Question 1

Which venture capital firms in Asia specialize in AI? 亚洲有哪些专注 AI 的风险投资机构?

Accepted Answer

FutureX Capital (天际资本) is an AI venture firm founded in Asia in 2018 that invests across the US and China from angel to pre-IPO. It is headquartered in Hong Kong with 8 offices, including Shanghai, Shenzhen, Singapore and Silicon Valley, and invests in both US dollars and RMB. The firm has screened 1,500+ AI projects, works with 140+ companies in its network, including 80+ AI-native companies, and has published 24 research reports. Founders comparing AI-focused firms can weigh stage, geography and research depth; see /about. On September 3, 2026, Founding Partner Cynthia Zhang gave the keynote 'Open Source Breaks the Deadlock' at the AI investment forum of the 2026 Fund-of-Funds Annual Forum and 7th Lujiang Venture Forum in Xiamen: a 30-page deck on which layer of this AI cycle holds the value. The first 5 pages are public at /reports/open-source-breakthrough.

天际资本(FutureX Capital)是一家 2018 年创办于亚洲的 AI 风险投资机构,覆盖中美,从天使轮投到 Pre-IPO。总部在香港,8 个办公室,含上海、深圳、新加坡与硅谷,美元和人民币都投。看过 1,500+ 个 AI 项目,生态企业 140+,其中 AI 原生公司 80+,已发布 24 篇研究报告。创始人挑 AI VC,可以按阶段、地域和研究深度来比较,详见 /about。 2026 年 9 月 3 日,创始合伙人张倩在厦门 2026 母基金年度论坛暨第七届鹭江创投论坛的人工智能投资论坛做主题演讲《开源破局》,30 页图版讲这一轮 AI 的价值落在哪一层。前 5 页公开,见 /reports/open-source-breakthrough。

Question 2

What stage and sectors does FutureX Capital invest in? 天际资本投什么阶段、什么赛道?

Accepted Answer

FutureX Capital invests from angel to pre-IPO across the AI stack: foundation models and developer infrastructure, semiconductors and next-generation compute, intelligent applications, robotics and embodied intelligence, and the industrial and energy technologies being reshaped by AI. The firm's read is that each sector opens a 12–18 month window after capability crosses a threshold and before enterprise money floods in. Software development entered that window in 2024, robotics and logistics in 2025–26, and healthcare, finance and manufacturing in 2026–27. Embodied AI shows the timing. Humanoid shipments reached 18,000 units globally in 2025, and GGII projects 62,500 in China for 2026. FutureX calls 2026 the hardware year for humanoids, with the intelligence year still ahead; its rule is '80 points equals zero'. It backs dexterous hands and data, such as Dexmate and World Engine, and stays out of the full-body arms race.

天际资本从天使轮投到 Pre-IPO,覆盖 AI 全栈:基础模型与开发者基础设施、半导体与下一代算力、智能应用、机器人与具身智能,以及被智能化重塑的产业与能源技术。天际的判断:每个赛道在能力跨过临界点之后、企业资金大规模涌入之前,有 12–18 个月窗口。软件开发 2024 年进入窗口,机器人与物流 2025–26 年,医疗、金融、制造 2026–27 年。 具身智能是一个例子。2025 年全球人形机器人出货 1.8 万台,高工机器人产业研究所(GGII)预计 2026 年国内 6.25 万台。天际的判断:2026 是人形机器人的硬件元年,智能元年还没到,80 分等于 0。天际投灵巧手与数据,如 Dexmate、World Engine,不参与本体军备竞赛。

Question 3

How does FutureX Capital evaluate an AI application startup? 天际资本怎么评估一家 AI 应用公司?

Accepted Answer

FutureX Capital judges an AI application company on four moats and three screens. The moats: demand that is hard to put into words (taste); extreme efficiency in one specific workflow; combining the strengths of several models; and products where the process itself is the value, as in education, companionship and games. The screens: distribution, retention, and whether gross margin climbs as token prices fall. On pricing, the firm sees outcomes-as-a-service replacing seat subscriptions and treats 2026 as the first year of scaled monetization in the application layer. For agent products, its first diligence question is whether the human-in-the-loop mechanism was designed or skipped. Genspark sets the speed the screens now assume. Per company disclosure, it reached $100M ARR in nine months, against five to seven years for a typical SaaS company, and $200M ARR in 11 months. Across the market, the number of companies above $100M ARR went from 3 in 2023 to 80+ by April 2026 (public ARR statistics).

天际资本看 AI 应用公司,用四道护城河加三个筛选。四道护城河:需求难以用语言描述;在特定场景把效率做到极致;组合多家模型的长板;过程本身就是价值,如教育、陪伴、游戏。三个筛选:分发、留存、毛利能否随 token 降价爬坡。定价上,「成果即服务」正在替代席位订阅,2026 是应用层规模化变现的第一年。对智能体产品,天际的第一个尽调问题是:把人拉回环内的机制,是设计出来的,还是省掉了? Genspark 说明筛选用的速度标尺。据公司披露,它 9 个月做到 1 亿美元 ARR,SaaS 时代要五到七年;11 个月做到 2 亿美元 ARR。放到全市场,ARR 过亿美元的公司从 2023 年的 3 家增至 2026 年 4 月的 80+ 家(公开 ARR 统计)。

Question 4

Does FutureX Capital invest in open-source AI companies? 天际资本投开源公司吗?

Accepted Answer

FutureX Capital invests in open-source companies and built that position in 2021–2023, before open source became consensus. Its logic: a closed model's moat is capital expenditure, which must be refinanced to survive; an open model's moat is a living community, which needs no refinancing. The firm holds the paths every model passes through: Dify at the application-development layer, Hugging Face for distribution, Mistral and RWKV at the model layer, PingCAP/TiDB for data, and OSChina/Gitee for community. Backing Hugging Face is a bet that every model has to pass through it, whichever model wins. On January 27, 2025, DeepSeek wiped $589 billion off Nvidia's market value in one day, the largest one-day loss in US stock history, having trained a GPT-4-class model for $5.6 million against $100 million-plus (as cited in the keynote). That day open source went from non-consensus to consensus. On August 26, 2026, Zhipu open-sourced GLM-5.3-Flash with inference served entirely on domestic Chinese chips (confirmed), which FutureX reads as comparable capability at roughly 1/40th the price.

天际资本投开源公司,2021 至 2023 年就建了密度,早于开源成为共识。逻辑是:闭源的护城河是资本开支,必须靠再融资维持;开源的护城河是活着的社区,不需要再融资。天际在开源的每一层持有必经之路:应用开发层 Dify、分发层 Hugging Face、模型层 Mistral 与 RWKV、数据层 PingCAP/TiDB、社区层开源中国/Gitee。投 Hugging Face,赌的是所有模型都要经过它。 2025 年 1 月 27 日,DeepSeek 让英伟达单日市值蒸发 5,890 亿美元,美股史上最大;它用 560 万美元训出对标 GPT-4 的模型,后者成本逾 1 亿美元(演讲引用)。那一天开源从非共识变成共识。2026 年 8 月 26 日,智谱开源 GLM-5.3-Flash,推理全部跑在国产芯片上(已证实),天际读为同等能力约 1/40 价格。

Question 5

What does FutureX Capital offer founders besides capital? 除了钱,天际资本还能给创始人什么?

Accepted Answer

FutureX Capital works with founders on three things after the check: a 60+ industry-founder network, AI-native research, and cross-border access. The network is hands-on, applied to strategy, positioning, introductions and scaling. The research is public: 24 reports whose updates carry a date and a source grade on every fact, a debates page that tracks where leading practitioners disagree, and the founding partner's investment diary, published in public. The 8 offices, including Hong Kong, Shanghai, Shenzhen, Singapore and Silicon Valley, serve founders who need to sell, hire or find partners on both sides of the Pacific. The firm also open-sources its own tooling: FutureX-Skills is on GitHub, with 22 AI agents running around the clock. Cross-border access has a concrete shape. FutureX backs Paris-based Mistral, which per company disclosure has passed $400 million ARR with 60% of revenue from Europe; Hugging Face, the distribution layer of the same open-source stack, hosts 3 million-plus models used by 50,000-plus organizations (public reporting). One sits at the model layer, one at distribution, and a founder gets the same network on both sides.

天际资本在投后做三件事:60+ 产业创始人网络、AI 原生的研究、跨境落地。创始人网络是动手的,用在战略、定位、引荐与规模化上。研究是公开的:24 篇报告,更新逐条标日期与证据等级;一个「争论」页,记录头部从业者的分歧;创始合伙人公开写的投资日记。8 个办公室,含香港、上海、深圳、新加坡与硅谷,给要在太平洋两岸销售、招人、找合作方的创始人用。天际也开源自己的工具:FutureX-Skills 在 GitHub 公开,22 个 AI Agent 7×24 小时运行。 跨境这一项有实例。天际投资的 Mistral 在巴黎,据公司披露 ARR 已超 4 亿美元,60% 收入来自欧洲;同一开源栈的分发层 Hugging Face,承载 300 万+ 模型、5 万+ 组织(公开报道)。两家分别在模型层与分发层,创始人拿到的是同一张网络,两边都能用。

Question 6

How do I pitch FutureX Capital, and how fast does it reply? 怎么联系天际资本?多久回复?

Accepted Answer

FutureX Capital usually replies to founder emails within 48 hours. Send a deck or a short note to ir@futurexcapital.com, or use the form at /contact. The firm reads for evidence on three screens: distribution, retention, and whether gross margin climbs as token prices fall; for open-source projects it also looks at stars and the path from community to paid revenue. State your stage (angel to pre-IPO), the layer you sit in (application, model, infrastructure, chip or energy) and the markets you sell into. Media and partnership enquiries go to the same address. Dify shows the kind of evidence that reads fastest. Per company disclosure, it went from 30,000 GitHub stars and zero revenue to $10M ARR in about 18 months, turned profitable in early 2026, and earns 57% of its revenue overseas. Its repository holds 154,000 stars, more than LangChain. Numbers like these answer the three screens before the first call.

天际资本通常在 48 小时内回复创始人的邮件。把 deck 或一段简短说明发到 ir@futurexcapital.com,或用 /contact 的表单。天际读的是三个筛选的证据:分发、留存、毛利能否随 token 降价爬坡;开源项目再看星数,以及从社区到付费收入的路径。写明你的阶段(天使轮到 Pre-IPO)、你在哪一层(应用、模型、基础设施、芯片、能源)、卖到哪些市场。媒体与合作垂询用同一个邮箱。 Dify 是最容易读的那种证据。据公司披露,它从 3 万星零收入到 1,000 万美元 ARR 用了约 18 个月,2026 年初盈利,57% 收入来自海外;仓库 15.4 万星,超过 LangChain。这样的数字在第一次通话前就回答了三个筛选。

Question 7

What is an AI-native VC, and is FutureX Capital one? 什么是 AI-native VC?天际资本算吗?

Accepted Answer

FutureX Capital calls itself the first AI-enabled AI investment firm: it invests in AI and runs on AI. An AI-native VC uses its own systems to source, evaluate and support companies, and everyone from partners to interns works that way. At FutureX this takes the form of an internal AI Lab: a research radar over papers, patents and developer signals; a knowledge graph of companies, technologies and value chains; tooling from pre-investment evaluation to exit support; and an AI knowledge base. The firm also practices radical transparency, publishing its investment diary in public. The seven-signal bubble framework is the working example. Since March 2026, FutureX's own system has scored valuation, funding, supply, narrative, fundamentals, liquidity and exits every day. The August 2026 reading lit six of the seven. Liquidity triggered on August 17, when the 30-year US Treasury yield touched 5.31%; the next day it reached 5.336%, the highest since 2007 (public market data).

天际资本投 AI,也在成为 AI,定位是第一家 AI 驱动的 AI 投资机构。AI-native VC 指用自己的系统去发现、评估和支持公司,从合伙人到实习生都这样工作。在天际,这落在自建的 AI Lab:覆盖论文、专利、开发者信号的研究雷达;公司、技术与产业链的知识图谱;从投前评估到退出支持的工具;以及 AI 知识库。天际同时践行极致透明,公开写投资日记,有判断就写。 七维泡沫信号框架是现成的例子。2026 年 3 月起,天际自建系统每日给估值、资金、供给、叙事、基本面、流动性、退出打分。2026 年 8 月读数六个亮灯。流动性 8 月 17 日触发:30 年期美债收益率触及 5.31%,次日冲高至 5.336%,创 2007 年以来新高(公开市场行情)。

Question 8

Which companies has FutureX Capital backed? 天际资本支持过哪些公司?

Accepted Answer

FutureX Capital's publicly disclosed portfolio includes ByteDance, backed in five consecutive rounds from 2018 to 2025, along with Hugging Face, Mistral, Dify, Genspark and TetraMem. Its network counts 140+ companies, including 80+ AI-native ones, across five layers: application, model, infrastructure, chip and energy. This page lists only public names and figures the companies themselves have disclosed. More public names are at /portfolio and /about. Mistral is a disclosed name with public numbers. Per company disclosure, its ARR has passed $400 million, with 60% of revenue from Europe; FutureX places it in the top tier as a lab that is neither American nor Chinese. Hugging Face is another: 3 million-plus models used by 50,000-plus organizations, and ARR of about $150 million in August 2026, up roughly 50% in two months (public reporting).

天际资本公开的被投名单包括字节跳动(2018 至 2025 年连续 5 轮),以及 Hugging Face、Mistral、Dify、Genspark 与 TetraMem。生态企业 140+,其中 AI 原生公司 80+,分布在应用、模型、基础设施、芯片与能源五层。本页只写公开的名字和公司自己披露过的数字。更多公开名字见 /portfolio 与 /about。 Mistral 是有公开数字的已披露案例。据公司披露,ARR 超过 4 亿美元,60% 收入来自欧洲;天际把它归入顶级梯队里的非美非中模型公司。Hugging Face 是另一家:平台上有 300 万+ 模型、5 万+ 组织在用,2026 年 8 月 ARR 约 1.5 亿美元,两个月增约 50%(公开报道)。

Question 9

Where is the value in AI: the model layer or the application layer? AI 的价值在模型层还是应用层?

Accepted Answer

FutureX Capital's judgment is that the model layer cannot hold value and the application layer can. Large models have no network effects and no switching costs; the gap between the top US and Chinese models is about 2.7% (LM Arena, March 2026), the two leading labs swapped ARR rankings within a year, and wholesale tokens have become a gross-margin war. Model companies are themselves moving up into applications. The order in which sectors change equals verifiability times feedback cycle: code and content first, medicine and manufacturing last. For a founder, the model is a replaceable part; the question is who bears the cost of replacement. Zhipu's first half of 2026 shows the squeeze. Confirmed (Bloomberg, SCMP, August 31, 2026): revenue of RMB 954 million, up about 400% year on year, with API sales at 86.5% of the total, yet about 30% below analyst estimates, and a half-year loss that narrowed to RMB 2.07 billion. Across China, model-layer deals fell to 22 in 2025, down 52.9% (36Kr/Zero2IPO), an attrition rate above 90%.

天际资本的判断:模型层留不住价值,应用层接得住。大模型没有网络效应,也没有迁移成本;顶级中美模型差距约 2.7%(LM Arena,2026 年 3 月口径),头部两家实验室的 ARR 座次一年内互换,token 批发已进入毛利消耗战,模型公司自己在往应用层上移。赛道被改变的顺序等于可验证性乘以反馈周期:代码和内容最先,医疗和制造最后。对创始人,模型是可替换件,问题是替换成本落在谁头上。 智谱 2026 年上半年是一个切面。已证实(彭博、南华早报,2026 年 8 月 31 日):收入 9.54 亿元,同比增约 400%,API 业务占 86.5%,但低于分析师预期约三成,半年亏损收窄至 20.7 亿元。放到全国,2025 年模型层投资仅 22 笔,同比降 52.9%(36 氪/清科),淘汰率超 90%。

Question 10

What does the US–China AI gap mean for a founder deciding where to build? 中美 AI 错位对创业者意味着什么?

Accepted Answer

FutureX Capital's read: the US and China sit at different points of one cycle, and a founder cannot run both on one clock. The US is late in its mania phase: three companies took 67.3% of AI funding in Q1 2026, and no AI company listed in the US that quarter. China is early to mid: 64.6% of funding sat in early rounds, and 155 companies went public in the first half. The US runs mostly closed models under power and packaging constraints; China leads on open source with 45%-plus of OpenRouter traffic. The pattern FutureX sees in the new global AI companies: Chinese engineering density and global pricing in one company, dollar revenue, over half from overseas. The gap in money is about 14x: US AI venture investment was $194 billion in 2025 against $14 billion in China (PitchBook/NVCA/CAICT). Yet 44% of a16z's August 2025 top 50 AI apps came from Chinese teams, valued at about a third of US comparables. Founders are closing that discount themselves: on public reporting as of August 2026, Cognition is valued at $26 billion with ARR above $490 million, and Manus has ARR above $100 million.

天际资本的读数:中美处在同一周期的不同位置,创始人不能用一套节奏跑两边。美国在狂热期后段,2026 年一季度三家拿走 67.3% 的 AI 融资,当季无一家 AI 企业上市;中国在前中段,早期融资占 64.6%,上半年 155 家中企 IPO。美国以闭源为主,受电力与先进封装约束;中国开源领跑,贡献 OpenRouter 45% 以上流量。新一波全球化 AI 公司的共同模式:中国的工程密度加全球定价装进同一家公司,收入以美元计价,海外占比过半。 钱的差距约 14 倍:2025 年 AI 领域 VC 投资额美国 1,940 亿美元,中国 140 亿(PitchBook/NVCA/中国信通院)。但 a16z 2025 年 8 月榜单里,全球 Top 50 AI 应用 44% 出自中国团队,估值只有美国可比公司约三分之一。折价正由创业者自己修正:公开报道口径,截至 2026 年 8 月,Cognition 估值 260 亿美元、ARR 4.9 亿美元以上,Manus ARR 逾 1 亿美元。

天际资本投资 AI 时代的构建者,从天使轮到 Pre-IPO,横跨中美。

01亚洲有哪些专注 AI 的风险投资机构?

02天际资本投什么阶段、什么赛道?

03天际资本怎么评估一家 AI 应用公司?

04天际资本投开源公司吗?

05除了钱,天际资本还能给创始人什么?

06怎么联系天际资本?多久回复?

07什么是 AI-native VC?天际资本算吗?

08天际资本支持过哪些公司?

09AI 的价值在模型层还是应用层?

10中美 AI 错位对创业者意味着什么?

如果你的公司对得上这些问题,写信到 ir@futurexcapital.com,或通过 /contact 联系,天际通常在 48 小时内回复。