Jacob Zhao

Crypto × AI | Research • Writing • Investing | INTJ
EIR @IOSGVC | ex-@ArweaveSCP @Mirana @OKX_Ventures @Indodax
📬 Subscribe →

[Jacob Zhao @0xjacobzhao]

https://t.co/g8GVKF9liN

[Jacob Zhao @0xjacobzhao]

My study notes on @Huawei 's Tau (τ) Scaling Law full industry chain with the help of Claude + ChatGPT.

[Huawei @Huawei]

HUAWEI's Tau (τ) Scaling Law is a new principle for guiding the future development of semiconductors. By 2031, HUAWEI's high-end chips are expected to feature a transistor density equivalent to 14 Å (1.4 nm) processes. Watch the livestream to learn more! https://t.co/0Zb73r23ZP

[Jacob Zhao @0xjacobzhao]

🚀 AI计算芯片新架构五家公司对比
🔹 @cerebras：晶圆级 WSE 以“物理规模”换集群效率，主攻低延迟推理——收入已跑通，但重资产承压。
🔹 @tenstorrent：RISC-V 开放架构直击 CUDA 生态垄断——IP 设计顶尖，软件栈与生态采纳待考。
🔹 @SambaNovaAI：全栈数据流赌政企/主权 AI 偏好集成平台——卖基建方案，非纯芯片创业。
🔹 @Etched：Transformer 专用 ASIC 赌架构不迁移——高赔率二元对赌，成则爆发，败则清零。
🔹 @dMatrix_AI：数字存内计算卡位解码加速瓶颈——PCIe 标准形态，部署摩擦最低、落地最快。

[Jacob Zhao @0xjacobzhao]

🚀 5 AI Compute Architectures
Not “the next Nvidia” — five distinct bets:
🔹 @cerebras: Wafer-scale WSE bets that physical scale can reduce distributed-cluster overhead in low-latency LLM inference — revenue-proven, capex-heavy.
🔹 @tenstorrent: RISC-V + Tensix cores bet that open architecture can compete with CUDA lock-in over time — visionary IP, adoption still unproven.
🔹 @SambaNovaAI: Full-stack dataflow systems bet enterprises and sovereign AI buyers will prefer integrated AI platforms over standalone chips — infra play, not pure silicon.
🔹 @Etched: Transformer-specific ASIC bets the architecture will not shift — massive upside if it holds, existential risk if it doesn’t.
🔹 @dMatrix_AI: Digital in-memory compute bets decode acceleration is one of the nearest inference bottlenecks — PCIe form factor = fastest path to deployment.

[Jacob Zhao @0xjacobzhao]

🚀 AI compute isn't just a "GPU story".
I mapped the landscape across two axes:
🧵Narrative Strength × Commercial Maturity.

[ ⬆️ THE NARRATIVE TIER ]
@Etched @UbertiGavin @czhu1729
@cerebras @andrewdfeldman $CBRS,
@tenstorrent @jimkxa,
@dMatrix_AI @sidsheth
@SambaNovaAI @RodrigoLiang
→ Architectural disruptors.
→ Attacking memory bandwidth, interconnect, latency, inference efficiency & TCO.
→ High upside, commercial validation in progress.

[ ⬇️ THE FUNDAMENTALS TIER ]
@nvidia , @AMD , $MRVL, @Broadcom
→ Proven, validated businesses.
→ Judged by revenue, margins, ecosystems & FCF.
→ Commercial maturity is the moat.
The Hyperscaler Hedge: @Google (TPU) & @awscloud (Trainium) are quietly de-risking their Nvidia dependency.

#AI #Semiconductors #NVIDIA #AIChips #Investing

[Jacob Zhao @0xjacobzhao]

中文版：AI芯片赛道： 叙事强度 × 商业成熟度

⬆️ 分水岭上方：叙事框架
@Etched @UbertiGavin @czhu1729
@cerebras @andrewdfeldman $CBRS
@tenstorrent @jimkxa
@dMatrix_AI @sidsheth
@SambaNovaAI @RodrigoLiang
→ 架构颠覆者
→ 主攻显存带宽、互联、延迟、推理效率与TCO
→ 赔率高，商业验证进行中

⬇️ 分水岭下方：基本面框架
@nvidia @AMD $MRVL @Broadcom
→ 已验证的成熟业务
→ 用收入、毛利、生态与自由现金流说话
→ 商业成熟度即护城河

[Jacob Zhao @0xjacobzhao]

🚀 AI compute isn't just a "GPU story".
I mapped the landscape across two axes:
🧵Narrative Strength × Commercial Maturity.

[ ⬆️ THE NARRATIVE TIER ]
@Etched @UbertiGavin @czhu1729
@cerebras @andrewdfeldman $CBRS,
@tenstorrent @jimkxa,
@dMatrix_AI @sidsheth
@SambaNovaAI @RodrigoLiang
→ Architectural disruptors.
→ Attacking memory bandwidth, interconnect, latency, inference efficiency & TCO.
→ High upside, commercial validation in progress.

[ ⬇️ THE FUNDAMENTALS TIER ]
@nvidia , @AMD , $MRVL, @Broadcom
→ Proven, validated businesses.
→ Judged by revenue, margins, ecosystems & FCF.
→ Commercial maturity is the moat.
The Hyperscaler Hedge: @Google (TPU) & @awscloud (Trainium) are quietly de-risking their Nvidia dependency.

#AI #Semiconductors #NVIDIA #AIChips #Investing

[Jacob Zhao @0xjacobzhao]

我的评级：

美国模型
• Claude — MIT：最硬核理工校，代码、长任务、Agent 依然最强势。
• ChatGPT — Stanford：综合最均衡，专业能力、产品成熟度、大众可用性都强。
• Gemini — 哥大：搜索、写作、Google 生态、NotebookLM 强，但纯理工硬推理不如 Claude 和 ChatGPT。
• Llama — UCLA：曾经的开源旗帜，历史地位高、影响力大，但现在不是最顶的综合王者。
• Grok — ASU：靠 X 数据和话题性出圈，流量感强，但综合专业能力不在第一梯队。

中国模型
• DeepSeek — 中科大：国内最像理科疯子的模型，推理、代码、Agent 都很硬。
• Qwen — 浙大：大而全的开源基础设施型选手，体系完整、生态广、落地强。
• Kimi — 复旦：长文本、搜索整理、知识工作体验强，文感和信息整合好。
• GLM — 北航：工程国家队气质最重，正规、稳、To B 和 Agent 工程化强。
• MiMo（小米）— 南方科大：新锐黑马，冲劲足，Agent 和多模态能力不差。
• MiniMax — 厦门大学：更精致、更综合，多模态、语音、视频、文本都强。
• 豆包 — 中传：国民知名度高，娱乐性、产品体验、多模态和视频感最强，但专业硬核度不够顶。
• 混元 — 深大：背靠腾讯，能力中上，商业体系和业务嵌入强，走务实路线。
• 文心 — 吉林大学：起得早、底子厚、老牌正规军，但现在行业中心感和极客心智都弱了

[Jacob Zhao @0xjacobzhao]

10/
🚀 Prediction market agents are still early —
but as liquidity, data, and agent capability scale,
→ they converge toward a new form of
automated finance built on probabilities

📚Follow BroadNotes for more deep dives:
Substack → https://t.co/O4LteAg3MN
Paragraph → https://t.co/1VHVqlBFZa
🤝 Supported by @IOSGVC https://t.co/jzzCj5Ulw8

[IOSG Ventures @IOSGVC]

https://t.co/CEZeEHRBhT

[Jacob Zhao @0xjacobzhao]

🚨 Just published a new research report:
“Turning Probability into an Asset: The Rise of Prediction Market Agents.”

1/ Prediction Markets → A Truth Layer:
Prediction markets turn dispersed information into tradable probability signals backed by real capital, evolving from betting tools into a potential global truth layer, led by the Polymarket / Kalshi duopoly amid competition between regulated and crypto-native models.

2/ Prediction Market Agents: Not about better AI predictions, but executable probabilistic portfolio management—turning probability mispricing into automated trades via Data → ML analysis → Strategy & risk → Execution.

3/Strategy & Risk: Agents should focus on markets with clear rules, liquidity, and structured information, using deterministic arbitrage (resolution arbitrage, Dutch book, cross-platform spreads) plus structured signals. Risk is controlled via rule-based position sizing and strict risk modules.

4/Business Model: A sustainable stack combines B2B infrastructure (data/execution/backtesting), strategy ecosystems, and Agent/Vault performance participation. Products may evolve from signal tools → semi-automated trading → managed vaults.

5/Industry Stage: The space is still early. Players fall into three groups: infrastructure frameworks, autonomous trading agents, and analytics/execution tools.

🚀 We may be approaching the breakout moment for Prediction Market Agents.

[Jacob Zhao @0xjacobzhao]

https://t.co/uhEznVoBE8

[Jacob Zhao @0xjacobzhao]

9/
🧱 The ecosystem is forming, but still early. No standardized product yet combines:
strategy generation + execution efficiency + risk management + business loop

1) 🏗️ Infrastructure
@Polymarket and @gnosis_ provide the only standardized agent frameworks (data + execution), but remain access layers—leaving strategy, risk, and full trading systems to developers; others like @Kalshi are still at the API/SDK level. @PolymarketBuild @shayne_coplan @LiamKovatch @elynch46 @shaykevin @mansourtarek_ @luanalopeslara @koeppelmann

2) 🤖Autonomous Agents
@autonolas (Olas Predict): one of the most advanced stacks, with Omenstrat + Polystrat enabling LLM-driven strategies, mispricing detection, and automated execution with built-in risk controls via @pearldotyou @david_enim @Valorianxyz @tannedoaksprout @contentwillvary @hirschhe

@UnifaiNetwork tail-risk Polymarket agent targeting >95% near-settlement spreads @sunny_unifAI

@NetworkNoya building a research → execution stack (Omnichain Vaults live), still early in full closed-loop development @OptimisticOmni

3) 🧰Prediction Market Tools: Still concentrated in the 📊 information/analysis layer—closer to signals and research than full agents; ⚡ execution, 📦 position, and 🛡️ risk remain user-driven

📊Market Analysis: @UsePolyseer @Valyuofficial @yorkeccak · @oddpool_alerts @c0delemons · @poly_data @VaingloriousETH · @hash_dive · @polyfactual @basedlourie · @predlyai · @Polysights @tre_poh · @polyradar_io· @_alphascope_

🚨Alerts / Whale Tracking: @StandDOTtrade @lastridgely @wolfeypackey
⚖️Arbitrage: @arbbets · @PolyScalping · @eventarbitrage · @PredictionHunt @JosephAndrewFr2
💻Aggregated Execution: @matchrxyz · Verso @agpkeleta · @tradefoxai @Prithvir12 @yoshi_eth2

🔗 Strong recommend Awesome Prediction Market Tools: https://t.co/hhRJJjqJvP

[Olas (formerly Autonolas) @autonolas]

Polystrat is "one of the first consumer-grade trading agents for Polymarket." 💯

That is the verdict from @0xjacobzhao's latest article diving into the future of AI in prediction markets. ⬇️

The research features Olas Predict and Olas' two trading agents: Omenstrat and the newly launched Polystrat for Polymarket.

🔗 Try Polystrat: https://t.co/82Hw8yG4Mc

[Jacob Zhao @0xjacobzhao]

🙏 特别感谢以下机构与媒体的支持转载
@IOSGVC :https://t.co/GekOdZIOot
@Foresight_News ：https://t.co/119AgkvHOk
@PANews ：https://t.co/ETDcX26fhI
@ChainCatcher_ ：https://t.co/GJDm6z7ihP
@BlockBeatsAsia ：https://t.co/0W9MXU8Msa
@ABMedia_Crypto ：https://t.co/sIB5LkWKHw
@TechFlowPost ：https://t.co/QKFQXaP2wV
@ME_News ：https://t.co/YofYfGnSzj

🧠 本文为「博闻札记 BroadNotes」系列内容。
如果对你有启发，欢迎关注我的深度内容专栏👇
Substack：https://t.co/4Osa4B4oQZ
Paragraph：https://t.co/tzV0MJiD2q

[Jacob Zhao @0xjacobzhao]

🚨《让概率成为资产：预测市场智能体前瞻》

📌核心观点
1/预测市场通过“真金白银的交易”把分散信息压缩成可交易的概率价格信号，正从类博彩的下注工具演化为可被金融与企业系统直接调用的“全球真相层”，并在 Polymarket/Kalshi 双寡头与合规分发 vs 加密原生两条路径竞争、以及各国监管分化的背景下加速发展。

2/预测市场智能体的核心不是“AI预测更准”，而是作为 可执行的概率资产管理系统，通过 信息层（多源数据）→ 分析层（LLM/ML识别错价）→ 策略层（仓位与风控）→ 执行层（多市场自动交易与套利） 的四层架构，将概率偏差高效转化为可自动化执行的交易机会。

3/预测市场智能体应只在规则清晰、流动性充足且信息结构化的市场中运行，以确定性套利（结算套利、Dutch Book、跨平台价差等）为核心策略、结构化信息与信号跟随为补充，并通过阶梯信心法 + 固定仓位上限的规则化资金管理与常驻风控模块，将高频数据处理与跨市场执行优势转化为可持续收益，同时系统性避开内幕主导或高操纵风险的市场。

4/预测市场智能体的商业模式为“基建B2B变现（数据/执行/回测）+ 策略生态分成（第三方策略调用与权重）+ Agent/Vault业绩参与（管理费+绩效费）”，对应产品形态从娱乐化入口引流到订阅/信号与半自动执行（当下最可行），再到高门槛的托管Vault。

5/当前预测市场智能体生态仍处早期、尚无策略生成/执行效率/风控/商业闭环都成熟的标准化产品，整体可分为基础设施框架（如 Polymarket Agents）、自主交易Agent（如 Olas Predict/Polystrat）与工具终端（研究信号、鲸鱼警报、套利发现、聚合执行等。

🚀 我们或正处在预测市场智能体爆发的前夜。

[Jacob Zhao @0xjacobzhao]

https://t.co/AOsSGT80FZ

[Jacob Zhao @0xjacobzhao]

@Polymarket @Kalshi @gnosis_ 10/
尽管当前预测市场智能体生态仍处早期、尚未出现成熟的标准化产品，但随着流动性、数据基础设施与智能体能力的提升，有望逐步演化为一种新的自动化金融形态。

[Jacob Zhao @0xjacobzhao]

@IOSGVC @PANews @ChainCatcher_ @ABMedia_Crypto @BlockBeatsAsia @MetaEraCN @Foresight_News @TechFlowPost 《让概率成为资产：预测市场智能体前瞻》https://t.co/8npzMeT06m

[Jacob Zhao @0xjacobzhao]

https://t.co/AOsSGT80FZ

[Jacob Zhao @0xjacobzhao]

@IOSGVC @PANews @ChainCatcher_ @ABMedia_Crypto @BlockBeatsAsia @MetaEraCN @Foresight_News @Foresight_News 《让概率成为资产，预测市场智能体前瞻》https://t.co/rWZGK4OVDj