Three seemingly isolated pieces of news unveil a common hidden thread: on one side, it has been reported that Polymarket, a leading global prediction market platform that has had ongoing regulatory friction in various regions, is quietly placing its bets towards the East, having appointed a representative in Japan and set a "long-term wager" to obtain a local license and begin operations by 2030; on the other side, across the ocean in the United States, a proposed AI executive order that was in preparation has been postponed by the Trump administration, and this delay is seen as a signal of internal disagreements on “whether, when, and how to regulate AI” within a regulatory system long influenced by lobbying from large tech companies; meanwhile, in a landscape where the regulatory path remains unclear, Chinese manufacturers have chosen to respond to uncertainty with speed—Zhipu and the TileRT team jointly launched the GLM-5.1 high-speed version API, claiming an end-to-end output of about 400 tokens/s, setting a new record for the official interface speed of large models, immediately followed by a notable surge in the AI sector of Hong Kong stocks on May 22, 2026, with Zhipu rising over 21%, MINIMAX-W and NetEase-S also showing significant gains, as capital directly votes for the "speed story" with prices. One represents a prediction market trying to seize a gap in future rules in a new jurisdiction, another reflects the indecision around AI regulation in the U.S., and the third indicates that Chinese AI manufacturers are accelerating in terms of computing power and engineering; they all point to the same thing: the interplay of policy rhythm and technological acceleration is together rewriting how global risks are priced and how expectations are decomposed and traded.

Polymarket Bets on Compliance in Japan

In recent years, as a leading global prediction market platform, Polymarket has repeatedly sparred with regulatory authorities across multiple jurisdictions, with almost every bout of growth needing to navigate the gray areas of regulation. This platform, known for "short-term price bets on the future," is now choosing to bet on an extremely long regulatory journey: according to a single source, Polymarket has appointed a representative in Japan and included in its roadmap the goal of obtaining local regulatory approval and initiating local operations by 2030. Insiders indicate that Polymarket sees Japan as a major business opportunity that has yet to be fully exploited, considering it one of the few markets left that might “rewrite the rulebook” in the context of tightening global compliance windows.

The problem is that Japan is not an easy regulatory nut to crack. The local rules around gambling, financial derivatives, and related internet businesses are detailed and conservative, and the prediction market, being a blend of betting on future events and the attributes of financial contracts and online platforms, has not yet been fully integrated into existing frameworks, which implies both opportunity and uncertainty. For Polymarket to find its “legal position” in such a system, it needs to convince regulators that its type of trading is closer to an information market than to a speculative game. It is not betting on a specific U.S. election or AI policy, but rather wagering that Japan will provide a replicable regulatory paradigm for prediction markets in the next few years—if successful, this could become an important reference point for similar platforms engaging with regulation globally; if thwarted, it would mean that the challenge of gaining institutional recognition for prediction markets within the mainstream financial system is far higher than market participants are willing to admit.

Trump Team Hits the Brakes on New AI Regulations

While Polymarket seeks the next "friendly regulatory enclave" in Japan, the United States is pressing the brakes. According to a single source, an AI-related executive order that the Trump administration had been preparing has been temporarily postponed, with no new release date provided. This pausing action has directly thrown the timeline of U.S. AI regulation into an uncertain black box. Briefings point out that this delay is likely not just a technical holdup, but an outward signal of internal discord within the Trump team over whether to prioritize unleashing innovation or to first establish regulatory boundaries.

This is not the first time; AI is merely the latest scene in the old script of U.S. tech regulation. In the long run, behind every round of rule rewriting, from internet platforms to mobile advertising, there have always been the footprints of lobbying by large tech companies: some bet on trying to blur boundaries to win innovation space, while others strive to incorporate compliance costs into the rules themselves, thereby raising barriers for new entrants. The postponement of the AI executive order means that, for the short term, there are no answers to which narratives can make it into White House documents or which business models will be tacitly approved. Until the executive order is fully implemented, U.S. AI companies can only navigate compliance costs and business boundaries in the gray area, and the capital markets can only present "predictions" through price: the valuation discount of AI-related assets will be more conservative, tech stocks will oscillate around narratives of "tightening/loosening regulation," and in the prediction market, the pricing of contracts surrounding "when the regulation will be issued and how strong it will be" is becoming a mirror for assessing the emotional and expected landscape of U.S. AI regulatory paths.

GLM-5.1 Soars to 400 tok/s in Chinese AI

While regulatory matters are still being tugged, engineers have already shifted the battleground to “who runs faster.” The GLM-5.1 high-speed version API launched by Zhipu (02513.HK) and the TileRT team is introduced in this context: the official statement is that it has restructured the GPU operating schedule mechanics, refurbishing the inference pathway without sacrificing flagship model capabilities. The result is an impressive figure—an end-to-end output speed of approximately 400 tokens/s, claiming to have set a new global record for the official interface of large models. It has been positioned as a flagship product balancing performance and speed, essentially squeezing more "engineering space" between the thresholds of computing power and user experience.

What truly changes with 400 tokens/s is the intuitive experience from the product side: moving from “waiting for the model to think for a while” to “conversations almost require no waiting.” For scenarios involving long document generation, multi-turn interactions, and complex tool calls, the enhancement in end-to-end speed means users will not be interrupted by delays at every step, applications can become "thicker" and more logically complex without collapsing under the weight of experience. For manufacturers, with the same GPU resources, if a single interaction ends faster and has higher throughput, the computing cost per request can decrease, amplifying the flexibility of the business model. On May 22, 2026, the Hong Kong stock market’s AI sector surged, with Zhipu's stock price rising over 21% in a single day; this marked a concentrated response from the capital market to “price speed”: while the U.S. is still debating whether to “hit the brakes,” Chinese manufacturers are starting to tell their story with end-to-end speed records, transforming part of the Sino-U.S. AI competition into a "speed war" centered around inference acceleration, system optimization, and engineering details.

Hong Kong Stocks' AI Surge: Zhipu Ignites Emotion

On May 22, 2026, the emotion was most vividly expressed on the market: Zhipu opened high and surged, pushing its daily rise to over 21%, while MINIMAX-W shot up over 11%, and NetEase-S was also included in the "AI beneficiaries" narrative, with an increase exceeding 7%. During this same period, Zhipu and the TileRT team had just launched the GLM-5.1 high-speed version API, with the official end-to-end output speed stated as approximately 400 tokens/s, and claiming to refresh the global record for the official interface speed of large models, instantly igniting the Hong Kong AI sector’s sensitivity to technological acceleration, with capital treating this “speed record” as the anchor point for the next wave of growth stories.

In this narrative, the trading logic is remarkably simple: faster inference speed indicates lower potential calling costs and more business scenarios that can be accommodated in a unit of time, thus the market automatically extrapolates the commercial ceiling of the model. A single interface speed figure is transformed into valuation and growth expectations, pushing up the pricing of a whole array of “AI concept stocks” through sector linkage. However, the amplification of positive information by the Hong Kong AI sector also indicates that the current momentum is still more about anticipated trading rather than validated trading—technological narratives can quickly lift stock prices but cannot automatically translate into sustainable revenues and profits; ultimately, what can support these gains must revert to product implementation, payment willingness, and the rhythm of financial reports.

The Next Act of Regulatory Games and Technological Acceleration

Zooming out, the three strands of clues above actually point to the same world map: Polymarket is betting on obtaining a Japanese license before 2030, choosing to prove the value of prediction mechanisms in one of the most stringent yet transparent markets; across the ocean, the Trump administration's delay in issuing the AI executive order leaves the regulatory boundaries in this largest hub of technology and capital in the U.S. still blurry; and in the East, Zhipu and TileRT have launched the GLM-5.1 high-speed version with an end-to-end speed of about 400 tokens/s, reinforcing the narrative with the surge of the Hong Kong AI sector on May 22, 2026, putting “speed” back at the center of valuation pricing. The key to the next act lies in how these three axes will synchronize or misalign: if Japan can indeed implement rules for prediction markets in the coming years, platforms like Polymarket may have the chance to move from the gray area to a compliance model; the timing of the U.S. AI executive order and the specifics of responsibilities and exemptions it entails will directly determine the game between tech giants and startups; on the Chinese side, whether technological breakthroughs like the GLM-5.1 high-speed version can quickly convert into verifiable commercial revenue will decide if the "AI sector" is merely a short-term theme or a long-term asset class. For investors, rather than being swayed by daily ups and downs, it is better to disentangle their risk exposure along these three timelines: configuring around the timetable of regulatory documents, the cadence of corporate products, and revenue realization, rather than focusing solely on speed records and headline news. The true victor will be determined by who can find an investment path that embraces time amidst the regulatory games and technological acceleration.

Join our community to discuss and become stronger together!
Official Telegram community: https://t.me/aicoincn
AiCoin Chinese Twitter: https://x.com/AiCoinzh
OKX Welfare Group: https://aicoin.com/link/chat?cid=l61eM4owQ
Binance Welfare Group: https://aicoin.com/link/chat?cid=ynr7d1P6Z

免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。