Anthropic's newly filed S-1 reveals a 4.3x improvement in revenue generated per dollar of compute and infrastructure expense since Q1 2025. The ratio rose from $0.40 in Q1 2025 to $1.72 in Q3 2026, hitting its highest point at $2.08 in Q2 2026. For AI companies, compute efficiency is becoming just as important as revenue growth. And Anthropic's latest numbers show meaningful progress on both... $AGIX was the first ETF to hold Anthropic through a direct private-company investment. With the company approaching its anticipated IPO, AGIX offers ETF exposure before its public-market debut. Anthropic is 2.44% of AGIX's portfolio as of date 10/30/2026. Holdings are subject to change. To learn more about how to get exposure to Anthropic and AGIX top 10 holdings visit:

3/3 Anthropic’s operating margin turned positive in Q2 2026. It improved from -249% in Q1 2025 to -35% in Q1 2026, before reaching +8% in Q2 2026. Operating margin measures how much revenue remains after operating expenses, before interest and taxes. Moving into positive territory marks a significant change in Anthropic’s reported operating performance. As Anthropic approaches its debut as the first publicly traded frontier AI model company, investors are without a playbook. Headline revenue alone won’t tell the full story. Customer diversification, compute efficiency, and expanding operating margins are the metrics that reveal some of Anthropic’s true strengths. Learn more at

2/3 Anthropic is generating more revenue for every dollar spent on compute and infrastructure. In Q1 2025, every $1 spent generated $0.40 in revenue. By Q4 2025, that reached $0.69. In Q2 2026, it hit $2.08. That’s more than 5x in five quarters. Compute is one of the major costs of running an AI business. Generating more revenue per dollar spent suggests Anthropic is getting more productive use out of that infrastructure. And that efficiency may help explain another major shift in the filing: its operating margin.

1/3 Anthropic’s S1 IPO filing reveals more than just revenue growth. Here are 3 financial updates that stood out: First, its revenue is becoming less concentrated. Its top five customers accounted for 36% of quarterly revenue in Q1 2026. By Q3, that share had fallen to 21%, with all other customers accounting for 79%. Why it matters: a broader customer base can make revenue less vulnerable to spending changes by a handful of large customers. But diversifying revenue is only one part of the equation. The next number reveals how much more revenue Anthropic is generating from its compute spending:

The humanoid robotics conversation has largely centered on which robot maker will win. But the companies supplying the motion systems underneath may matter just as much, regardless of which platform comes out on top. Actuators, the motors, reducers, sensors, and controllers that turn AI commands into physical movement, are the single largest cost driver in humanoid hardware, and every robot needs dozens of them. Could the actuator supply chain be one of the most overlooked opportunities in physical AI? Join KraneShares CIO @ahern_brendan, Senior Investment Strategist Derek Yan, CFA, and Max Chen of Worldline Capital for a webinar on the actuator bottleneck and how $TORK and $KOID seek to capture the physical AI opportunity at different levels of the stack. Register: TORK: KOID:

A robot's intelligence is only useful if its body can execute the task. Recognizing an object and deciding how to pick it up are software challenges. Actually reaching for it, gripping it, and moving it without damage requires actuators, the muscles and joints that translate AI commands into physical movement. Actuation and mechanical components represent an estimated 55% of humanoid robot hardware costs, according to TrendForce, and with humanoids requiring dozens of actuators per unit, the motion system supply chain could become one of the most significant bottlenecks in physical AI. $TORK is designed to provide targeted exposure to the companies building those motion systems, including manufacturers of actuators, precision reducers, motors, controllers, and sensors, across humanoid robotics and the established industrial markets they already serve. For investors who already hold $KOID, TORK offers a more concentrated allocation to what we believe is one of the most critical layers of the physical AI stack. Read our latest article: TORK: KOID:

Three potential AI spin-offs could put China’s tech giants in focus... @AlibabaGroup T-Head AI chip unit, @Baidu_Inc Kunlun AI chip business and Kuaishou’s Kling AI video unit are three potential catalysts to watch. Kling is reportedly targeting a Hong Kong listing in Q1 2027, while November earnings could bring more clarity on the broader pipeline. Separate listings could also give investors a clearer picture of what these AI businesses are worth, potentially unlocking value that is harder to see within their much larger parent companies. Three to watch for $KWEB See KWEB’s top 10 holdings and learn more about the fund here:

One of China’s most closely watched AI companies is preparing for a potential IPO on Shanghai’s STAR Market... DeepSeek’s IPO could mark the next phase of China’s AI boom, as it prepares to enter a market already home to leading semiconductor and robotics companies like CXMT and Unitree. At a potential $75 billion valuation, DeepSeek could become the fifth-largest company listed on the STAR Market, based on September 28, 2026 market prices. $KSTR tracks the STAR 50 Index, giving investors exposure to companies listed on the STAR Market. Due to its size, DeepSeek could be eligible for the STAR Market’s fast-track inclusion process, potentially giving it a quicker path into major indexes. Learn more about DeepSeek’s potential IPO and what DeepSeek’s potential IPO could mean for KSTR here: A Complete Guide to DeepSeek’s 2026 IPO: What It Means For China ETF KSTR - KraneShares $KSTR does not hold DeepSeek as of 10/7/26. An IPO and future inclusion in KSTR are not assured. Holdings are subject to change. Top 10 holdings:

China's stock market has seen a deep divergence in 2026. Mainland hardware technology stocks have delivered solid performance, while internet and consumer platforms have sold off despite their strong relevance to the AI story. Meanwhile, China's IPO pipeline is filling up fast. After CXMT and @UnitreeRobotics made their debuts, Yangtze Memory (YMTC), @deepseek_ai, @Kimi_Moonshot, and others are still to come. Could a peaking capex cycle drive a profitability rebound for internet giants like $BABA, and could that spark a re-rating? Join KraneShares CIO @ahern_brendan and Head of International @Dr_XiaolinChen for a Q3 update on China's market, the upcoming IPO pipeline, and how $KWEB and $KSTR seek to capture both sides of the opportunity. Register:

"Actuators will be the GPUs of tomorrow. GPUs are in short supply right now, but actuators are the future," Derek Yan told @SergeBerger on the Steady Wealth Podcast from KraneShares' Switzerland event, From Digital to Physical AI: Humanoids, China AI, & Emerging Technologies. Derek discussed the shift from digital to physical AI alongside a live robot demonstration and made the case that we may be approaching a ChatGPT-style inflection point for humanoid robotics and highlighted $KOID as a way to capture the full physical AI ecosystem, from the brain and integrators down to the critical components like actuators and harmonic drives that represent today's biggest production bottleneck. He also pointed to $AGIX as the vehicle bringing private AI leaders like @Anthropic into an ETF wrapper. Thank you, Serge, for having us on! Watch: KOID: AGIX:

Korea: If Taiwan is critical for making advanced chips, Korea is critical for the memory those chips need. Samsung and SK Hynix are two of the world’s leading producers of high-bandwidth memory (HBM). HBM helps AI chips process large amounts of data quickly. And as AI systems get bigger, they need more and more of it. What’s also changing is how that memory is being sold. Major customers are signing multi-year contracts to secure HBM supply, creating a growing backlog of demand. That’s very different from the traditional memory market, where demand and prices could change sharply from one cycle to the next. 3/6

1/6: 75% of semiconductor production capacity and material supply is concentrated in Asia. And when you look at where the biggest AI infrastructure bottlenecks are, four countries stand out: Taiwan. Korea. China. Japan. Each plays a different role in building the infrastructure AI needs to scale. Let’s break down how KAIT gets exposure to all four markets and the role each plays in the AI buildout:

$0.18 vs. around $4 per million tokens... That’s the price gap between some China-based LLMs and leading U.S. models. For coding assistants, customer service, enterprise search and internal automation, the most expensive model may not be necessary for every workload. It raises a simple question: do businesses need a Ferrari for every trip to the grocery store? China’s LLM providers are offering open-weight models that can be customized, fine-tuned internally and run on a company’s own infrastructure.

We caught @ahern_brendan before he headed to @CNBC to get a quick preview of what he’ll be discussing ahead of this week’s Trump-Xi meeting.   Here’s what’s on our CIO’s radar 👇

China's AI models may be becoming the new global stars. Certain US and global technology firms have reported using lower-cost, China-developed large language models (LLMs) for certain applications. We believe this development could significantly benefit China's internet sector by providing AI leaders, including $KWEB holdings $BABA, @Zai_org, @TencentGlobal, and @MiniMax_AI, with a stable and diversified revenue stream. KWEB has also expanded its index methodology to include LLM providers, data center operators, and enterprise cloud firms. We believe these additions reflect the evolution of web-based businesses globally in recent years and mean that KWEB may fully represent AI-enabled internet development in China. Our Summer 2026 China Internet Quarterly Earnings Report breaks down what this means for investors, plus a full subsector analysis and earnings update across KWEB's top 10 holdings. Read the full report: KWEB Top 10 Holdings:

The humanoid robots at the 2026 World Robot Conference in Beijing may be the most eye-catching part of the story. But behind them is a much larger ecosystem, connecting components, manufacturers, applications, and infrastructure across logistics, retail, and disaster response. China isn't just building robots. It's building the infrastructure to make them work.

$31.6 TRILLION. That’s how much Oxford Economics thinks the world could spend on AI datacenters through 2050. If AI adoption accelerates? $50 TRILLION. In comparison, the entire US Interstate Highway System cost about $704B in today’s dollars. That means the AI infrastructure bill could be 45–71x larger. But here’s the real question: What companies will power this massive AI infrastructure buildout? Explore the companies we believe are driving the AI infrastructure boom with $KAIT See KAIT’s top 10 holdings and learn more about the fund here:

In 2016, AlphaGo made a move no human understood. Move 37 looked like a mistake. It wasn’t. Now, we may be seeing Move 37 happen again, this time in the physical world. Joe and Cole sat down to break down what really happened with the Tiangong Omni robot at the Beijing Humanoid Robot Games, where its unconventional running technique broke all records in the 400m. Was it a mistake? Or did the robot figure out something we hadn’t?

AI spending reached $765 BILLION in 2026. But where is all that money going? 👀 As GPUs, servers and datacenters scale, one part of the AI infrastructure is becoming increasingly critical: connectivity and networking. The market is estimated to grow from $15B in 2026 to $156B by 2028. That’s more than 10x growth in just two years. $LUMA gives investors exposure to this critical layer of the AI infrastructure buildout ⚡ See LUMA's top 10 holdings and learn more about the fund here:

Breaking news🚨 One of the biggest names in humanoid robotics goes public tomorrow 🚀 @UnitreeRobotics is set to list on the STAR Market on August 19. KraneShares @ahern_brendan shares his thoughts on what this could mean for the future of humanoid robotics 👏