Recently, OpenCode disclosed that an old system, which was about to be decommissioned, was breached, and the third-party model APIs hosted on it may have had their calling paths and keys quietly exposed. This meant that who was using which models, and how many automated decisions were embedded in which business scenarios, were all revealed to the attackers. What is even more distressing is not the embarrassment of being a "generation behind" in the technology stack, but rather that this system, which should have already exited the stage, exposed cracks in the trust of AI infrastructure at the edge of its handover – users entrusted computational power access and the core of their business to others. Once the backend link is hacked, the losses are often difficult to quantify in time, and as of now, the specific affected users and amounts have not been disclosed. After the incident, OpenAI joined forces with OpenCode to restore services for the affected OpenAI API customers and refund the API allocations consumed during the attack, attempting to remedy trust with compensation and response speed. However, a larger question has emerged: in the intertwined era of AI and cryptocurrency, who truly pays for "infrastructure risk"? On the other hand, Chinese Online released statistics in June of this year, showing that approximately 55% of domestic AI Token consumption was directed towards short drama and video generation, quickly simplified by public opinion to the accusation that "over half of domestic computational power is used to produce feel-good dramas." In response, Liu Xing, the public relations head of Doubao, stood up and dismantled this claim, stating that it is "somewhat unreliable" and revealing that about 10% of ByteDance's total computational power is used for video generation model training and inference. With the same group of machines and two narrative sets, the industry stands divided on whether "computational power should be spent on entertainment or research." Meanwhile, multiple headhunters have mentioned another race accelerating amid tightening hiring conditions: with a general reduction in hiring quotas, domestic giants are willing to break salary caps for top-tier fresh AI graduates in the foundational model direction, with annual salaries ranging from 2 million to 5 million yuan, pushing some newcomers to income levels higher than long-term employees, leading to "salary inversion," which has become a new sensitive term within companies. Zooming out to Seoul, South Korea's tax policy on crypto assets has already been delayed three times, with the latest postponement pushing the actual implementation time to 2027. There are also new proposals attempting to delay it further to 2029, and senior officials from the finance department are preparing to negotiate with major cryptocurrency exchanges. Yoo Young-jun, head of the Digital Finance Policy Bureau of the Financial Services Commission, is expected to personally attend these discussions. This tug-of-war between regulation and the market mirrors the battles in the AI world surrounding security incidents, data metrics, and talent costs: security incidents reshape boundaries, data narratives influence public opinion, top talent drives up costs, and every fluctuation in the tax system quietly rewrites where capital is willing to invest.
OpenCode Old System Breached: API Trust Crisis
The system that has been breached was an old system that was meant to be taken offline. OpenCode continues to emphasize in public information that it "is not the main system," trying to contain the fire to a corner of the data center, but what truly incites fear is the content it hosts – the third-party model APIs accessed by users. For many developers, this not only includes calling credentials but also implicit portraits of calling habits, traffic scales, and even business structures, such that when a “possible leak” occurs, it means others may have the opportunity to understand your model dependencies and traffic lifelines. More troubling is that OpenCode has not disclosed the specific number of affected users and the amount of losses, and this information vacuum itself amplifies panic, prompting even those not affected to start questioning: where exactly is their API hosted?
OpenAI's involvement has become one of the few "certainties" in this crisis. After the incident was disclosed, it quickly collaborated with OpenCode to help the affected OpenAI API customers restore service and promised to refund the API allocations consumed during the attack – without disclosing the total amount or specifying how many accounts were involved, but the stance is very clear: even if the technical responsibility does not fully lie on its side, it will first recover the customers' losses. The problem is, in a world where numerous AI applications and some cryptocurrency-related tools heavily rely on hosted model APIs and third-party platforms, a seemingly “side-system” security incident can easily be interpreted by the market as a fragile sample of the entire supply chain: today it is OpenCode's old system; will the next time be a certain node host, wallet interface, or another model platform? Once such doubts take root in the minds of developers and financing parties, what is genuinely eroded is not just a segment of code or a certain account balance, but the entire trust threshold of AI and cryptocurrency ecosystems towards this foundational infrastructure of hosted APIs.
“Feel-Good Dramas Consume Computational Power?” Data Metric Battle
Once the trust in hosted APIs is breached, the market naturally shifts its gaze toward another foundational clue: where is computational power actually being used? The set of data disclosed by Chinese Online in June of this year was originally just a breakdown of the usage of “domestic AI Token consumption” – approximately 55% was used for short drama and video generation, about 24% for e-commerce advertising, and about 15% for research and development. However, along the dissemination chain, the "Token consumption structure" was gradually simplified into a slogan in the ensuing verbal battles, ultimately leaving behind a prominent tag: "over half of domestic computational power is used to produce feel-good dramas." Other scenarios were reduced to background noise, and "55%" transformed from a statistical result into ammunition for emotional narratives.
The backlash was equally swift. Liu Xing, the head of public relations at Doubao, directly called out in public that the claim that “over half of computational power is used to produce feel-good dramas” is "somewhat unreliable," and presented a comparative data set from within ByteDance: only about 10% of the company's total computational power is used for video generation model training and inference. The issue isn’t merely “who is telling the truth,” but “what they each calculate.” Chinese Online talks about the distribution of AI Token consumption, while Liu Xing discusses the allocation ratio of the company’s overall computational power; the former is measured by Token consumption, the latter describes based on computational power pool allocation. The same industry is placed in different denominators, leading the media and markets to see completely different worlds: under one metric, feel-good dramas consumed half of the territory; under another, video is merely a relatively limited attempt. For capital and regulators, this metric dispute itself is a signal – in heavy asset industries like AI and cryptocurrency, those willing and able to present a clear and consistent statistical account have a better opportunity to assert the initiative in resource allocation and rule negotiations.
New Recruits with 5 Million Yuan Salaries: AI Talent War Burns Through Salary Tables
While the outside world is still arguing over “where the computational power went,” what has silently been rewritten within the companies is the labor cost table. Descriptions from multiple headhunters to the media show a high degree of consistency: overall hiring quotas are tightening, ordinary positions are being delayed, but as long as it is top-tier AI talent, especially fresh graduates in the foundational model direction, the existing salary framework can be discarded at any time. Some graduates enter the workforce and immediately have their annual salary packages raised to about 2 million yuan; reports suggest that ByteDance’s Seed project and a few top positions at Xiaohongshu have had their highest tiers pulled directly into the 4 to 5 million yuan range, while Baidu places its top configuration between 2 million and 3 million yuan. The direct consequence of these figures is what headhunters describe as “salary inversion”: in the same group, newly graduated recruits already earn more than older employees with higher job ranks who have worked on projects for years, the originally stacked salary gradients based on seniority and rank have been suddenly punctured by AI positions creating a towering peak.
This inversion is not simply a “favoritism effect,” but a panic-driven rush by companies under the combined pressures of computational power and time: foundational model R&D is highly tied to computational investments, and those willing and able to pay high premiums for a few individuals can burn expensive clusters into visible technological moats sooner. The chain reaction is also quietly developing on the other side – upstream AI infrastructure providers face clients who have already paid massive costs for top teams, making it harder for them to easily cut costs or shut down, thereby increasing their reliance on stable supply and service negotiations. Meanwhile, on the entrepreneurial and AI + cryptocurrency project side, capital and founders are forced to reassess the combination costs of labor and computational power, as many teams that expected to “exchange equity for stories” will discover, upon seeing recruits with 5 million yuan salaries, that the market price for talent has already been marked by the large companies, and this elevated starting point is rewriting who is qualified to tell the next round of AI narratives.
South Korea Discusses Crypto Tax Delay Again: Tug-of-War Between Regulation and Market
As major companies redefine the threshold of “who is qualified to tell AI stories” with salary tags, the regulatory bodies in South Korea are repeatedly rewriting the time frame for “when cryptocurrency taxes will begin.” The plan for taxing cryptocurrency assets has already been delayed three times, with the latest postponement moving the nominal effective date to 2027. This back-and-forth indicates that even a government known for administrative efficiency continually clashes with market sentiment, technological cycles, and compliance costs in real execution. Recently, there have been proposals to further delay the implementation of crypto taxes until 2029, which is still in the legislative process, publicly admitting that the regulatory body is still assessing how much tax burden the market can bear, especially at a stage where AI and cryptocurrency assets are rapidly converging and business model changes occur faster than regulations.
What’s even more intriguing is that for an additional “look” over the next two years, senior officials from South Korea’s finance department are planning to bring major cryptocurrency exchanges to the table to discuss crypto taxes and a package of regulatory issues, with Yoo Young-jun, head of the Digital Finance Policy Bureau, expected to personally attend. For enterprises and investors laying the groundwork for AI + cryptocurrency, this dual approach of scheduling to 2027, or even possibly delaying to 2029 while also initiating high-level negotiations, implies that the tax system is no longer just a “rule” that will descend one day in the future, but a continuously fluctuating variable. It not only influences whether projects will keep their core computational power and assets locally or migrate to other jurisdictions but also forces funds and entrepreneurial teams estimating returns to include the “policy discount rate” as an implicit parameter in their models, recalculating who has the stamina to stand until the end in this long-term game crossing AI, security, and cryptocurrency tax.
Common Threads of Security Vulnerabilities, Computational Power Controversies, and Taxation Games
From the breach of a soon-to-be-decommissioned old system and the potential leak of hosted model APIs, to the back-and-forth tugging of the narrative surrounding “over half of domestic computational power is used for feel-good dramas” driven by data from Chinese Online and the doubts raised by Doubao's Liu Xing, and onto fresh graduates with annual salaries from 2 million to 5 million yuan, older employees faced with salary inversion, and South Korea's crypto tax being postponed three times with a suggestion to delay it further to 2029 – these seemingly unrelated pieces of news actually revolve around two main threads: trust and resource allocation. The OpenCode intrusion incident reminds all hosting platforms that as long as there is an old system without security patches, the trust of the entire hosting chain will collapse; the refund of the API allocations consumed during the attack to affected OpenAI API customers by OpenAI and OpenCode is essentially providing the market with an expected repair path through a “safety compensation mechanism.” After Chinese Online disclosed the AI Token consumption structure, it was simplified by public opinion to “feel-good-drama computational power accounts for half the territory,” while Doubao countered with “only about 10% of total computational power is used for video generation.” This misalignment of statistical metrics and narrative choices directly undermines the external confidence in the actual flow of computational power and Tokens. Meanwhile, despite the overall tightening of recruitment in large companies, they are still willing to offer annual salaries of 2 million to 5 million yuan for fresh graduates in foundational models, even surpassing long-tenured higher-ranking employees, indicating that resources are concentrating at a very high premium towards a select few top teams and key positions. Furthermore, the repeated postponement of South Korea's crypto tax and the need for regulatory bodies to negotiate with exchange executives turn the tax system itself into a variable that affects where capital and computational power land. Both AI and cryptocurrency are high-investment, high-uncertainty industries. In a phase of frequent security incidents, chaotic data metrics, skyrocketing talent prices, and fluctuating tax burdens, the opportunities for their convergence and systemic risks will only coexist in the long term. The signals worth watching in the future include whether security compensation mechanisms will be established as an industry norm like the refund of API allocations this time, whether more institutions will publicly disclose the structure of computational power and Token consumption with increasingly unified metrics, whether top AI talent will continue to flood to a few large companies or begin to redistribute across various institutions and regions, and when and in what form cryptocurrency taxes in critical markets like South Korea will truly land. Only when these four threads gradually clarify will the trust foundation and resource map surrounding AI and cryptocurrency have the potential to stabilize.
Join our community to discuss together and grow stronger!
AiCoin exclusive Hyperliquid benefits: https://app.hyperliquid.xyz/join/AICOIN88
AiCoin exclusive Aster benefits: https://www.asterdex.com/zh-CN/referral/9C50e2
On-chain Telegram community: https://t.me/AiCoinWhaleData
On-chain community: https://www.aicoin.com/link/chat?cid=N6OVMor5g
AiCoin on-chain Twitter: https://x.com/aicoinwhaledata
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[email protected]，本平台相关工作人员将会进行核查。
