首页  >>  来自播客: a16z 更新   反馈  

a16z - AI, Infrastructure, and the Next Investment Cycle

发布时间:   原节目
这期于2026年9月30日发布的播客深入探讨了科技市场的当前状况、大规模人工智能建设、其早期采用情况以及对各行业的未来影响。由David George主持,同事Sarah Wang、Alex Emmerman和Santiago Rodriguez参与,讨论围绕A16Z最新市场现状演示中的25张关键幻灯片展开。 David George强调,全球市值最高的10家公司中,有8家是美国科技公司。自近四年前JotGPT发布以来,市场飙升了90%,年化增长率为17%。尽管处于这个热门时期,市场并未出现泡沫,因为业绩是由基本盈利驱动的,而非虚高的市盈率。标普500指数的市盈率仍低于20倍,与互联网泡沫时代形成鲜明对比。科技现在占美国资本支出的55%,占总股票市值的40%,甚至超过了铁路建设占GDP的比例。 这种被称作“万物周期”的大规模投资,远远超出了人工智能和数据中心。到2040年,全球基础设施需求预计将达到90万亿美元,包括电力、水、道路和交通。Alphabet、亚马逊、Meta、微软和甲骨文等超大规模企业正在将短期运营现金流投入到产能建设中,预计2026年资本支出将达到7800亿美元,并从2027年起每年超过1万亿美元。这种需求持续超过供给,受到人工智能及其通过现有互联网、云和移动基础设施即时触达数十亿用户的推动。Alex Emmerman指出,“计算需求是一个模式破坏者”,像OpenAI这样的公司表现出永不满足的需求,甚至暂停了新的专业版订阅。 尽管由于这种建设,超大规模企业的自由现金流目前受到抑制,但预测从2028年起将强劲复苏。值得注意的是,数据中心,尽管经常受到批评,实际上可以降低电价;一项美国研究显示,数据中心容量增加10%与居民电价下降40个基点相关,因为它们稳定并分摊了固定的电网成本。 人工智能已产生可观的收入和节省,但其采用仍处于“极其早期”阶段。OpenAI和Anthropic的合计年化收入已大幅超越了历史软件巨头的增长率。尽管标普500公司中有69%已部署了实时AI,但只有30%报告了可量化的影响,且只有2%随着时间推移跟踪AI指标。这表明有很大的深入整合空间。Sarah Wang强调,模型能力与实际应用之间的这种差距为应用层公司带来了巨大机遇。 由于受“杰文斯悖论”启发的技术等创新,AI模型的成本正在暴跌,使代理在更广泛的任务中变得经济划算。一些工作负载成本降低了10倍,而微调模型可以降低60%的成本并具有更低的延迟。坊间传闻,投资组合中顶级的AI用户支出是中位数用户的20倍,一些前瞻性公司将多达10%的人员预算分配给AI工具。可量化的案例研究包括Chime连续四年每年将服务成本降低超过10%,以及Shopify通过其AI助手将客户保留率提高8%。甚至像ServiceNow这样的老牌企业也在蓬勃发展,报告的AI年度合同价值(ACV)超过10亿美元。 在消费领域,采用也仍处于初期阶段,只有2%的美国家庭拥有付费AI订阅。然而,这些订阅表现出卓越的保留率,随着产品价值的增加,它们通常呈现“微笑曲线”。讨论承认AI代理有潜力重塑消费平台,对传统的广告密集型模式构成挑战,但可能创造更多的整体消费和商品交易总额(GMV)。 公共软件市场已转向增长较慢但利润更高的公司,只有30%的上市软件公司增长率达到20%或更高。这要求现有SaaS公司利用其强大的分销渠道来整合AI并推动收入增长加速。网络安全、可观测性和垂直软件领域表现最佳,反映了由AI驱动的安全需求和专业化应用带来的需求增长。 私募市场正迎来前所未有的规模,六家顶级公司(Anthropic、OpenAI、Databricks、Stripe、Waymo、Revolut)的总估值达到2.4万亿美元——超过过去十年(不包括SpaceX)所有首次公开募股(IPO)的总市值。这使得公司能够在产品开发中进行更长期的“大胆尝试”。员工信心很高,只有58%的人参与了要约收购,表明他们对公司的未来充满信心。风险投资交易活动也反映了这一点,AI相关公司占2026年美国风险投资交易的86%。 播客最后展望了未来的机遇,充满兴奋:长期运行的消费者代理、机器人技术(潜力可能超过大型语言模型)、自动化(自动驾驶安全性提高14倍)、生物学中的AI(药物发现、个人健康)、企业AI在编码之外更深层次的普及,以及“美国活力”——用新的科技供应商重塑工业和国防部门。发言者对AI对经济的变革性、生产力提升影响持乐观态度。

The podcast, published on September 30, 2026, delves into the current state of the tech market, the massive AI buildout, its early adoption, and future implications for various sectors. Hosted by David George with colleagues Sarah Wang, Alex Emmerman, and Santiago Rodriguez, the discussion centers on 25 key slides from A16Z's latest state of markets presentation. David George highlights that eight of the top 10 most valued companies globally are U.S. tech firms. Since JotGPT's release nearly four years prior, the market has soared by 90%, an annualized 17%. Despite this hot period, the market isn't in a bubble, as performance is driven by fundamental earnings, not inflated multiples. The S&P 500 earnings multiple remains below 20 times, contrasting sharply with the dot-com era. Tech now accounts for 55% of U.S. capital spending and 40% of the aggregate stock market value, exceeding even the buildout of railroads as a percentage of GDP. This massive investment, termed the "everything cycle," extends far beyond AI and data centers. Global infrastructure needs are estimated at $90 trillion through 2040, encompassing power, water, roads, and transit. Hyperscalers like Alphabet, Amazon, Meta, Microsoft, and Oracle are pouring near-term operating cash flow into capacity, with CapEx projected to hit $780 billion in 2026 and exceed $1 trillion annually from 2027. This demand consistently outstrips supply, driven by AI and its immediate access to billions of users through existing internet, cloud, and mobile infrastructure. Alex Emmerman notes that the "demand for compute is a model buster," with companies like OpenAI demonstrating insatiable demand, even pausing new pro subscriptions. While free cash flow for hyperscalers is currently depressed due to this buildout, forecasts predict a strong recovery from 2028. Notably, data centers, often criticized, can actually lower electricity rates; a U.S. study showed a 10% increase in data center capacity correlated with a 40 basis point drop in residential rates, as they stabilize and spread fixed grid costs. AI is already generating major revenue and savings, yet adoption remains "extremely early." OpenAI and Anthropic's combined annualized revenue has dramatically surpassed the growth rates of historical software giants. Despite 69% of S&P 500 companies having live AI deployments, only 30% report quantifiable impact, and only 2% track AI's metric over time. This indicates significant room for deeper integration. Sarah Wang emphasizes that this gap between model capabilities and actual use presents a huge opportunity for application-layer companies. Costs for AI models are plummeting due to innovations like Jevons paradox-inspired techniques, making agents economical for a wider range of tasks. Some workloads have become 10x cheaper, while fine-tuned models can be 60% cheaper with lower latency. Anecdotally, top AI users in portfolios spend 20 times more than median users, and some forward-leaning companies allocate up to 10% of their headcount budget to AI tools. Quantifiable case studies include Chime reducing cost to serve by over 10% annually for four years, and Shopify boosting customer retention by 8% through its AI sidekick. Even incumbents like ServiceNow are thriving, reporting over $1 billion in AI ACV. In the consumer space, adoption is also nascent, with only 2% of U.S. households having a paying AI subscription. However, these subscriptions show exceptional retention, often "smiling" as product value increases. The discussion acknowledges the potential for AI agents to reshape consumer platforms, posing challenges to traditional advertising-heavy models but potentially creating more overall consumption and Gross Merchandise Volume (GMV). The public software market has seen a shift towards slower-growing, more profitable companies, with only 30% of public software firms growing at 20% or more. This necessitates existing SaaS companies to leverage their strong distribution to integrate AI and drive revenue growth acceleration. Cybersecurity, observability, and vertical software segments have performed best, reflecting increased demand due to AI-driven security needs and specialized applications. Private markets are seeing unprecedented scale, with six top companies (Anthropic, OpenAI, Databricks, Stripe, Waymo, Revolut) collectively valued at $2.4 trillion—more than the combined market cap of IPOs in the last decade (excluding SpaceX). This allows companies to pursue longer-term, "bigger swings" in product development. Employee conviction is high, with only 58% participation in tender offers, indicating strong belief in their companies' future. VC deal activity reflects this, with AI-related companies accounting for 86% of U.S. VC deals in 2026. The podcast concludes with excitement for future opportunities: long-running consumer agents, robotics (potentially larger than LLMs), autonomy (self-driving becoming 14x safer), AI in biology (drug discovery, personal health), deeper enterprise AI diffusion beyond coding, and "American dynamism"—a retooling of industrial and defense sectors with new tech vendors. The speakers are optimistic about AI's transformative, productivity-enhancing impact on the economy.