독서리서치레이더투자 프레임워크
로그인 / 회원가입
전체기술제품투자상장시장기타

독서

2026-10-3171 게시물

새로고침

Kuaishou Announces Completion of ~20 Billion RMB Capital Increase for Kling AI; Alibaba, Tencent, and Baidu Join Shareholder Structure

주제 · 快手可灵AI融资중대
2026-10-31 10:55 GMT+8

On October 30, Kuaishou Technology issued an announcement on the Hong Kong Stock Exchange stating that all conditions under the capital increase agreement for its video generation business, Beijing Kling AI, have been met. All investors have paid their capital contributions in full as agreed, and the share purchase has been completed. According to the announcement, the total capital contribution paid by all investors upon completion is approximately RMB 19.80184 billion (approximately USD 2.90533 billion), marking the largest single financing event in China's AI video sector to date.

The transaction introduced several prominent institutions and internet giants. Under the newly disclosed equity structure, controlling shareholder Lucky Labs will hold 66.68% of the equity and 52.45% of the voting rights. Kuaishou co-founder, Chairman, and CEO Cheng Yixiao holds 1.01% of the equity and 0.79% of the voting rights. Notably, Gai Kun (also known as Yu Yue), Director and CEO of Beijing Kling AI, holds only 3.02% of the equity but enjoys 23.72% of the voting rights through a special share arrangement, a design intended to ensure management control during independent operations. Previously, on September 30, Gai Kun, formerly head of Kuaishou's Community Science Line, was appointed as Director and CEO of Kling AI.

Major tech companies are also among the shareholders: Alibaba holds 1.38% of the equity and 1.09% of the voting rights; Shanghai Qishan, a subsidiary of Tencent, holds 0.56% of the equity and 0.44% of the voting rights; and Baidu holds 0.28% of the equity and 0.22% of the voting rights. Additionally, the capital increase introduced several new shareholders and adjusted subscription arrangements for some existing investors. For instance, Kingdee Software (China), Wenzhou Ruida, and Gongqingcheng Junkun obtained corresponding repurchase rights through rights and obligations transfer agreements. Meanwhile, Dakang Lanwu and BFI MIDCO signed termination agreements with Beijing Kling AI because they failed to pay their capital contributions within the agreed timeframe and are no longer investors in this round.

Kling AI is becoming one of the most standout businesses under Kuaishou. In Kuaishou's Q2 2026 financial report released on August 19, the company disclosed that Kling AI's operating revenue exceeded RMB 850 million in the second quarter, representing a year-over-year growth of over 200% and a quarter-over-quarter growth of 30.8%. On the product front, Kling launched the industry's first native 4K direct output video generation model in April, and fully rolled out Kling 4.0 in October, supporting native video generation up to 30 seconds long and multi-keyframe references, further consolidating its technological leadership.

출처:zhidx.com

리서치

DoorDash CEO says third-party AI agents generated only 'several hundred' orders

주제 · DoorDash중대
2026-10-30 21:54 GMT+8

DoorDash CEO Tony Xu revealed during the company's third-quarter earnings call that it has received only "several hundred" orders from third-party AI agents such as Muse, Instinct, and Fo. This stands in stark contrast to the company's reported average of 3 million daily deliveries in its 2025 report.

Xu noted that users are still predominantly placing orders through the DoorDash app. However, the company aims to learn more about agentic use cases and develop healthy relationships with third-party agent developers, mentioning potential future opportunities for advertising within these agents.

While DoorDash has launched its own internal AI tools, including a text-to-order agent and a chatbot, the actual commercial impact of external AI agents remains minimal at this stage.

출처:TechCrunch

리서치

Big Tech's AI Spending Boom Faces Investor Backlash Over Negative Cash Flow and Debt

주제 · AI资本支出与回报争议구조적
2026-10-31 05:17 GMT+8

Over the past two weeks, Alphabet, Amazon, Meta, and Microsoft have outlined plans to spend more than $740 billion on capital expenditures this year, which Wall Street is reading as AI infrastructure spending, including data centers and chips. However, as we enter the fourth quarter, investor enthusiasm for these outlays is giving way to concerns about negative free cash flow, mounting debt, and whether these enormous investments will ever generate enough returns.

The anxiety spilled across the market on Friday. The Nasdaq fell 2.9%, and chip stocks took a drubbing: Nvidia dropped 4%, Broadcom more than 5%, and Qualcomm 11%. Alphabet, which plummeted 12% on Thursday, its worst day on record, wiping out $500 billion in market value, fell again.

Alphabet provided the clearest example of this shift. Google’s parent company reported negative free cash flow of $9.1 billion in the third quarter, its second consecutive quarter in the red, even as quarterly capex grew to $51.4 billion. Alphabet expects to spend up to $205 billion this year, with spending projected to keep rising in 2027. Meanwhile, the wild success of AI Mode is eating into Google’s money-maker, with Search advertising growth slowing to 12%, its weakest pace in five quarters.

Amazon is in a similar boat after reporting negative free cash flow of $42.4 billion over the past 12 months, compared with $14.8 billion in positive free cash flow the year before. The company says it expects to spend $220 billion in capex this year. Even though revenue at its cloud business, AWS, grew 39%, Amazon’s stock tumbled 7% after it reported earnings. Meta is also burning cash, reporting negative free cash flow of $2.1 billion in the third quarter as capex outpaced its operating cash, guiding to as much as $145 billion in capex this year.

It’s worth noting that the money to pay for all of this AI infrastructure isn’t just coming from profits. It’s also showing up in stock and bond sales. According to S&P Global Market Intelligence, Microsoft, Alphabet, Amazon, Meta, and Oracle raised a combined $326 billion in debt in 2026, triple the amount raised last year. This change in financing structure has intensified investor skepticism about the sustainability of the AI spending boom.

출처:TechCrunch

리서치

Infineon Announces €1.1 Billion Investment in World's First 300mm GaN Fab and 900V Switch Supply for AI Data Centers

중대
2026-10-31 17:00 GMT+8

German chipmaker Infineon Technologies announced on October 27, 2026, that it will construct a new mass production facility for gallium nitride (GaN) power semiconductors using 300mm wafers at its site in Villach, Austria. According to the company, this will be the "world's first" such facility. Infineon plans to invest approximately €1.1 billion by 2034, with production expected to begin by the end of 2027. The investment falls under the support scope of the European Chips Act; the European Commission has approved €319 million in state aid, with the Austrian government's formal subsidy decision expected soon.

The shift to 300mm wafers is primarily aimed at reducing costs and increasing capacity. Infineon stated that compared to traditional 200mm wafers, the 300mm process can increase the number of chips obtained per wafer by 2.3 times, significantly lowering manufacturing costs. This move is designed to accelerate the commercialization of GaN technology in broader applications, particularly in AI data centers where efficiency demands are high.

In addition to capacity expansion, Infineon announced it will supply 900V-rated GaN monolithic bidirectional switches (BDS) to US energy technology firm Enphase Energy. Described as a "world first," the device integrates bidirectional voltage blocking and current conduction functions and is planned for use in solid-state transformers (SST) for AI data centers. By reducing the number of power conversion stages and component count, these high-voltage GaN devices help achieve higher efficiency and power density in power conversion, adapting to the growing power consumption of AI server racks.

Japanese competitor Renesas Electronics is also actively positioning itself in the AI data center power market. According to EE Times Japan, Renesas is developing a 20kW-class power system that directly converts 800V DC to 6V and has already provided samples to major OEMs and ODMs. The system uses 650V-rated GaN devices and targets an efficiency of over 96% at 20kW output. Although Renesas' efficiency figures are currently predictions based on component loss calculations and actual hardware evaluation is ongoing, this indicates that Japanese semiconductor firms are responding to the power challenges of AI infrastructure through technological innovation.

출처:EE Times Japan

리서치

Liu Zhiyuan, Chief Scientist at ModelBest, Proposes "Density Law": Open-Source Large Model Intelligence Density Doubles Approximately Every 100 Days

주제 · 面壁智能密度法则중대
2026-10-31 11:31 GMT+8

Liu Zhiyuan, Co-founder and Chief Scientist of ModelBest (面壁智能), presented a testable hypothesis at the 8th Intelligent Driving Conference held in Suzhou, China: the intelligence density of global open-source large models increases by approximately one doubling every 100 days, meaning that half the parameters and half the computational cost can achieve similar intelligence capabilities as before every 100 days. He termed this observation the "Density Law."

Liu Zhiyuan believes that the core change in large models over the past eight years has not been solely the continuous expansion of parameter scale, but rather the formation of a general technical route capable of continuously absorbing different types of data. However, as models continue to expand, data and computing power are becoming new constraints. On the data side, starting around 2024, large-scale reinforcement learning has driven models from imitation learning toward exploratory learning, enabling models to actively explore and generate new training data; on the computing side, if capability improvements rely solely on expanding parameter scales, computational demands will ultimately become unsustainable. Therefore, large models need to increase "intelligence density," i.e., achieving the same level of intelligence with fewer parameters and less computing power.

Liu Zhiyuan analogizes the Density Law to Moore's Law in the semiconductor industry: in the future, models will continue to shrink on one hand, while terminal computing power for devices such as smartphones, cars, and robots will continue to improve on the other. These two trends will eventually converge, allowing large models that previously could only run on servers to gradually enter terminals. According to the Density Law, if intelligence density continues to double approximately every 100 days, within less than two years, the number of parameters and computing power required for equivalent capabilities may drop by two orders of magnitude, compressing some capabilities of originally trillion-parameter models into the hundred-billion range and enabling deployment on terminal devices.

He categorizes future AI into three forms: super intelligence, cloud-side intelligence, and edge-side intelligence, among which edge-side intelligence will become an important carrier for physical AI. There are three additional trends for artificial intelligence entering the physical world: high intelligence density, which reduces the parameters and computing power required for model operation; full-modal perception, which understands the real world through vision, sound, and other sensors; and autonomous agents, which continuously make decisions, learn, and act in open environments.

ModelBest has already applied edge-side models to scenarios such as smartphones, automotive cockpits, satellites, and robots, and is collaborating with companies including Huawei and Samsung. Liu Zhiyuan believes that in the future, large models will continue to enter terminals such as mobile phones, PCs, cars, robots, wearable devices, and IoT devices, making edge-side intelligence a critical pathway for AI to enter the physical world.

출처:leiphone.com

리서치

Novo Nordisk VP Loïc Giraud: Hand Data to Business Users, and 8,000 Agents Can Run

주제 · 诺和诺德AI智能体部署중대
2026-10-30 17:05 GMT+8

Novo Nordisk operates in 80 countries. When Loïc Giraud joined the company in 2024 as Vice President of Data and AI Foundation, its subsidiaries and affiliates were already using the Snowflake platform to handle sales, marketing, finance, and procurement data, but the teams and technology were fragmented. He chose a different approach: first centralizing the Snowflake teams and technology into a single global function, then gradually integrating more datasets, and finally introducing AI capabilities. His core belief is that data must be placed in the hands of end users so they can build their own solutions.

Giraud believes agentic AI makes this process simple, provided there are good guardrails and controls in place. The practical path involves centralization, which creates a richer platform, followed by the introduction of capabilities like CoWork (natural language data querying) and CoCo (coding agents). Once data is ingested and processed, it can be pushed to ChatGPT, Snowflake services, or other systems. He notes that approximately 25,000 people use this platform directly or indirectly.

Specific applications demonstrate how this logic is implemented. In the United States, Novo Nordisk uses CoWork agents to help sales representatives prepare for visits with healthcare professionals; reps can query databases using natural language to access information that helps position the right drugs. Translation is another scenario: product materials written in English need to be translated into multiple languages for use across 80 affiliates, a process CoWork automates. Data engineers use CoCo to build new service connections and data pipelines, which Giraud says reduces the number of engineers needed to focus on product development.

The scale figures are significant: nearly 8,000 agents are running across the business, including for productivity tasks and research and development; over 60% of employees use AI every week. Giraud attributes this expansion to putting tools in the hands of users, supported by proper education and controls, rather than relying on the IT team to build them one by one—he stated that if his team had to build 8,000 agents themselves, it would have taken a long time. He highlighted Snowflake’s agent control plane—an orchestration layer that helps professionals connect AI models and data with strict governance—as a direction for continued advancement. It should be noted that all these figures come from Novo Nordisk’s own reports and have not been independently verified; furthermore, this solution is deeply tied to the Snowflake platform, and its portability depends on whether an organization has already selected similar data infrastructure.

출처:diginomica

리서치

Nvidia Negotiates Residual Value Guarantee for Nscale's West Virginia Data Center Construction Debt

주제 · Nscale融资与担保중대
2026-10-31 09:20 GMT+8

According to The Information and corroborated by Bloomberg, US chipmaker Nvidia is negotiating an arrangement to guarantee billions of dollars in debt financing for British AI cloud provider Nscale’s planned data center in West Virginia. This proposal marks a shift in how Nvidia supports the AI infrastructure ecosystem: while its previous financial backing primarily covered chip purchases and operational leases, the proposed “residual value guarantee” implies that if Nscale defaults during the construction phase, Nvidia would temporarily take over the facility and help find a new buyer or tenant.

The guarantee aims to help Nscale secure financing for electricity and construction costs at its West Virginia campus. The first 460-megawatt building of this project has already secured a $44.6 billion contract from Anthropic and is expected to use Nvidia’s next-generation Vera Rubin chips. Nscale is currently seeking a US listing with a target valuation of up to $35 billion, but faces significant headwinds: Microsoft recently canceled a lease agreement for Nscale’s Loughton site in the UK due to grid power delays, and Nscale’s Norwegian projects have encountered cooling leaks and regulatory hurdles. As of August, Nscale had deployed only about 25,000 advanced AI chips, far below its planned nearly 500,000.

Nvidia’s move reflects the industry’s response to growing investor anxiety over massive spending. According to The Information, Nvidia has extended roughly $170 billion in investments and credit support to neoclouds and large customers, including $1 billion in convertible financing committed to Nscale in September. With peers like CoreWeave and Nebius seeing share prices fall and bond yields spike, market focus has shifted from pure demand strength to the financing capabilities of builders. By assuming some construction risk, Nvidia attempts to maintain delivery timelines for its key customers, thereby protecting its own sales outlook.

출처:TechCrunch

리서치

Is AI VC Concentration a Power Law or a Yellow Light: Two LP Heads Explain Their Decision Frameworks

주제 · AI风投集中度争议중대
2026-10-31 12:16 GMT+8

Limited partners in the venture capital industry — those who provide capital to VC funds — find themselves in a contradictory position: they feel both excited and uneasy. TechCrunch editor-in-chief Connie Loizos recently spoke with two LP heads: Beezer Clarkson from Zurich-based LGT Capital Partners, a diversified asset manager, and Amit Bhatti from Chapel Hill, North Carolina-based TrueBridge, a firm focused solely on venture capital. Their decision frameworks reflect an attribution logic among capital providers during the AI venture capital cycle.

The most striking argument concerns concentration. Traditionally, LPs manage risk by diversifying across many VC firms. But currently, the underlying portfolios of almost all LPs have converged onto just two companies: OpenAI and Anthropic. Bhatti’s response was surprisingly positive: he believes venture capital has always been a power-law business, and concentration is a good thing, not a bad one. If a fund reaches maturity with many small positions remaining, something has gone wrong. Funds older than ten years are converging toward a few names — lots of Databricks, lots of SpaceX, lots of Stripe — and in this generation of funds, he hopes to see lots of Anthropic and lots of OpenAI. He says this is a feature, not a bug; the difference lies in how quickly the convergence is happening. Clarkson adds that early-stage opportunities still show considerable diversity because check sizes are smaller, and concentration mainly appears in later stages.

Pacing is a yellow light acknowledged by both. Clarkson notes that the normal fund cycle used to be two to two-and-a-half years, but some funds are now returning to raise new capital on a one-year cycle, which is fast. Some funds deploy capital faster because they take concentrated, large-check positions, which could be a worrying signal. Bhatti says their normal cycle is 24 to 36 months, but 12 to 18 months isn’t crazy for some rapidly advancing funds. Many funds are also upsizing mid-cycle, which is definitely unusual — partly because funding rounds have become so large. If Anthropic is raising a massive amount of capital and you allocate 20% of your fund to it, you will obviously spend your commitment ahead of schedule. Seed-round amounts have also nearly doubled, meaning if you want to lead a round, you must write checks twice as large.

However, they distinguish key differences between the current cycle and the last period of accelerated investing. Bhatti says that last time, when pacing accelerated, much capital flowed into zero-revenue crypto moonshots. Now, they see many companies growing from zero to $100 million in revenue, which was completely unheard of before. Clarkson points out that 2021 vintages are still early (five years in), but they argue against doing one-year funds. Some may catch a piece of an AI company and look excellent, but the odds of getting it right within one year are lower, and valuations at the time were extreme.

Changes in liquidity structure are another important judgment factor. Bhatti says the secondary market is now the third leg of liquidity: IPOs, M&A, and secondaries. Private equity has long had a developed secondary market, but venture capital did not — it was once a dirty word, sold only when distressed. After people got stuck in 2021, they realized they should have sold earlier. Now, more managers use it as a tool and more judiciously: if you still believe it’s a power-law company, you probably won’t exit entirely; but if you can sell 20% of your position and return the fund, you probably should. Clarkson adds that the secondary market has become much more sophisticated in venture over the past five years. Although concentrated in the top handful of companies, it is far more prevalent than before, which is good. LPs are also selling GP stakes and parts of their portfolios, and the taboo around it has weakened.

출처:TechCrunch

리서치

Andrew Chen, Head of Speedrun at US Venture Capital Firm a16z: Which Founder Assumptions Are Fundamental in the AI Era?

중대
2026-10-31 00:02 GMT+8

Andrew Chen leads the Speedrun program at US venture capital firm a16z—a fund that writes initial checks of up to $1 million into brand-new startups. A core part of his work involves assessing which founder backgrounds warrant investment. He poses a question he considers more valuable than asking "which new archetype will succeed": Of the founder assumptions previously treated as standard, which are fundamental constraints of building a company, and which are merely products of specific technology cycles and distribution channels?

He lists five founder profiles that may gain new opportunities due to AI—the solo founder, the non-technical founder, the domain expert, the young entrepreneur, and the content creator—and provides both pro and con arguments for each. Solo founders can use AI to accomplish tasks that previously required two or three co-founders, but Ben Horowitz offers a counterpoint: if a team also gains the same leverage from AI, the team might still outperform a single individual. Non-technical founders can now build prototypes, but there remains a significant gap between a demo and a reliable, scalable product; furthermore, when everyone can produce a first version, technical judgment may become even more critical. Domain experts deeply understand problems within their industries, yet hoteliers did not create Airbnb, nor did the finance industry create Kalshi—sometimes an outsider’s perspective is required to fundamentally transform an industry rather than make incremental improvements. Young entrepreneurs understand their peers and excel at consumer products, but their advantages are limited in sectors like enterprise software and healthcare, where experience and relationships matter. Content creators bring their own distribution channels, but audience attention yields only a temporary spike in exposure; building a sustainable growth engine is a different challenge.

Chen further explains why founder profiles are reshuffled in every technology cycle. He analyzes this process across three levels. The most obvious is pattern matching: if a particular category proves to generate many winners, investors observe the backgrounds of leading founders and seek similar profiles, assuming that demographic shares similar advantages. Less obvious is the alignment between technology and distribution channels: in the Web 2.0 era, viral email invitations gave college students with strong social skills a natural advantage; the iPhone unlocked GPS and mapping capabilities, making on-demand services possible; and App Store editorial picks favored developers with strong design sensibilities. The most subtle level is the self-reinforcement of investment consensus: when all investors agree on what a founder of a given era should look like, projects fitting that profile find it easier to secure follow-on funding. Even if the business model has not fully validated, founders may receive multiple rounds of "unfair" financing. This dynamic is particularly pronounced in seed and pre-seed rounds, where deals are often fragmented, requiring many individuals to individually agree in order to assemble a stack of small checks.

출처:a16z.news

리서치

US Portfolio Manager Ben Carlson: S&P 500 Earnings Growth Far Outpaces Share Prices This Year, Valuations Are Actually Compressing

주제 · AI资本开支与回报争议중대
2026-10-31 04:30 GMT+8

In 2026, S&P 500 earnings growth approached 30%, while the index rose by only about 12%. This means that valuations did not expand as they typically do in a bull market; instead, they compressed. In an article published on October 30, Ben Carlson, a portfolio manager at US-based Ritholtz Wealth Management, highlighted this phenomenon as a core observation, noting that it challenges a widely held intuition: that rising markets must be accompanied by rising valuations.

Carlson first presented the trajectory of earnings expectations. At the beginning of the year, analysts estimated full-year S&P 500 earnings at slightly above $310. After several quarters of earnings reports beating expectations, this estimate has been revised upward by nearly 20%. He emphasized that this is not a long-term forecast for earnings three, five, or even ten years out, but simply for the current year, 2026. With earnings growth around 30% and price appreciation around 12%, the gap between the two directly points to declining valuations.

To illustrate that this divergence is not anomalous, Carlson looked back at data on S&P 500 earnings growth versus total returns by decade since the 1930s. He found that the two sometimes move in sync and sometimes diverge significantly: the 1940s, 1960s, 2000s, 2010s, and 2020s (thus far) were roughly in line; in the 1930s, the stock market performed poorly but earnings were even worse; in the 1970s, earnings far outpaced the stock market; and in the 1950s, 1980s, and 1990s, the opposite was true, with stock market gains far exceeding earnings growth. He calculated that the average absolute difference between the two numbers is close to 25%, with a median of 17%. In other words, significant divergence between earnings and returns is the norm, not the exception.

Carlson also cited more specific industry cases to illustrate how market pricing reflects expectations rather than purely fundamentals. According to statistics from The Wall Street Journal, ten companies accounted for about two-thirds of profit growth this year, with just Micron and Nvidia making up one-third of that growth. Yardeni Research pointed out that semiconductor industry profits grew nearly 200%, yet their share prices rose only about 40% year-to-date, and have currently pulled back more than 20% from their highs. It appears the market does not fully trust the earnings of these semiconductor companies.

Regarding why the market remains restrained despite strong earnings growth, Carlson proposed two possibilities. One is that investors worry that AI capital expenditure is inflating earnings figures, which may fall back down in the future. A simpler explanation is that the relationship between earnings and share prices is often broken anyway, and short-term disconnects do not require special reasons.

Carlson's conclusion is that while fundamentals drive the stock market in the long run, the market and earnings are not always synchronized in the short run. Most of the time, no one knows what has already been priced in. This makes investing in the stock market both exciting and frustrating—it is precisely this uncertainty that makes equity investment simultaneously attractive and infuriating.

출처:A Wealth of Common Sense

리서치

Anthropic Removes Five-Hour Rolling Limit for Top-Tier Claude Code Users; Plans Renamed

주제 · Claude初创计划중대
2026-10-31 04:14 GMT+8

Anthropic announced on October 30, 2026, that it is removing the "rolling five-hour usage cap" for users of its AI coding tool, Claude Code, on the highest consumer tier. This change addresses interruptions faced by developers during long-running debugging sessions or large refactors.

The adjustment coincides with a renaming of Anthropic's top consumer plans: the former Max 5x is now simply Max, and Max 20x becomes Ultra. The monthly subscription prices remain unchanged at $100 and $200 respectively. Current subscribers do not need to take any action to switch to the new names.

However, the removal of the five-hour limit is not immediate for all Ultra users. Anthropic stated that some accounts must undergo verification to prevent fraud and abuse. This verification happens automatically based on account and payment history, with no option to request or expedite it. The cap is lifted only after this check is completed.

While the session-level constraint is gone, the weekly total usage allowance remains in place. Ultra users have a weekly quota ten times that of Pro users, compared to five times for Max users. This means that without the forced pauses from the five-hour cap, users may exhaust their weekly allowance sooner. Once the weekly limit is reached, users must wait for a reset or purchase additional API credits to continue working.

Additionally, the updated pricing page indicates that Max and Ultra plans now bundle monthly Claude Platform API credits equal to their subscription price ($100 and $200). This is a significant consideration for developers who use Claude Code alongside direct API calls for application building.

출처:The New Stack

리서치

Google, Microsoft, Anthropic and OpenAI Back Unified AI Crawler Opt-Out Standard

주제 · AI训练爬虫退出标准중대
2026-10-31 10:00 GMT+8

U.S. technology companies Google (Alphabet Inc.), Microsoft Corp., Anthropic and OpenAI announced on October 30, 2026, that they plan to support an upcoming website crawling standard. The initiative aims to simplify the process for website owners to prevent artificial intelligence developers from using their content to train models.

Currently, websites use a text file called "robots.txt" to manage search engine crawler behavior. To block specific AI companies from collecting training data, site administrators must add separate exclusion entries for each AI provider and every distinct scraper tool they use. The new standard will allow websites to block all scrapers collecting data for generative AI models with a single robots.txt instruction. This setting would not apply to models that perform exclusively non-generative tasks.

The four companies plan to incorporate this standard into the Robots Exclusion Protocol, which is maintained by the Internet Engineering Task Force (IETF), and have already submitted a draft proposal to the organization. Additionally, Microsoft stated in a blog post that it will offer a setting called "AI Grounding," allowing sites to prevent AI models from accessing pages to answer user questions (retrieval-augmented generation scenarios), not just for training data collection. OpenAI also plans to offer similar controls.

According to reports, OpenAI and Microsoft intend to support the new crawling standard in the coming months, while Google will follow suit sometime next year. Meta Platforms Inc. expressed support for the idea behind the technology but has not yet committed to implementing it.

출처:SiliconANGLE

리서치
다음 →
로그인 / 회원가입
독서리서치레이더투자 프레임워크

These judgments’ boundaries come from the two LPs themselves. They call the current situation a yellow light, not a green light. Bhatti says it gives him a little heartburn. Clarkson says one-year vintage funds are typically harder to get right. Bhatti notes that if things turn, concentrated holdings would be less desirable — but for now, no one complains because potential returns are too large. In other words, this framework holds only if AI companies’ revenue growth and valuations continue to materialize; if that premise changes, the power-law convergence logic shifts from a feature back to a risk.

Chen concludes that some old rules will survive: coding may become easier, but sales, recruiting, judgment, and persistence will not; domain expertise may help identify problems, but it might also make it harder to rethink an industry from scratch; a solo founder might be able to do the work of ten people, but a team of three might do the work of thirty. However, some rules are almost certainly disappearing, meaning founder profiles that looked unusual or even poor five years ago may suddenly possess real advantages. He warns that the mistake lies in replacing old pattern-matching with new labels—claiming "solo founders are good now," "domain experts are good now," or "creators are good now." The true point is that when underlying technology changes this drastically, the founder profiles that work must change too, rather than simply applying a new set of tags.