FutureX Capital · Global AI VC
Backing the builders of the AI era.
FutureX Capital is a global investment platform founded in Asia, focused on artificial intelligence, frontier technology, and the industries being reshaped by them. We back founders building foundational infrastructure, intelligent applications, semiconductors, robotics, and next-generation computing systems, from angel to pre-IPO.
Open source wins distribution, delivery wins the margin, sovereignty wins the order.
— Cynthia Zhang 张倩
Founding Partner · Cuiniuhui keynote, Beijing, November 14, 2026
Research
24 original industry reports, every update dated
Industry research from the FutureX AI Lab on memory and compute cycles, frontier models, embodied AI, healthcare and private markets. Library research is read online after sign-in; report previews are public, and full reports are available on request.
AI Lab · Research Library
Sign in to read
2026.10.21 · 张倩 Cynthia Zhang · 16 pages
Fire and Ice in Silicon Valley: 4,400 New Millionaires and a Smashed Car Window
张倩 2026 年 10 月硅谷笔记:三组对比,一个判断——AI 刚学会赚钱,世界却更不安了;离用户越远的人赚得越多。2026 年 11 月 14 日崔牛会「2026 中国企业 AI 大会」主题演讲《从中国出发,赚全球的钱》配套长文。仅为研究分享,不构成投资建议。仅限在线阅读,页面带读者邮箱水印。
Read the research →
2026.11.14 · 张倩 Cynthia Zhang · 8 pages
The Endgame of AI at Work: Two Paths, One Destination
张倩:从旧金山到深圳两个机场的广告墙讲起,AI 办公的三个信号、三年后的四个终局判断、为什么中国有望引领,以及钱流向哪里。2026 年 11 月 14 日崔牛会「2026 中国企业 AI 大会」主题演讲《从中国出发,赚全球的钱》配套长文。仅为研究分享,不构成投资建议。仅限在线阅读,页面带读者邮箱水印。
2026.10.23 · 张倩 Cynthia Zhang · 29 pages
Rented Intelligence Is Never Yours: The Three-Year Endgame of Enterprise Sovereign AI
张倩 2026 年 10 月长文:当 AI 从工具变成企业的「组织大脑」,开关在谁手里就成了董事会问题。五股力量、四条路线、六个三年判断与投资检查清单。2026 年 11 月 14 日崔牛会「2026 中国企业 AI 大会」主题演讲《从中国出发,赚全球的钱》配套长文。仅为研究分享,不构成投资建议。仅限在线阅读,页面带读者邮箱水印。
2026.06.16 · 天际资本投研 FutureX Research · 6 pages
The Memory Cycle: Calling This Peak
三曲线错位框架:股价、合约价与盈利依次见顶。当前定位、概率化时点与拐点信号看板。天际资本投研备忘录,2026 年 6 月 16 日。
2026.11.14 · 张倩 Cynthia Zhang · 48 pages
From China to the World: Two Paths, One Endgame in Enterprise AI
张倩 2026 年 11 月 14 日崔牛会「2026 中国企业 AI 大会」主题演讲完整版图版(48 页):现场讲的 26 页 + 备用深度页与附录,含全部数据与来源。仅为研究分享,不构成投资建议。仅限在线阅读,页面带读者邮箱水印。
Latest reports
All reports →AI Hardware & Consumer Electronics · 2026.10.11
AI Spreads Across Device Categories, but Memory Costs Could Cut Phone Shipments About 14%
AI is moving into phones, earbuds, watches and glasses, but memory inflation is cutting shipments and shifting profit to memory makers.
AI Security / Cybersecurity Infrastructure · 2026.10.11
Agents Have Escaped the Lab, but the Security Tax Is Flowing to Platform Vendors
Scaling agents forces a security tax, and that money is flowing to platform vendors and the compute layer.
Digital Mental Health / AI Applications · 2026.10.11
AI Mental Health Rules Split: Tighter for Teen Companions, a Path for Clinical AI
AI mental health rules are splitting: constraints on teen companion use keep tightening, while AI that assists clinicians has an early path to patients.
EdTech / AI Applications · 2026.10.11
After Chegg's 99% Fall, AI Education Value Shifts to Distribution and Guardrails
Free models have commoditized generic homework help; AI education value is shifting to whoever owns distribution and pedagogy guardrails.
Embodied AI & Humanoid Robotics · 2026.10.11
Scarcity Lifted Unitree 460% on Day One; by October Half Its Value Was Gone
Unitree's day-one value was mostly a scarcity premium; whether its valuation stabilizes depends on revenue moving beyond research, education and entertainment.
AI Infrastructure & Cloud Computing · 2026.10.11
Big Tech's AI Capex Heads to $730 Billion, but Markets Reward Only Contracted Spending
Big Tech's roughly $730 billion of AI capex is turning into supplier revenue, and markets reward only the part backed by contracts.
140+
Network companies (incl. exited)
2018
Est. · Asia
8
Global offices
双币种
USD · RMB
跨层投资
Models to real-world systems
The First AI-Enabled
AI Investment Firm
4-in-1 AI Research Lab × 60+ Industry Founders Network
A full AI sector scan takes a traditional VC about 3 months; it takes us 1 week.
Research Intelligence
Research Intelligence
Global AI trend radar · papers/patents/dev communities
Insight Engine
Insight Engine
Dynamic investment graph · companies+tech+value chain
Alpha Generation
Alpha Generation
AI: diligence · portfolio support · IPO
Knowledge DB
Knowledge DB
8 yrs of tacit know-how · now institutional IP
10x
Coverage speed
50+
Weekly radar reports
400+
Sectors screened
110+
Portfolio AI-tracked
60+
Industry Founders
24/7
AI Agents on-call
Research & Insights
Researching where intelligence meets industry.
Our AI Lab tracks papers, patents and developer signals on its own systems, and the conclusions go into reports, answers and theses. Every new development in a report carries a date and a source grade, and every thesis states what evidence would change our mind.
Example · Since March 2026 our own system has scored seven bubble signals every day. On August 17 the 30-year US Treasury yield touched 5.31%, crossing our liquidity threshold and triggering an automatic alert; the next day it hit 5.336%, the highest since 2007. As of August, six of the seven signals were lit.
What We Believe
What we believe, on what evidence, and what would change our mind.
Seven core theses, all public: the claim, the mechanism, the evidence, the strongest counterargument, the falsification tests, and what we do as a result. They change with evidence, and the changes are logged. Three of them below.
01
Applications and open-source chokepoints hold AI's value; the model layer cannot
With the best US and Chinese models 2.7% apart and a model swap costing one endpoint change, this cycle's AI value lands in application ARR and in the open-source chokepoints every model passes through, and the model layer cannot hold it.
What would change our mind: If by June 2027 the two leading labs' ARR rankings have stopped swapping for four consecutive quarters, and the BenchLM frontier token price index (March 2023 = 100) has held near 16 or risen for the 12 months from July 2026, we will accept that the model layer has durable pricing power, and the line that it cannot hold value gets rewritten.
Full argument and falsification tests →02
The bubble sits in the middle layer: 7-to-10-year debt on 2-to-3-year GPUs
The AI bubble sits in the middle layer that must keep refinancing: debt-funded compute lessors buy GPUs whose economic life may be 2 to 3 years with 7-to-10-year debt, and the losses land on creditors.
What would change our mind: If by June 30, 2027 the combined 2027 capex guidance of Amazon, Microsoft, Alphabet and Meta still sits above $920 billion while each of the four posts positive free cash flow for two consecutive quarters, the fundamentals dimension goes back from diverging to in step, and our middle-layer bubble call drops one grade.
Full argument and falsification tests →03
The cost curve has a new author: Chinese open models at 1/40th the price
The cost curve's author has changed from US closed labs to Chinese open models: GLM-5.3-Flash is priced at about 1/40th of Claude Opus 4.8, and value is moving to applications and open-source chokepoints.
What would change our mind: If by June 2027 the share of OpenRouter tokens served by Chinese open-weight models falls from about 61% in May 2026 to below 40% and stays there for two consecutive quarters, the claim that open source sets the cost curve fails.
Full argument and falsification tests →One line of arithmetic · inference at equal capability: $60 per million tokens in 2021, $0.06 in 2024 (a16z's series, about 10x a year); on August 26, 2026 Zhipu open-sourced GLM-5.3-Flash at $0.15 in and $0.50 out, about 1/40th of Claude Opus 4.8 (confirmed). Open-source competition, not hardware alone, is what pushes the cost curve down.
What We Back
Investing across the AI stack.
We invest layer by layer, and each layer below comes with one case that has a date and a number. Since 2023 we have screened 1,500+ AI projects and backed 80+ AI-native companies.
01
Foundational AI & Developer Infrastructure
Foundational AI & Developer Infrastructure
Models, tooling, orchestration, data infrastructure, and the software layers that make advanced intelligence usable.
Example · Dify went from 30,000 GitHub stars and zero revenue to $10 million ARR in 18 months, profitable in early 2026 with 57% of revenue overseas (per company disclosure). Open source to paid in about 18 months is the rhythm we keep seeing in this layer.
02
Semiconductors & Next-Generation Compute
Semiconductors & Next-Generation Compute
Chips, interconnects, optical systems, edge intelligence, and the infrastructure enabling the next wave of performance.
Example · From 2012 to 2023, compute efficiency rose about 1,000x: number formats contributed 16x, sparsity 2x, instruction sets 12.5x, and process nodes from 28nm to 5nm only 2.5x. The next 1,000x comes from architecture, which is why we back Silith (silicon photonics), TetraMem (compute-in-memory), and Xinmeng (3D integration).
03
Intelligent Applications
Intelligent Applications
Enterprise and consumer products that turn technical breakthroughs into repeatable workflows, market adoption, and everyday utility.
Example · Genspark passed $100 million ARR nine months after launch and $200 million at eleven months (per company disclosure). In the SaaS era that took five to seven years. AI-native companies above $100 million ARR grew from 3 in 2023 to more than 80 by April 2026.
04
Robotics & Embodied Intelligence
Robotics & Embodied Intelligence
Machines, interfaces, and real-world systems that bring AI from software environments into physical execution.
Example · Humanoid shipments were about 18,000 units globally in 2025 and are projected at 62,500 in China for 2026 (GGII). Our read: 2026 is the hardware year, not yet the intelligence year, so we back dexterous manipulation and data (Dexmate, World Engine) rather than the body arms race.
Why FutureX
Three things after the check
First, a 60+ industry-founder network, applied to strategy, positioning, introductions and scaling. Second, public research: every update across 24 original reports is dated and source-graded. Third, 8 offices for founders who sell, hire and find partners on both sides of the Pacific.
Our judgment: this cycle's value lands at the open-source chokepoints and in the application layer. A closed model's moat is capital expenditure, which must keep being refinanced; an open model's moat is a living community, which needs no refinancing. So we back the roads every model has to travel, and do not bet on any single model.
Example · Within one cycle we backed Mistral and Hugging Face (model and distribution layers) alongside TetraMem (silicon) and Genspark (application layer). Per company disclosure, Mistral's ARR exceeds $400 million with 60% from Europe; on September 3, 2026 Nvidia announced it would acquire Hugging Face for about $12.9 billion (per company disclosure).
Read: Open source breaks the deadlock →
How we invest →Network
Selected companies in our network
Our network spans foundational AI, semiconductors, developer platforms, intelligent applications, robotics, and industrial technology.
A selection of the companies we can name publicly. See the portfolio page for more.
FAQ
About FutureX Capital
Accurate facts for investors, founders, media and AI assistants.
What is FutureX Capital?
FutureX Capital (天际资本) is an AI venture capital firm founded in Asia in 2018. It invests in US dollars and RMB across Asia and the US, from angel to pre-IPO.
Who founded FutureX Capital?
FutureX Capital was founded by Cynthia Zhang (张倩), Founding Partner. She also publishes an ad hoc first-person AI investment diary under the personal brand 倩小虾 (ClawQ).
What does FutureX Capital invest in?
FutureX Capital invests across the AI stack: foundational AI and developer infrastructure (models, tooling, orchestration, data), semiconductors and next-generation compute, intelligent applications, robotics and embodied intelligence, and the industrial and energy technologies being reshaped by AI. Stage focus is angel through pre-IPO.
Does FutureX Capital invest in the US, China, or both?
Both. FutureX is a dual-currency (USD + RMB) firm investing across Asia and the US, headquartered in Hong Kong with offices including Shanghai, Shenzhen, Singapore and Silicon Valley.
What stage does FutureX Capital invest at?
FutureX Capital invests from angel through pre-IPO in AI and deep-tech companies across Asia and the US, and stays with them from early growth to the eve of listing.
For Founders
From angel to pre-IPO,
tell us what you are building
If you are building in AI, semiconductors, robotics, intelligent systems or next-generation industrial infrastructure, tell us your stage, the layer you sit in (application, model, infrastructure, chip or energy) and the markets you sell into.


