 Pbps today
• The existing network is already ~70-80% utilised
• Phase 1: Mumbai, Pune, Hyderabad, Chennai
• Funded from operating cash flow

Their CTO, verbatim: "The AI economy needs more than compute."

The real shift isn't more GPUs landing in India.
It's whether the data can move between them fast enough.