Chi-Hua Chien saw Facebook coming — now he says the real AI winners won’t be selling AI
Chi-Hua Chien saw Facebook coming — now he says the real AI winners won’t be selling AI
Chi-Hua Chien saw Facebook coming — now he says the real AI winners won’t be selling AI 简志华（Chi-Hua Chien）曾预见 Facebook 的崛起——如今他表示，AI 时代的真正赢家将不会是那些兜售 AI 的公司。
Chi-Hua Chien has spent more than two decades as a venture capitalist, but he thinks like a cultural anthropologist. As a co-founder of Goodwater Capital, a firm focused exclusively on consumer and prosumer technology, he has built a portfolio spanning entertainment, healthcare, fintech, and live experiences — with investments in companies like MIDI Health, Fever, and Monzo. 简志华从事风险投资已超过二十年，但他思考问题的方式却像一位文化人类学家。作为 Goodwater Capital 的联合创始人，他专注于消费级和专业级技术领域，其投资组合涵盖了娱乐、医疗、金融科技和现场体验等行业，投资对象包括 MIDI Health、Fever 和 Monzo 等公司。
He was also, as a 27-year-old associate at Accel, the person who initially found a six-person company launched from Harvard called The Facebook. That ability to read human behavior at scale informs everything from his view that Americans will never trust a single app with both their social lives and their finances, to his belief that the gap between the most advanced AI model and what you can run on your phone — once as wide as two years — will shrink to three months within the next year. 27 岁时，身为 Accel 投资经理的他，曾发掘了当时仅有六人、从哈佛起步的 Facebook。这种大规模洞察人类行为的能力，贯穿了他的所有观点：从他认为美国人永远不会将社交生活和财务状况同时托付给同一个 App，到他坚信最先进的 AI 模型与手机本地运行模型之间的差距（曾长达两年）将在明年缩短至三个月。
These days, he is also willing to say out loud what many in venture capital are only thinking: that the commoditization of the model layer is already underway and that the biggest winners of the AI era won’t be the companies selling AI at all. We talked last week. This interview has been edited for length and clarity. 如今，他愿意公开说出许多风投圈人士只敢在心里想的事情：模型层的商品化已经开始，而 AI 时代的真正赢家根本不会是那些兜售 AI 的公司。我们上周进行了交谈。为简洁明了，本文对采访内容进行了编辑。
More founders and investors have been publicly sharing their grievances about VCs lately. What’s changed? 最近，越来越多的创始人和投资者公开表达对风投机构的不满。发生了什么变化？
It’s part of the meme-ification of everything — you’re seeing what’s happening in the political realm bleeding over into the business side, and it’s probably also the sign of some peakiness in the market. The reason you’re seeing some of these outspoken investors talking more publicly is because venture firms have largely vertically integrated, so the really big ones have enough capital that they’re not necessarily looking for syndicate partners. There used to be decorum around wanting to preserve good relationships with other co-investors, because you got to work with them at different points along the line. As the firms have gotten bigger and vertically integrated, there’s less of that need. 这是万物“模因化”（meme-ification）的一部分——你看到政治领域的现象正蔓延到商业领域，这也可能是市场过热的迹象。之所以看到一些直言不讳的投资者更公开地发表言论，是因为风投机构在很大程度上实现了垂直整合，那些真正的大型机构拥有充足的资本，不一定需要寻找联合投资伙伴。过去，为了维护与其他共同投资者的良好关系，大家会保持某种礼节，因为你未来可能在不同阶段与他们合作。随着机构规模扩大并实现垂直整合，这种需求减少了。
What about the “fast follow” rounds — where firms invest a large chunk at one valuation and a smaller amount weeks later at a much higher one, making the headline number look more impressive than it really is? Is this really new? How pervasive is it? 关于“快速跟投”轮次——即机构先以一个估值投入大笔资金，几周后再以高得多的估值投入少量资金，使账面数字看起来比实际更亮眼——这真的是新现象吗？它有多普遍？
I think it’s been going on for quite some time. The best companies raise successive rounds very quickly — there might only be three to six months between rounds now, and valuations change really quickly … Valuations are being marketed very aggressively as a way of demonstrating market leadership, attracting talent, potentially blocking out competition. There’s probably some element of frothiness, because what these fast financings are most illustrative of is there’s way more demand than there is supply. An investor can come in, set a price, complete a financing, and then a couple of weeks later there’s still excess demand — and the company can immediately price a new round at a higher price. 我认为这种情况已经存在很长一段时间了。最优秀的公司融资节奏非常快——现在轮次之间可能只有三到六个月，估值变化也非常迅速……估值被极其激进地营销，作为展示市场领导地位、吸引人才以及潜在地阻击竞争对手的手段。这其中可能存在泡沫成分，因为这些快速融资最能说明的是：需求远大于供给。投资者可以入场、定价、完成融资，几周后如果需求依然过剩，公司就可以立即以更高的价格进行新一轮融资。
You’ve argued that infrastructure companies get commoditized and that applications capture most of the value over time. Are we already seeing that play out in this cycle? 你曾指出基础设施公司会走向商品化，而应用层最终会捕获大部分价值。我们在这一周期中已经看到这种情况了吗？
If you look at the PC cycle, the web cycle, and the mobile cycle, they all follow fairly consistent patterns. Infrastructure market caps actually peaked in the year 2000 — but you fast-forward 25, 26 years later, and in nominal dollar terms, the market cap of those infrastructure companies has not surpassed the 2000 peak. In the web era, infrastructure new entrants produced $400 billion of new market cap. Application companies created $3.1 trillion — 88% of the new value. In the mobile era, it’s very similar: Infrastructure produced about $700 billion, while application companies produced $3.7 trillion. Companies like Netflix, Spotify, Meta, Uber, Airbnb. 如果你观察 PC 周期、Web 周期和移动互联网周期，它们都遵循相当一致的模式。基础设施公司的市值实际上在 2000 年达到顶峰——但快进 25、26 年后，以名义美元计算，这些基础设施公司的市值仍未超过 2000 年的峰值。在 Web 时代，基础设施领域的新进入者创造了 4000 亿美元的新市值，而应用公司创造了 3.1 万亿美元——占新价值的 88%。在移动时代，情况非常相似：基础设施创造了约 7000 亿美元，而应用公司创造了 3.7 万亿美元。比如 Netflix、Spotify、Meta、Uber 和 Airbnb 等公司。
And [last week] you saw something pretty interesting: Google announced that their subscription AI product is dropping price from $7.99 a month to $4.99 a month and doubling the storage. We’re already in the era of price competition — and companies like Google, with structural advantages in vertical integration and distribution, can start bundling and price competing for the average consumer. 而且（上周）你看到了一个非常有趣的现象：谷歌宣布其 AI 订阅产品价格从每月 7.99 美元降至 4.99 美元，并将存储空间翻倍。我们已经进入了价格竞争时代——像谷歌这样在垂直整合和分发渠道上拥有结构性优势的公司，已经可以开始通过捆绑销售和价格战来争夺普通消费者。
You keep coming back to personalization as a through line. Is that what separates the next wave of winners? 你一直强调“个性化”是核心主线。这就是下一波赢家的分水岭吗？
Hyper-personalization definitely is a key through line, because what does personalization give you? If done right, it gives you higher customer satisfaction, deeper engagement, and higher ARPUs over time. We have entertainment companies in our portfolio — companies like Triumph and Ritten and Flow GPT — where the customer is not saying, “This is an AI application.” They’re saying it’s an entertainment application. These companies are going into 100 million, 400 million, 600 million of ARR very quickly, at great margins, because AI makes the experience more customizable and more personalized — but it’s not the fundamental capability they’re selling. 超个性化绝对是一条关键主线，因为个性化能带来什么？如果做得好，它能带来更高的客户满意度、更深度的参与感，以及随时间增长的更高每用户平均收入（ARPU）。我们的投资组合中有一些娱乐公司——比如 Triumph、Ritten 和 Flow GPT——客户不会说“这是一个 AI 应用”，他们会说这是一个娱乐应用。这些公司正迅速实现 1 亿、4 亿、6 亿美元的年度经常性收入（ARR），且利润率极高，因为 AI 让体验变得更具定制化和个性化——但这并不是它们兜售的核心能力。
We also have a women’s health company called Midi Health. One of the fundamental constraints in women’s health is that there aren’t that many providers well trained in hormone replacement therapy for perimenopausal women. By using AI, they’re able to substantially expand the supply of care and treat hundreds of thousands of patients that otherwise couldn’t be reached. And they can do it cost effectively, which expands access to a market that was previously supply constrained. You can play that forward across every supply-constrained category where human expertise is the bottleneck. 我们还有一家名为 Midi Health 的女性健康公司。女性健康领域的一个根本制约因素是，受过良好培训、能为围绝经期女性提供激素替代疗法的医疗服务提供者并不多。通过使用 AI，他们能够大幅扩大医疗供给，治疗数十万原本无法触达的患者。而且他们能以高性价比的方式实现这一点，从而扩大了此前受供给限制的市场准入。你可以将这种模式推广到每一个以人类专业知识为瓶颈的供给受限领域。
How far away are we from AI that feels truly personal and ambient? 我们距离那种真正个性化且无处不在的 AI 还有多远？
I don’t think we’re very far away at all. You can run locally now on your phone AI models that are as good as the best models were about six months ago — and that lag is shrinking. You go back two years ago, the lag between what you could run locally and what was in the cloud with the frontier models might have been 18 to 24 months. It’s now six months. It’s probably getting down to three months. 我认为距离并不遥远。现在你可以在手机上本地运行 AI 模型，其性能相当于六个月前最顶尖的模型——而且这种滞后正在缩小。回到两年前，本地运行模型与云端前沿模型之间的差距可能长达 18 到 24 个月。现在是六个月，而且很可能很快会缩短到三个月。