本周简而言之主要项目Anthropic的IPO即将考验人工智能热潮在过去的十年里，人们感觉从未有过上市的完美时机--环境几乎总是不稳定，私募市场也相对舒适。
The Week in Short The Main Item Anthropic's IPO Is About to Test the AI Boom Over the past decade, it’s never felt like there’s been a perfect time to go public — the conditions are almost always rocky and the private markets have been so comparatively cozy.
但当你需要像前沿实验室一样多的资金时，在某个时候你必须堵住鼻子并跳下去。
But when you need as much capital as the frontier labs do, at some point you have to plug your nose and jump.
上周，Anthropic和OpenAI的泄密、独家新闻和公告层出不穷，他们试图弄清楚如何获得下一批急需的资金。
The last week has been a whirlwind of leaks, scoops, and announcements from Anthropic and OpenAI as they try to figure out how to land their next tranche of desperately needed capital.
周一，该公司推出了一份令人印象深刻的独家新闻，获得了似乎是Anthropic招股说明书草案的内容。其中的数字显示了其惊人的支出速度。去年，Anthropic报告营业亏损约80亿美元，收入为46亿美元。它还承诺在未来几年内在云、计算和基础设施上花费5180亿美元。
On Monday pulled off a downright impressive scoop, getting its hands on what appears to be a draft of Anthropic’s prospectus. The numbers in it showed its staggering rate of spending. Last year Anthropic reported an operating loss of around $8 billion on revenue of $4.6 billion. It also was committed to spending $518 billion on cloud, computing, and infrastructure in the coming years.
需要注意的是，这些数字都是在Anthropic的Opus 4.6模型于2026年2月推出之前的;这是一个怪物模型，开启了该公司的代理编码主导地位以及随后的“tokenmaxing”热潮，从而推动了其收入。在没有Opus 4.6的情况下判断人择，就像之前评估Marvel一样。
The caveat here is these numbers are all before Anthropic’s Opus 4.6 model came out in February 2026; that was the monster model that kicked off the company’s agentic coding dominance and the subsequent “tokenmaxxing” craze that rocket-powered its revenue. Judging Anthropic without Opus 4.6 would be like evaluating Marvel before .
尽管如此，对于Anthropic的收入增长率以及该公司是否能够在IPO中达到2万亿美元的估值，人们一直存在疑问。它已经推迟了预期的10月首次亮相，现在希望在感恩节前的11月开始交易。
Still, there have been swirling questions about Anthropic’s revenue growth rate and whether the company would be able to hit a $2 trillion valuation in an IPO. It has already postponed an expected October debut and now is looking to start trading in November before Thanksgiving.
一般来说，公共市场似乎不太欢迎。软银子公司SB Energy在9月份推迟了其IPO。本周，Oura以市场状况为由无限期推迟了IPO。
The public markets in general don’t seem too welcoming. SB Energy, the SoftBank subsidiary that builds AI data centers, delayed its IPO in September. And this week Oura postponed its IPO indefinitely, citing market conditions.
与此同时，OpenAI正在努力尽可能长时间地等待时机。该公司正在就新一轮私募融资进行早期谈判，目标是至少300亿美元，估值为1.4万亿美元，预计该公司将持续到2027年IPO。萨姆·奥尔特曼（Sam Altman）一直在利用延迟来应对对车型安全性的担忧，而不是基础业务的任何问题--为此，多家媒体本周报道称，该公司的年化收入已接近700亿美元。
OpenAI, meanwhile, is trying to bide its time as long as possible. The company is in early talks to raise a new private round, targeting at least $30 billion at a $1.4 trillion valuation, which is supposed to last the company until a 2027 IPO. Sam Altman has been spinning the delay as a reaction to concerns over model safety rather than anything wrong with the underlying business — to that end, several outlets reported this week that the company’s annualized revenue had neared $70 billion.
接下来的几周对于评估人工智能业务（以及总体经济）的健康状况将非常有启发性。如果最新的Anthropic财务报告在完整发布时显示一家公司的收入增长仍然蓬勃发展，那么这个周期还有更多的跑道，我们可能会看到对该行业其他部分的下游影响。如果我们开始看到经济放缓，那么在可预见的未来，事情将变得艰难。AI Mega交易AMD是黑马。是在盈利吗？
The next few weeks will be very instructive in assessing the health of the AI business (and the economy writ large). If the latest Anthropic financials, when they get released in full, show a company with still-booming revenue growth, then there’s much more runway to this cycle and we’re likely to see a downstream effect on the rest of the industry. If we start to see a slowdown, then things will be rough for the foreseeable future. AI Mega Deals AMD Is a Dark Horse. Is it Gaining?
芯片制造商AMD本周宣布，将以82亿美元的全股票交易收购计算机视觉先驱李飞飞博士的人工智能初创公司World Labs。这是AMD长期担任首席执行官的Lisa Su博士的又一明智之举，因为该公司寻求赢得超大规模的市场份额并扩展到新的垂直领域。
Chipmaker AMD announced this week that it would be acquiring computer vision pioneer Dr. Fei-Fei Li’s AI startup World Labs for $8.2 billion in an all-stock deal. It is yet another shrewd move by AMD’s longtime CEO Dr. Lisa Su as the firm seeks to both win hyperscaler market share and expand into new verticals.
世界实验室恰如其分地创建了世界模型--模拟的多模式和交互式环境，以一致的方式生成自己。这个领域的其他大公司包括Odyssey和前Meta首席科学家Yann LeCun的AMI实验室。AMD表示，收购该公司是为了帮助其更好地了解随着“对计算基础设施的需求变得更加多样化”，人工智能工作负载如何演变。
World Labs, fittingly, creates world models — simulated multi-modal and interactive environments that generate themselves in a consistent manner. The other big players in this space include Odyssey and former Meta chief scientist Yann LeCun’s AMI Labs. AMD says it is buying the firm to help it better understand how AI workloads are evolving as “demands on compute infrastructure become more diverse.”
在Substack讨论此次收购的帖子中，李指出苏是该公司的“好朋友”和早期投资者。她还指出，两家公司早在2025年就曾联手“对AMD图形处理器进行模型训练和推理优化”。在这样一个男性主导的行业中，AMD现在有两名女性担任其最高级技术职位，这既令人震惊又令人兴奋，因为该公司正在继续重塑其作为英伟达有力竞争对手的说法。
In a Substack post discussing the acquisition, Li noted that Su was a “great friend” and an early investor in the firm. She also noted that the two companies had teamed up back in 2025 to conduct “model training and inference optimization on AMD GPUs.”In such a male-dominated industry, it is both striking and exciting that AMD now has two women in its most senior technical roles as the firm continues to rebuild its claim to being a serious NVIDIA competitor.
收购世界实验室是本周的新闻，但随之而来的背景故事是，AMD首款机架级人工智能解决方案Helios将于今年年底开始首次发货。投资者很高兴看到这些系统在全球上市; HPE股价本周大幅上涨，此前该公司宣布以12亿美元的价格向云基础设施公司Vultr提供Helios机架。
The World Labs acquisition is the news of the week, but the background story which bubbles along is that initial shipments of AMD’s first rack-scale AI solution, Helios, will begin by the year’s end. Investors are excited to see these systems out in the world; HPE stock rose sharply this week after it announced a $1.2 billion deal to provide Helios racks to cloud infrastructure company Vultr.
AMD的企业级图形处理器产品线（Instinct）也拥有极其健康的需求，仅OpenAI、Meta和Anthropic对这些芯片的需求就达到了14吉瓦。作为参考，麦肯锡预计到2026年全球人工智能工作负载总容量将达到62千兆瓦。Anthropic的两千兆瓦合作伙伴关系也将采取Helios机架的形式。
AMD’s enterprise-grade GPU product line, called Instinct, also boasts exceedingly healthy demand, with 14 gigawatts of demand for these chips from OpenAI, Meta and Anthropic alone. For reference, McKinsey projects that there is 62 gigawatts of total global AI workload capacity in 2026. Anthropic’s two gigawatt partnership will also take the form of Helios racks.
这并不是夸大AMD的地位;它远远落后于英伟达，市值仅为1万亿美元，具有更高的P/E倍数，并且缺乏CUDA生态系统护城河。尽管如此，AMD已经连续四个季度轻松超出了分析师的预期，苏博士可以获得当之无愧的胜利。
This is not to overstate AMD’s position; it lags well behind NVIDIA at a mere $1 trillion market cap, has a much higher P/E multiple and lacks the CUDA ecosystem moat. Nonetheless, AMD has handily beat analyst expectations four quarters in a row, and Dr. Su can take a deserved victory lap.
新人播客Reid Hoffman谈ChatGPT、数据中心和为什么美国需要人工智能一大排行榜最大的风险投资公司支持最多的独角兽如果您有足够大的战争基金，就更容易为最好的交易付费。
Newcomer Podcast Reid Hoffman on ChatGPT, Data Centers & Why America Needs AI One Big Chart The Biggest VC Firms Backed the Most Unicorns If you have a big enough war chest, it’s easier to pay up for the best deals.
根据Odin的丹·格雷（Dan Gray）汇编的数据，2010年至2020年获得最多资本的大型基金也倾向于支持最多的未来独角兽公司。
Megafunds that secured the most capital from 2010 to 2020 also tended to back the most future unicorns, according to data compiled by Dan Gray over at Odin.
Accel是顶级企业中的一个异类--它筹集的资金不到100亿美元，但投资的独角兽数量与a16 z和红杉几乎相同，这两家公司在同一时间段内的收入都远超过100亿美元。
Accel is the one outlier at the top — it raised less than $10 billion but invested in close to the same number of unicorns as a16z and Sequoia, which both brought in well over $10 billion during the same time frame.
五笔著名交易Listen Labs、EliseAI、Armadin、Varda Space、Hertha Metals
Five Notable Deals Listen Labs, EliseAI, Armadin, Varda Space, Hertha Metals