137Labs Global

The 2026 FIFA World Cup final has just concluded — Spain defeated Argentina 1-0 in extra time on July 19 to lift the trophy. What was a global sporting spectacle also served as a major stress test and growth catalyst for prediction markets.
Polymarket saw over $5 billion in World Cup-related trading volume, with the Winner market alone surpassing $4.2 billion. Kalshi contributed $1–1.2 billion on its Winner market, helping the platform set new all-time records in June, with some weekly sports volumes exceeding tens of billions of dollars and millions of new users joining. Meanwhile, AI Agent projects rapidly integrated during the tournament, offering automated analysis, signal generation, and autonomous betting — further amplifying trading activity.
In just 39 days, prediction market fee volume surged 58.3%, even as TVL slightly declined. This perfectly illustrates the nature of event-driven markets: massive capital rotates at high frequency to chase pricing inefficiencies rather than long-term locking. From the U.S. election cycle to the World Cup, prediction markets are evolving from seasonal political tools into year-round financial infrastructure.
This article, based on on-chain data from DeFiLlama and cross-referenced with reports from The Block, Dune Analytics, and industry sources, provides a comprehensive review of the sector’s history, World Cup performance, platform landscape, and Agent innovations — helping you quickly grasp the explosive logic and future potential of this emerging market.
The Explosive Path of Prediction Markets: From Seasonal to Evergreen
Prediction markets didn’t explode overnight — they climbed through multiple stepwise leaps. According to DeFiLlama
2020–Early 2023: Experimental phase. Monthly fees hovered near zero, with TVL fluctuating between tens of millions of dollars — mostly early-stage concept validation.
Q3 2023 (Aug–Oct): First real breakout. Monthly fees jumped from ~$100K to $979K, $3.126M, and $3.118M, driven by U.S. election anticipation on Polymarket.
Oct–Nov 2024: Election peak. TVL hit a record $375 million in a single month.
Late 2025–2026: Structural shift. Monthly fees climbed from $3.429M in Aug 2025 to a record $17.06M in Jan 2026. TVL stabilized in the $500–600M range, and market cap peaked at $13.03 billion in June 2026.
The core drivers expanded from politics to crypto price forecasts and macro events, ultimately turning prediction markets into evergreen tradable assets during major sporting events like the 2026 World Cup. By May 2026, the “World Cup Winner” contract had already become one of the largest single markets on both Polymarket and Kalshi, with total notional volume reaching billions of dollars.
World Cup Performance: Volume, Capital Behavior, and Emerging Players
Comparing the 39-day World Cup period (June 11 – July 19) with an equal pre-tournament window, DeFiLlama on-chain data shows clear patterns:
Cumulative fees: Rose from $42.83M to $67.78M (+58.3%).Cumulative revenue: Rose from $30.52M to $34.62M (+13.4%).Average TVL: Slightly declined from $506.7M to $497.8M (-1.8%).
The key insight: The World Cup drove velocity, not new capital lock-in. This matches the DNA of event-driven exchanges — major events activate existing funds for frequent betting rather than long-term deposits.
Kalshi (regulated platform, off-chain): As a CFTC-registered exchange, Kalshi shone during the tournament. Its Winner market exceeded $1–1.2 billion, helping drive record monthly volume (parts of June approached or exceeded $30 billion). It added millions of new users, with daily revenue repeatedly hitting records. Kalshi’s compliance-first approach makes it ideal for U.S. retail and institutional users, offering clear regulatory protection and fiat settlement.
Polymarket: Solidified its on-chain leadership. The Winner market surpassed $4.2 billion, with total World Cup-related volume around $5 billion+. Politics, crypto, and macro remained core, but sports and events saw significant uplift. Long-tail underdogs attracted huge notional volume, highlighting the speculative nature.
Other notable projects during the World Cup(DeFiLlama & public data):
Predict Fun (predict.fun): Strong performer. World Cup Winner market alone exceeded $2.7 billion, with platform cumulative volume surpassing $18 billion. User activity doubled around kickoff, daily new users hit ~6,000, and TVL jumped over $5M in a single day. 30-day prediction volume ~$356M, fees/revenue ~$712K–$761K (primarily on BSC and Blast). Its “prediction + yield” model proved highly effective.Limitless Exchange ($LMTS, token launched): Built on Base with CLOB mechanics. 30-day prediction volume ~$134M, fees ~$1.8M.InsightX: 30-day prediction volume ~$418M, fees ~$1.04M.OPINION: Focused on macro data, predictions, and news with AI Oracle (multi-agent system). Strong 30-day volume in the $400M range.Others like Overtime and PancakeSwap Prediction contributed in sports/BNB Chain niches.
These projects collectively boosted tournament-wide activity, showing the sector’s growing fragmentation. AI Agents further enhanced platforms with automated execution.
Current Landscape: Polymarket Leads, Kalshi Dominates Compliance, Emerging Players Differentiate
DeFiLlama tracks 109 prediction market protocols with total TVL ~$3.98B, 30-day fees ~$53.9M, and market cap ~$11.89B. The top tier shows clear concentration and differentiation:
Polymarket: Undisputed leader (92%+ of category fees, 84%+ of TVL). Dominates politics, crypto prices, and macro events with deepest global liquidity.Kalshi: Compliance leader. Strong in sports, ideal for U.S. users with regulatory clarity and institutional appeal.Predict Fun: Deeply integrated with Binance ecosystem (wallet support + traffic). “Prediction + yield” model; strong sports/event focus.Limitless Exchange ($LMTS): Built on Base with CLOB for high-performance short-term markets.InsightX & OPINION: Rising macro/AI-focused players with solid volume.Rain: Arbitrum-based, permissionless + AI resolution; previously hit top-3 TVL with major liquidity commitment.Others (Overtime, PancakeSwap Prediction, etc.): Niche sports/chain supplements.
Multi-Dimensional Comparison:
Volume & Liquidity: Polymarket leads global events; Kalshi excels in sports & compliance; emerging players shine in verticals.TVL & Capital: High turnover dominates; ecosystem-linked projects (e.g., Predict Fun) saw sharp TVL spikes during events.Categories & Users: Polymarket = broad/global; Kalshi = U.S./sports; Predict Fun = Binance users; OPINION = macro/institutional insights.Regulation & Fees: Kalshi offers compliance edge; on-chain projects provide transparency and low costs.Accuracy & Behavior: Smart money drives efficiency; retail often loses long-term.Agent Integration: AI Agents add automation layers across all platforms.
Kalshi owns compliance, Polymarket owns global liquidity, and emerging players differentiate through ecosystem ties and innovation (especially AI Oracles).
Conclusion
The World Cup may be over, but the prediction market story is just getting started. With the tournament behind us, we’re immediately stepping into the 2026 U.S. Midterms, the 2028 Presidential Election cycle, ongoing macro releases, AI milestones, and evolving crypto narratives. Political and macro themes are ready to take the baton, while AI Oracles, Agent automation, and cross-platform innovation quietly reshape the sector’s long-term architecture.
Whether you’re a seasoned liquidity hunter or exploring new mechanisms, now is an excellent time to position for the next phase. Stay attuned to real-world events, leverage platform strengths, and manage risk wisely — the next wave of opportunities in prediction markets could be even bigger than the World Cup.
Data as of July 20, 2026, sourced from DeFiLlama, The Block, Dune Analytics, and public market observations. Markets involve risk — trade responsibly.

I. Why Did Kimi Suddenly Suspend New User Subscriptions?
In mid-July, just two days after Moonshot AI released its open-weight large language model Kimi K3, the company made an announcement that quickly became one of the most discussed stories in China's AI industry. Due to sustained high utilization of its computing resources, Moonshot said it would temporarily suspend new consumer subscriptions, prioritizing existing GPU capacity for current paying members while accelerating the deployment of additional compute clusters. Unlike the familiar outages of the traditional internet era—when services went offline because of server failures or database issues—the explanation this time reflected a defining characteristic of the generative AI age: the product itself was functioning as intended, but demand for the model had outpaced the infrastructure needed to serve it.
This seemingly routine operational decision revealed a much larger shift taking place across the AI industry. In the traditional software business, rapid user growth generally leads to stronger network effects and improving unit economics. Although companies must continue expanding their server capacity, the marginal cost of serving each additional user typically declines as scale increases. Large language models operate under a very different economic reality. Every prompt submitted by a user, every line of code generated, every document summarized, every image interpreted, and every autonomous agent task executed requires significant GPU resources to perform inference. Unlike conventional software, AI services incur substantial computational costs every time they are used. As a result, a growing user base does not necessarily reduce costs through scale—in many cases, it increases operating expenses at an equally rapid pace.
For frontier AI companies that remain heavily invested in research and model development, this creates a paradox. The more capable the model becomes, the more complex the tasks users entrust to it. The more valuable the product becomes, the more frequently users rely on it. Success therefore brings not only greater visibility and adoption but also rapidly rising inference costs and increasing pressure on cash flow. In this context, Kimi's decision to pause new subscriptions was far more than a temporary shortage of GPUs. It became a visible reminder that the competitive dynamics of large language models are entering a new phase. As model capabilities continue to converge and the technological gap between open and proprietary models narrows, long-term competitiveness will depend less on benchmark rankings or parameter counts and more on whether companies can deliver AI services that are affordable, reliable, scalable, and economically sustainable. In other words, competition is gradually shifting from building better models to running better AI businesses.
Only two years ago, discussions surrounding China's leading AI startups largely revolved around familiar questions: Which company had trained the strongest model? Who had raised the largest funding round? Which startup was closest to OpenAI in terms of technological capability? Today, however, the conversation has become considerably more pragmatic. If an AI model succeeds in attracting millions of users, can the company actually support that success operationally? Can its infrastructure keep pace with demand? Can its revenue eventually offset the enormous costs of inference? The compute constraints encountered by Moonshot AI are unlikely to be unique. They represent a challenge that virtually every frontier model developer will eventually face. Moonshot simply became one of the first companies willing to acknowledge that reality publicly.
From an industry perspective, the explosive reception of Kimi K3 was hardly unexpected. Since DeepSeek sparked a new wave of enthusiasm for open-weight reasoning models, developers across China have shown growing interest in high-performance open-source AI systems. Kimi K3 built upon Moonshot's long-standing strength in handling extremely long contexts while delivering substantial improvements in coding, reasoning, and agent capabilities. As a result, developers, enterprise customers, and ordinary users all rushed to evaluate the model almost simultaneously, generating an enormous surge in inference requests that quickly pushed available computing resources to their limits.
Yet stepping back from the immediate event reveals a more interesting question. The real issue is not simply why Kimi ran short of GPUs, but why another company widely regarded as one of China's leading foundation-model developers—MiniMax—has rarely encountered similar public discussions.
The answer is not that MiniMax possesses unlimited computing resources, nor that demand for its models is significantly lower. Rather, the two companies have pursued fundamentally different strategies since their inception. Moonshot AI has focused almost exclusively on turning Kimi into a universal AI assistant—an intelligent gateway for information retrieval, knowledge work, and future productivity. Consequently, nearly all of the company's brand recognition, user traffic, and product identity are concentrated around a single flagship application. MiniMax, by contrast, has never placed all of its bets on one conversational interface. Instead, it has built a portfolio of AI products spanning virtual companions, video generation, speech synthesis, music generation, enterprise APIs, and other multimodal services, allowing its underlying foundation models to reach users through multiple products and multiple markets.
Viewed from this perspective, Kimi's subscription pause was more than an isolated operational adjustment. It served as one of the clearest signals yet that China's large-model industry is entering a new stage of development. And it naturally raises an equally important question: why have two companies operating at the forefront of the same technological wave chosen such fundamentally different paths toward building sustainable AI businesses?

II. Why Did Kimi Suddenly Explode in Popularity? A Delayed "Stress Test" by the Real World
A little more than a year ago, Moonshot AI was still considered one of the more understated startups in China's fast-growing AI landscape. While competitors such as MiniMax were already attracting attention with products like AI companions and Zhipu AI was aggressively expanding its enterprise business, Moonshot pursued a far more focused strategy. Rather than launching multiple consumer applications in search of rapid user growth, the company concentrated almost all of its resources on a single ambition: building a truly general-purpose AI assistant.
That strategy reflected the background of its founding team. CEO Yang Zhilin, a longtime researcher in large language models, chose to prioritize advances in core model capabilities instead of chasing short-lived trends such as AI-generated art or digital avatars. Moonshot devoted substantial effort to one area in particular—long-context reasoning. At a time when most competing models supported only tens of thousands of tokens, Kimi introduced the ability to process documents containing hundreds of thousands—and eventually millions—of Chinese characters in a single session. This gave it a meaningful advantage in use cases such as reading research papers, analyzing financial reports, reviewing legal contracts, and summarizing lengthy technical documents, quickly attracting knowledge workers, researchers, lawyers, and software engineers.
For a considerable period, Kimi became almost synonymous with long-context AI. Many users were drawn to the product not because they believed it was necessarily more intelligent than ChatGPT, but because it could ingest an entire PDF, analyze hundreds of pages at once, and produce concise, useful summaries. In an increasingly crowded AI market, this distinctive positioning allowed Moonshot to avoid direct competition with global incumbents while cultivating a loyal user base around a clearly differentiated capability.
But exceptional context length alone was never going to be enough to sustain a frontier model company. As OpenAI, Anthropic, Google DeepMind, and more recently DeepSeek continued to push advances in reasoning, coding, and autonomous agents, expectations for large language models began to evolve. Users no longer wanted AI systems that could simply read more; they wanted systems that could reason more deeply, generate higher-quality code, plan complex workflows, and execute increasingly sophisticated tasks. Coding assistants, advanced mathematical reasoning, agentic workflows, and tool use gradually became the new benchmarks by which leading models were judged.
It was against this backdrop that Kimi K3 was introduced.
The significance of K3 was not merely that it contained more parameters than previous generations. More importantly, it represented a substantial leap in general reasoning capability. Across coding, web development, long-horizon planning, and agent-based task execution, K3 quickly established itself among the strongest open models available. Its programming performance, in particular, attracted widespread attention from developers, many of whom began comparing it directly with Claude and GPT-series models. For Moonshot, this marked an important transition: Kimi was no longer simply competing to become China's best AI assistant—it was entering the global race for frontier foundation models.
Moonshot also adopted a notably more open strategy by releasing K3 with open weights. This decision significantly lowered the barrier for developers to deploy, fine-tune, and integrate the model into their own applications, allowing K3 to spread rapidly through the global open-source community. The move strengthened Moonshot's technical reputation and expanded its developer ecosystem, but it also produced an inevitable consequence: an explosion in real-world usage.
Open-weight models are a double-edged sword. The better the model performs, the more developers want to experiment with it. Every deployment, benchmark comparison, coding project, and agent workflow translates into additional inference requests. Unlike training, which is performed once, inference occurs continuously, every time users interact with the model. Success therefore generates not only greater visibility but also exponentially increasing computational demand.
From this perspective, Kimi's decision to suspend new subscriptions should not be viewed as an isolated incident. It was, in many ways, the first large-scale stress test of Moonshot's infrastructure under real-world conditions.
For years, frontier AI companies have relied heavily on benchmark scores such as MMLU, HumanEval, and other standardized evaluations to demonstrate technological progress. While these benchmarks remain useful indicators of model capability, they ultimately represent carefully designed testing environments rather than the unpredictable complexity of everyday use. What truly determines whether a foundation model is ready for production is not how it performs on a leaderboard, but whether it can continue delivering reliable service when confronted with millions of simultaneous users solving real problems.
That is precisely what happened after Kimi K3 was released. Developers immediately began using it to write software, build AI agents, automate workflows, and prototype applications, while ordinary consumers relied on it for studying, translation, office productivity, content creation, and countless other tasks. Unlike a simple chatbot conversation, a sophisticated agent workflow may require dozens—or even hundreds—of sequential inference steps and multiple tool invocations before producing a final answer. What users perceive as a single response often represents an extensive chain of computational operations running behind the scenes.
This also explains why user numbers alone have become an increasingly misleading metric for measuring the infrastructure demands of modern AI systems. A casual user who asks a handful of questions each day consumes only a fraction of the computing resources required by a developer running autonomous coding agents for hours at a time. In the age of generative AI, it is no longer the number of registered users that determines infrastructure pressure, but the volume, complexity, and economic value of inference requests.
Seen from this perspective, Moonshot's compute bottleneck can almost be interpreted as a high-class problem. It demonstrates that Kimi has evolved beyond being a technological curiosity and has become a genuine productivity tool that users depend on for real work. At the same time, however, that success introduces an even greater challenge. For Moonshot, the next stage is no longer simply building a more capable model—it is finding a sustainable way to ensure that rapidly growing inference costs can ultimately be supported by equally sustainable revenue.

III. Why Didn't MiniMax Become Another Kimi?
If public attention were the only measure of success, Moonshot AI would undoubtedly be one of China's most visible AI startups over the past year. From Kimi's early popularity driven by its long-context capabilities, to the global developer buzz surrounding the release of Kimi K3, and most recently its decision to suspend new user subscriptions due to compute constraints, nearly every major move by the company has become headline news. Yet when the focus shifts away from media attention and toward long-term business development, another company—MiniMax—has quietly been following a very different trajectory.
The contrast is not rooted in differences in technological capability. MiniMax is widely recognized as one of China's leading foundation model companies, and its models consistently rank among the country's strongest. The real distinction lies in how the two companies have answered a more fundamental question: How should a foundation model company build a sustainable business?
For many international users, MiniMax first gained recognition through Talkie, its AI companion app. At a time when the industry was overwhelmingly focused on building general-purpose AI assistants in the wake of ChatGPT's success, MiniMax chose a direction that initially seemed unconventional. Rather than competing head-on in productivity tools, it invested heavily in AI companionship. Talkie allows users to interact with AI characters with distinct personalities, creating long-term conversational relationships that emphasize emotional engagement rather than task completion. While the concept inevitably invited comparisons with Character.AI, MiniMax differentiated itself through stronger language models, richer character design, and more sophisticated long-term conversational memory.
The strategy proved unexpectedly successful. Talkie quickly attracted users across North America, Japan, and other overseas markets, validating an important assumption: people are willing to pay for AI not only to improve productivity, but also to satisfy emotional and social needs. This insight allowed MiniMax to establish one of its earliest recurring revenue streams while many competitors remained focused almost exclusively on showcasing model performance.
The company later introduced Xingye , a Chinese-language AI companion built on the same underlying technology. Although AI companionship remains one of the more controversial sectors within consumer AI, it enabled MiniMax to accumulate a loyal subscription base and gain practical experience in monetization, user retention, and long-term consumer engagement much earlier than many of its peers.
If Talkie represented MiniMax's vision for emotional interaction, Hailuo AI reflected its ambitions in the era of multimodal content generation.
Over the past year, the center of gravity in generative AI has gradually shifted. Users increasingly expect AI not merely to answer questions, but to produce content directly. Following the rapid development of image generation, AI video has emerged as one of the industry's most fiercely contested frontiers. From OpenAI's Sora to Google's Veo and a growing number of Chinese competitors, video generation has become a strategic battleground for nearly every foundation model company.
MiniMax was among the earliest Chinese startups to invest heavily in this space. Rather than simply replicating products developed overseas, Hailuo focused on practical creative workflows, including filmmaking, advertising, digital marketing, and short-form content production. Its latest video models are capable of generating longer sequences, maintaining stronger temporal consistency, and producing more controllable human motion—capabilities that make the platform increasingly useful for professional creators rather than merely demonstrating technical possibility.
As a result, visiting MiniMax's website today feels fundamentally different from visiting the homepage of a traditional chatbot company. Instead of presenting a single conversational interface, the company offers an expanding portfolio of AI products built upon the same foundation models. Text generation, video creation, speech synthesis, music generation, AI agents, and enterprise APIs all coexist within a unified technological ecosystem while targeting entirely different user groups.
This reflects a fundamentally different product philosophy.
Moonshot AI believes that one exceptionally capable AI assistant can eventually become the universal interface for knowledge work, allowing users to search, read, write, code, and collaborate through a single intelligent entry point. MiniMax, by contrast, views the foundation model less as a destination than as infrastructure. In its vision, large language models should function like an operating system or cloud platform—powering numerous applications across different markets rather than funneling every user into the same chat window.
Neither strategy is inherently superior. However, they produce very different commercial dynamics.
Moonshot benefits from extraordinary brand concentration. Kimi has become synonymous with the company itself, enabling every product improvement to reinforce a single, highly recognizable identity. The downside is equally obvious: virtually all user traffic and inference demand accumulate around one product, meaning that every major breakthrough in model capability immediately translates into greater pressure on the underlying compute infrastructure.
MiniMax presents almost the opposite picture. Many users know Talkie without realizing it is developed by MiniMax. Others generate AI videos through Hailuo without recognizing that the same company provides the underlying model. Enterprise customers may integrate MiniMax APIs into their products while never interacting with its consumer-facing applications at all. Although this fragmented brand presence may appear weaker from a marketing perspective, it distributes both user demand and commercial risk across multiple products instead of concentrating them in a single application.
More importantly, it also creates a more diversified revenue structure.
Moonshot's current commercial model revolves primarily around Kimi subscriptions, developer APIs, and future enterprise services. Consequently, the financial performance of the company remains closely tied to the continued growth of a single flagship product. MiniMax, however, generates revenue through multiple channels simultaneously—including AI companion subscriptions, video generation services, developer APIs, and enterprise AI solutions. When one business enters a period of heavy investment, other product lines can continue generating cash flow, providing greater flexibility in managing the substantial costs associated with large-scale inference.
This does not mean MiniMax is free from computational challenges. AI video generation is significantly more GPU-intensive than text generation, while conversational companion products require persistent inference to maintain long-term memory and user engagement. The difference is that these costs are distributed across multiple businesses and markets rather than concentrated in a single consumer application.
In many ways, the two companies embody two distinct visions of what a foundation model company should become. Moonshot is building what it hopes will evolve into a universal AI assistant—the primary gateway to knowledge work. MiniMax is building something closer to an AI ecosystem, allowing its models to power applications across entertainment, creativity, communication, and enterprise software. As the industry moves beyond technological experimentation and into large-scale commercialization, the market will ultimately determine which approach proves more sustainable. The real question is no longer which company can build the better model, but which one can build the better business around it.

IV. Two Diverging Paths to Commercialization
Just two years ago, China's leading AI startups were largely competing on the same playing field. Success was measured by familiar metrics: model size, benchmark scores, context length, and training capabilities. Investors rewarded companies that built increasingly powerful foundation models, assuming that commercial opportunities would naturally follow once the technology matured.
That assumption is now being tested.
As foundation models continue to converge in capability, technical leadership alone is becoming less sufficient as a competitive advantage. Increasingly, the question facing every frontier AI company is no longer who can build the smartest model, but who can operate it as a sustainable business. Foundation models are evolving from research achievements into commercial infrastructure, and that shift changes almost every aspect of competition.
This is where Moonshot AI and MiniMax begin to represent two fundamentally different visions of the AI industry.
Moonshot's strategy closely resembles the early ambitions of OpenAI. Its goal is to make Kimi the universal interface for knowledge work—a single AI assistant capable of searching information, reading documents, writing reports, generating code, and eventually coordinating increasingly sophisticated AI agents. Everything the company builds ultimately strengthens one product and one brand.
The advantages of this strategy are obvious. Concentrating resources around a single flagship product creates an exceptionally strong brand identity. Every improvement to the model reinforces user recognition, while every new capability increases the value of the same platform. History offers many examples of this approach succeeding. Google became synonymous with internet search because users consistently returned to a single entry point. WeChat evolved into China's dominant digital ecosystem by first becoming the primary gateway to mobile communication. Moonshot appears to be pursuing a similar ambition for the AI era: to make Kimi the place where users begin virtually every knowledge-intensive task.
Yet building a universal AI gateway comes with equally significant challenges.
When millions of users rely on the same application, every improvement in model capability leads directly to higher inference demand. More users generate more conversations, more agent workflows, and more complex reasoning tasks, all of which consume increasingly expensive GPU resources. Kimi's decision to suspend new subscriptions illustrates this reality. The company's infrastructure simply expanded more slowly than demand, forcing it to temporarily limit growth in order to preserve service quality for existing users.
MiniMax has taken a markedly different approach.
Rather than building a single destination product, it treats foundation models as shared infrastructure capable of supporting many independent businesses. Talkie serves users seeking AI companionship. Hailuo targets creators producing AI-generated video. Speech models address enterprise voice applications. Music generation explores creative production, while APIs connect developers building their own AI-powered services. Instead of directing all traffic into one conversational interface, MiniMax distributes its technology across multiple products, customer groups, and revenue streams.
In many respects, this resembles the evolution of cloud computing. Most AWS customers are not interested in the underlying infrastructure itself; they purchase storage, databases, or machine learning services depending on their specific needs. Likewise, MiniMax is positioning its models less as consumer products and more as a technological foundation powering a diverse ecosystem of AI applications.
These strategic differences naturally extend to commercialization.
Moonshot's revenue today depends primarily on Kimi subscriptions, developer APIs, and future enterprise offerings. As Kimi grows, revenue has the potential to grow alongside it—but so do infrastructure costs. The company's commercial success remains tightly coupled to the continued expansion of a single product.
MiniMax's revenue model is inherently more diversified. AI companionship subscriptions, video generation services, enterprise AI solutions, developer APIs, and future multimodal products each contribute to different parts of the business. Although every product requires significant investment in research and computing infrastructure, no single business bears the entire financial burden. The result is a portfolio approach that distributes both opportunity and risk.
The contrast also reflects two different philosophies of capital allocation.
During the first wave of the generative AI boom, investors focused overwhelmingly on technological capability. Teams with stronger researchers and more advanced models attracted larger funding rounds because the prevailing belief was that commercialization could always come later.
That mindset is beginning to change.
Investors are now asking more practical questions. How quickly can inference costs decline? What percentage of users are willing to pay? How predictable are recurring revenues? Can the company eventually finance its own infrastructure instead of relying on venture capital?
In this context, MiniMax's move toward the public capital markets signals a willingness to have its commercial performance evaluated by investors, while Moonshot, still privately held, retains greater flexibility to prioritize long-term technological development without the pressure of quarterly earnings expectations.
Ultimately, the competition between these companies is no longer about who publishes the larger model or achieves the highest benchmark score. Those milestones may determine who reaches the frontier of AI research, but they are unlikely to determine who builds the most enduring business. The next phase of competition will be decided by something far more difficult: whether a company can transform extraordinary intelligence into a commercially sustainable enterprise.

V. Conclusion
If the defining keyword of the global AI industry over the past two years has been "models," then the keyword shaping its next stage is increasingly becoming "operations." As foundation models continue to converge in capability, they are evolving from scarce technological breakthroughs into digital infrastructure, shifting competition away from parameter counts and benchmark scores toward inference costs, product experience, business models, and operational efficiency. Moonshot AI's decision to suspend new user subscriptions due to compute constraints may appear to be a routine operational adjustment, but it reflects a broader reality of the foundation model era: the more capable a model becomes, the higher its inference costs. AI companies are no longer judged solely by their ability to build smarter models, but by their ability to turn every inference into sustainable commercial value. In this sense, Moonshot's vision of building Kimi into a universal AI gateway and MiniMax's strategy of developing a diversified portfolio of AI products represent two different answers to the same commercial challenge. More importantly, they illustrate how China's foundation model industry is moving from a competition driven by technological breakthroughs to one defined by business execution. In the years ahead, what will ultimately determine a company's competitiveness will not simply be the quality of its models, but its ability to transform intelligence into a sustainable business.

في 14 و15 يوليو، ظهر رئيس مجلس الاحتياطي الفيدرالي كيفن وورش أمام الكونغرس الأمريكي لمدة يومين متتاليين. وأدلى بشهادته أمام لجنة الخدمات المالية في مجلس النواب، ولجنة المصارف والإسكان والشؤون الحضرية في مجلس الشيوخ بشأن تقرير مجلس الاحتياطي الفيدرالي نصف السنوي للسياسة النقدية.
بصفته أول شهادة برلمانية للإدلاء بها منذ توليه المنصب، اجتذبت جلسات الاستماع اهتمامًا كبيرًا من الأسواق المالية العالمية.
بالنسبة للمستثمرين، وفرت جلسات الاستماع أكثر من مجرد تقييم شامل لاقتصاد الولايات المتحدة. كما أتاح ذلك فرصة مهمة لدراسة الاتجاه المستقبلي للسياسة النقدية. وعلى مدار يومين، تناول وورش التضخم وأسعار الفائدة والميزانية العمومية لمجلس الاحتياطي الفيدرالي واستقلال البنك المركزي والأثر الاقتصادي للذكاء الاصطناعي، مُرسلًا عدة إشارات سياساتية لافتة.

في 16 يوليو 2026، أصدرت شركة تايوان لصناعة أشباه الموصلات (TSMC)، الرائدة عالميًا في مجال «التمركز النقي» لخدمات تصنيع أشباه الموصلات، تقرير أرباحها للربع الثاني، وكان السبب وراء جذب هذه المجموعة من النتائج اهتمامًا وثيقًا من الأسواق الرأسمالية العالمية وسلسلة إمداد أشباه الموصلات وكبار المسؤولين التنفيذيين عبر صناعة التكنولوجيا، ليس فقط كون TSMC شركة ضخمة الحجم وذات أهمية استراتيجية هائلة، بل أيضًا أنه في ظل استمرار التوسع في الإنفاق على البنية التحتية للذكاء الاصطناعي، وتصاعد حدة المنافسة في تقنيات العمليات المتقدمة، وتسارع إعادة هيكلة سلسلة إمداد أشباه الموصلات العالمية، أصبحت TSMC واحدة من أهم النوافذ التي يمكن للمستثمرين والمشاركين في الصناعة من خلالها تقييم القوة الحقيقية لطلب الذكاء الاصطناعي، واختناقات الإنتاج التي تؤثر في الرقائق المتقدمة، والدورة الأوسع لإنفاق رأس المال في قطاع التكنولوجيا.

هل انغمَرت بالتغطيات الإخبارية اليوم عن شركة تشانغشين تكنولوجي؟ حوالي 15 يوليو، سريعة الانتشار أخبار عن منصة Hyperliquid (المبنية على بروتوكول Trade.xyz HIP-3) بأنها على وشك إطلاق عقود CXMT الدائمة، ما هيمن بسرعة على محادثات مجموعات الاستثمار وعلى وسائل التواصل الاجتماعي.
باعتبارها أكبر شركة لتصنيع رقائق DRAM في الصين، والرابعة عالميًا، حدّدت شركة تشانغشين تكنولوجي سعر الاكتتاب عند 8.66 رنمينبي للسهم في 14 يوليو. وستبدأ الاكتتابات عبر الإنترنت وغير المتصلة في 16 يوليو، ومن المتوقع إدراجها في بورصة شنغهاي ضمن سوق STAR حوالي 27 يوليو (رمز السهم: 688825.SH). ولا يُعد هذا الحدث فحسب أكبر حدث اكتتاب عام أولي (IPO) محوري في سوق رأس المال الصيني في عام 2026، بل هو أيضًا ثاني أكبر حدث في تاريخ سوق STAR، وأكبر طرح لأسهم جديدة في أسهم A هذا العام، مع تمويل مخطط بقيمة 295 مليار رنمينبي، ما يجعله واحدًا من أكبر اكتتابات آسيا في السنة.

في 14 يوليو (بالتوقيت المحلي)، أصدرت وزارة إحصاءات العمل الأميركية (BLS) تقرير مؤشر أسعار المستهلكين (CPI) لشهر يونيو 2026. وباعتباره واحدًا من أكثر المؤشرات الاقتصادية الكلية متابعةً في الأسواق المالية العالمية كل شهر، فإن مؤشر CPI لا يعكس فقط التغيرات في أسعار المستهلكين عبر الولايات المتحدة، بل يعمل أيضًا كمعيار رئيسي لتقييم التوجه المستقبلي للسياسة النقدية لدى الاحتياطي الفيدرالي. ونتيجة لذلك، كان المستثمرون والاقتصاديون وصانعو السياسات ومشاركو السوق يراقبون الإصدار عن كثب. وأظهرت أحدث الأرقام أن التضخم العام تراجع بشكل ملحوظ مقارنةً بالأشهر السابقة، حيث جاء كل من مؤشر CPI الرئيسي وCPI الأساسي دون توقعات السوق. وقد عززت البيانات فورًا معنويات السوق، إذ دفعت مؤشرات الأسهم الأميركية الرئيسية إلى الارتفاع، وخفضت عوائد سندات الخزانة، وأضعفت الدولار الأميركي، وذلك بعد أن خفّض المستثمرون توقعاتهم بشأن المزيد من التشديد النقدي من جانب الاحتياطي الفيدرالي. ومع ذلك، وفي ظل التفاؤل المتجدد، برز سؤال أكثر جوهرية: هل يشير هذا التحسن إلى أن التضخم في الولايات المتحدة أخيرًا بات تحت السيطرة، أم أنه يعكس فقط تراجعًا مؤقتًا مدفوعًا بعوامل قصيرة الأجل؟

تعيش سوق الأسهم في كوريا الجنوبية رحلة درامية على طريقة الأفعوانية في عام 2026. فقد قفز مؤشر KOSPI بأكثر من 85% خلال النصف الأول من العام، مسجلاً مستوى قياسيًا بلغ 9,385 نقطة في يونيو. ومع ذلك، في منتصف يوليو، وبسبب تصفيات رافعة مالية ضخمة وتدفقات رأسمالية أجنبية خارجية، تراجع بسرعة إلى ما يُعد سوقًا هابطة فنيًا.
من ذروته، انخفض المؤشر بنحو 27% تقريبًا (ليغلق عند حوالي 6,857 نقطة اعتبارًا من 14 يوليو)، بينما انخفضت شركة SK Hynix (000660.KS) بأكثر من 30% عن أعلى مستوى لها مؤخرًا (مع تسجيل أكبر هبوط في يوم واحد بلغ 15.4%). كانت ديون الهامش قد بلغت مستوى قياسيًا بلغ 38.6 تريليون وون كوري.

مقدمة
خلال العامين الماضيين، ركّزت المناقشات في الأسواق حول الذكاء الاصطناعي بشكل أساسي على ما إذا كانت قدرات النماذج يمكنها الاستمرار في التحسن، وما إذا كان بإمكان الذكاء الاصطناعي التوليدي أن يصبح تقنية ذات أغراض عامة comparable للإنترنت والهواتف الذكية، وكيف قد تُحوِّل نماذج اللغات الكبيرة صناعات مثل البحث والبرمجيات والإعلانات وخدمات المؤسسات؛ ومع ذلك، ومع تقدّم أداء النماذج، وتسارع تبنّي المستخدمين، ودخول البنية التحتية للحوسبة العالمية مرحلة بناء مكثفة، تحوّلت أنظار المستثمرين بشكل متزايد نحو العوائد التجارية، فأضحى السؤال المركزي الآن هو ما إذا كانت الكميات الهائلة من رأس المال التي تستثمرها كبرى شركات التكنولوجيا في الذكاء الاصطناعي يمكن تحويلها في النهاية إلى إيرادات مستقرة وأرباح وتدفقات نقدية حرة.

مرّت ثلاثة أيام منذ أن قدّمت شركة SK Hynix الكورية الجنوبية الرائدة في مجال الذاكرة أول ظهور لها في الولايات المتحدة عبر الإدراج. يستعرض هذا المقال الحدث من خلال أربع زوايا رئيسية: تفاصيل الإدراج والقوة التجارية لدى SK Hynix، والدعم الناجم عن ارتفاع الإنفاق الرأسمالي المخصص للذكاء الاصطناعي في قطاع الذاكرة، وتحليل Goldman Sachs لسلسلة قيمة الذكاء الاصطناعي، والقيمة الاستراتيجية لـ SK Hynix مع رؤى مستقبلية.
سعّرت شركة SK Hynix إعلاناتها الإقليمية (ADRs) بسعر 149 دولاراً. بدأت التداولات تحت الرمز SKHYV ثم تم لاحقاً تغييرها إلى SKHY. رفعت الشركة مبلغاً قياسياً قدره 26.5 مليار دولار في أكبر طرح أولي (IPO) في الولايات المتحدة لشركة أجنبية. وارتفعت قيمة إعلاناتها الإقليمية بنحو 13% في أول يوم، ما دفع القيمة السوقية إلى أكثر من 1.2 تريليون دولار. تُستخدم العوائد بشكل رئيسي لتوسيع مصانع تصنيع الرقائق (wafer fabs) في كوريا، والاستحواذ على معدات EUV متقدمة لتسريع خطة بناء سعة رقائق الذاكرة من نوع HBM. وباعتبارها مورّداً رئيسياً لشركة NVIDIA وغيرها من مسرّعات الذكاء الاصطناعي عالية المستوى، فإن منتجاتها تشغل بالفعل موقعاً حاسماً ضمن سلاسل إمداد مراكز البيانات العالمية.

١. دورة أسعار الفائدة التي خلقت سوقًا جديدًا
إذا عدنا إلى عام ٢٠٢١، كانت فكرة الأصول الواقعية في العالم، أو RWA، لا تزال إلى حد كبير نظرية. كانت تشير إلى عملية تمثيل الأصول من الاقتصاد التقليدي على سلاسل الكتل، وإتاحة إصدارها وحيازتها وتحويلها بصيغة مُرمّزة (tokenized). ومع ذلك، في ذلك الوقت لم تكن الفكرة قد تطورت بعد إلى سوق تجارية مقنعة. فمن جهة، أبقى مجلس الاحتياطي الفيدرالي أسعار الفائدة قريبة من الصفر لسنوات، بينما ظلت عوائد أذون الخزانة الأمريكية قصيرة الأجل أقل من ١٪، ما ترك دخلًا تقليديًا منخفض المخاطر بالكاد كان مغريًا بما يكفي للانتقال إلى السلسلة. ومن جهة أخرى، كانت صناعة العملات المشفرة تتوسع بسرعة؛ إذ حققت قطاعات مثل التمويل اللامركزي (DeFi) والرموز غير القابلة للاستبدال (NFTs) وGameFi وغيرها من القطاعات المضارِبة عوائد درامية. كان المستثمرون أكثر اهتمامًا بالأصول التي يمكن أن ترتفع عشرة أضعاف أو حتى مئة ضعف، مقارنةً بالمنتجات التي توفر بضعة نقاط مئوية فقط من الدخل السنوي. في ظل تلك الظروف، بدا أن RWA أقرب إلى اتجاه تكنولوجي منه إلى مشروع مالي قابل للحياة.

في 8 يوليو 2026، أصدرت "Binance Research" هذا التقرير التفصيلي بعنوان: "Stablecoins: Transforming The Financial Landscape". يعاين التقرير بشكل منهجي كيف تطورت العملات المستقرة من مجرد جسور داخل منظومة العملات المشفرة إلى بنية تحتية أساسية تعيد تشكيل النظام المالي العالمي.
يقدم هذا المقال شرحًا تفصيليًا للأفكار الرئيسية للتقرير والبيانات الحاسمة والاتجاهات الناشئة، مما يساعد القراء على فهم التحولات البنيوية التي تجري في مجال العملات المستقرة وما تترتب عليها من آثار عميقة للأفراد والمؤسسات وللمشهد المالي الأوسع.

بعد أسبوع واحد فقط من إطلاقها الرئيسي العام على الشبكة الرئيسية في 1 يوليو، تجاوز إجمالي القيمة المقفلة (TVL) لسلسلة Robinhood Chain عتبة 100 مليون دولار، حيث بلغ ذروته قرب 106 ملايين دولار، مع زيادة خلال 24 ساعة وصلت إلى 159% في إحدى اللحظات. وقد دفعت هذه الزيادة الانفجارية، التي غذّتها بروتوكولات إقراض DeFi وزادها تضخّم سيولة تداول عملات الميم، بسرعة بسلسلة الطبقة الثانية الجديدة إلى دائرة الضوء.
في 8 يوليو، شهدت عملة ميم على السلسلة (on-chain) ارتفاعًا مؤقتًا في القيمة السوقية لتتجاوز 110 ملايين دولار قبل أن تتراجع إلى نحو 104 ملايين دولار. وقد حققت مكاسب تتجاوز 13.9x خلال 24 ساعة (وكانت أعلى عند ذروة اللحظات)، مع وصول حجم التداول اليومي إلى مئات ملايين الدولارات. كما قفز إجمالي حجم التداول في DEX الخاص بسلسلة Robinhood Chain إلى ما يزيد عن 500 مليون دولار، ما جعلها واحدة من أكثر المواضيع سخونة في السوق.

أدى صدور محاضر اجتماع اللجنة الفيدرالية للسوق المفتوحة التابعـة للاحتياطي الفيدرالي (FOMC) لشهر يونيو إلى تحويل انتباه المستثمرين بعيدًا عن السؤال المألوف «متى ستبدأ تخفيضات الفائدة؟»، نحو قضية أكثر جوهرية: ما الذي يدفع التضخم في المرحلة المقبلة من الاقتصاد الأمريكي؟
في حين اتفق صانعو السياسة بالإجماع على ترك أسعار الفائدة دون تغيير، تكشف المحاضر عن تزايد القلق من أن الضغوط التضخمية تتغير وتتطور بدل أن تختفي. وبعيدًا عن الأجور والطلب الاستهلاكي، أشار مسؤولو الاحتياطي الفيدرالي بشكل متزايد إلى ثلاثة مصادر ناشئة لتضخم قد يكون مستمرًا: استثمارات الذكاء الاصطناعي، وارتفاع أسعار الطاقة، وارتفاع الرسوم الجمركية.

في يوليو 2026، كشفت روبنhood النقاب عن أحد أكثر التوسعات المنتج طموحًا في تاريخها خلال فعالية "The World is Flat" في لندن. أطلقت الشركة رسميًا Robinhood Chain، بلوكتشين الطبقة الثانية الخاص بها المبني على Arbitrum Orbit، إلى جانب مجموعة من المنتجات الجديدة، بما في ذلك الأسهم المرمّزة، والإقراض اللامركزي، ووكلاء تداول مدعومين بالذكاء الاصطناعي، والعقود الآجلة الدائمة، واستراتيجية توسع عالمي مُسرّعة.
للوهلة الأولى، قد يبدو الأمر وكأنه مجرد منصة مالية أخرى تطلق بلوكتشينها الخاصة. لكن عند التعمق أكثر، يتضح أن روبنhood تحاول القيام بشيء أكبر بكثير: التحول من وسيط عبر الإنترنت إلى طبقة البنية التحتية للجيل القادم من التمويل العالمي.

مقدمة
في أوائل يوليو، أثارت تقارير تفيد بأن شركة Meta كانت تقوم ببناء أعمال للحوسبة السحابية وتستعد لبيع سعة الحوسبة الخاصة بالذكاء الاصطناعي إلى عملاء خارجيين رد فعل حاد بشكل غير معتاد عبر قطاع البنية التحتية للذكاء الاصطناعي. كان رد فعل السوق غير متناظر بشكل لافت: فقد ارتفعت أسهم Meta بشكل قوي، بينما تكبدت شركات تأجير الحوسبة الخاصة بالذكاء الاصطناعي مثل CoreWeave وNebius خسائر كبيرة. وفي الوقت نفسه، تعرضت تقريبًا كل منظومة الأجهزة الخاصة بالذكاء الاصطناعي—بما في ذلك AMD وMicron وSanDisk وASML وTSMC وSamsung Electronics وSK hynix—لضغط بيع واسع النطاق. وعلى السطح، بدا الأمر وكأنه مجرد شركة تقنية أخرى تتوسع في مجال عمل جديد. غير أن الحقيقة هي أن ما كان السوق يراهن عليه لم يكن ما إذا كانت Meta تنوي تحويل موارد وحدات معالجة الرسوميات (GPU) إلى نشاط تجاري، بل ما إذا كان أحد الافتراضات الأساسية التي دعمت صناعة الذكاء الاصطناعي خلال العامين الماضيين قد بدأ في التغيير.

أمس، سيطرت بسرعة على النقاشات داخل مجتمع العملات المشفرة ومؤسسي الأسهم في الولايات المتحدة، إعلاناتٌ جديدة عن عملة مستقرة (ستابل كوين) واعدة بالاستقرار. فقد قامت أكثر من 140 شركة ومؤسسة بالاشتراك في تقديم Open USD (OUSD)، بينما انخفضت أسهم Circle (CRCL) فوراً بنحو 17.5%. وفي الوقت نفسه، أدت أحدث إعادة موازنة لمؤشر Russell إلى زيادة ضغط البيع من الصناديق السلبية. يتجاوز هذا الحدث إطلاق منتج واحد فقط. إنه يعبّر عن تسارع تكامل العملات المستقرة من أدواتٍ نشأت داخل عالم التشفير إلى البنية التحتية للدفع المالي السائدة، مع دفع السوق إلى إعادة تقييم الأثر التنافسي الحقيقي للشركات المالية العملاقة التقليدية وهي تدخل هذا المجال. فيما يلي تفصيل شامل للحدث وOUSD وتداعياته من عدة وجهات نظر.

لماذا عادت غرايسكيل لتوجيه اهتمامها نحو سولانا؟
خلال السنوات القليلة الماضية، كلمتان كانتا غالبًا ما تحددان سولانا: الأداء و"ال ميم كوينز".
باعتبارها واحدة من أبرز سلاسل الكتل (Layer 1) من دورة السوق السابقة، ارتفعت سولانا إلى دائرة الضوء بفضل إنتاجيتها العالية، وتكاليف المعاملات المنخفضة، وإنهاءها شبه الفوري للمعاملات (finality). وفي الوقت نفسه، حوّلت مشاريع النظام البيئي مثل BONK وdogwifhat (WIF) وPump.fun سولانا إلى مركز طفرة الميم كوينز. ومع ذلك، فقد طغت هذه النظرة أيضًا على تحول أعمق يجري داخل الشبكة.

في أواخر يونيو 2026، قدمت شركة Chainalysis إطارًا جديدًا يُسمى «Blockchain Tracing Ontology»، بهدف وضع طريقة أكثر توحيدًا وشفافية لوصف استخبارات البلوك تشين. وبدلًا من إطلاق منتج تحليلات آخر أو أداة استقصائية، تحاول الشركة شيئًا أكثر جوهرية بكثير: إعادة تعريف كيفية هيكلة بيانات التتبّع على البلوك تشين وتفسيرها وتوصيلها.
على الرغم من أن الإطار ما يزال في مرحلة المقترح، فقد أثار بالفعل نقاشًا مهمًا عبر صناعة الأصول الرقمية. وتكمن في جوهره مسألة بسيطة ولكنها بعيدة المدى: هل تحتاج «استخبارات البلوك تشين» إلى لغة مشتركة؟