This article is crafted based on publicly accessible information and is intended purely for the purpose of informational exchange. It does not, in any way, constitute investment advice.

“A single day in the AI realm can feel like a decade in the human world.” As we emerge from the 2025 National Day holiday, the notion of a “hyper-accelerated society” becomes increasingly tangible. Historically, each industrial revolution has ushered in exponential social transformations, and AI is following suit. Despite our psychological readiness, the rapidity of AI's impact is startling, compressing time and accelerating societal progress at an unprecedented pace.

During this National Day break, the intensified investment in data center infrastructure and OpenAI's continued valuation surge are not merely technological advancements but a complex interplay of capital, computing power, and geopolitical rivalry. It appears that a socioeconomic manifestation of the “scaling law” in the AI era is gradually taking form.

What is increasingly unsettling is that while humans take a holiday “exile,” AI is quietly “recharging” — in the literal sense of consuming vast amounts of power. This serves as a stark reminder that, when measured against human progress, AI's advancement is exacerbating inequality. The future, in this context, seems “unimaginable.”

Below, we delve into several pivotal events that unfolded during the holiday, each acting as a strategic move on the global AI chessboard, demanding our undivided attention.

01 Computing Power Monopoly 2.0: OpenAI's Strategic Alliance with AMD

Following the “century partnership” forged between NVIDIA and Intel, OpenAI is not content to stand on the sidelines and has begun viewing AMD as a strategic alternative.

On October 6th, at its annual developer conference (DevDay), OpenAI unveiled a multi-year agreement with AMD. Under this pact, AMD will supply AI chips and grant OpenAI the option to acquire approximately 10% of its shares.

This deal, valued in the tens of billions of dollars, not only solidifies OpenAI's leadership in generative AI but also underscores the industry's insatiable appetite for computing resources.

This demand stems from the exponential growth in AI model training and inference requirements. It is estimated that training models at the GPT-5 level necessitates tens of thousands of GPUs, yet the anticipated milestones remain elusive. Post-GPT-5, an even more “desperate” (fengkuang) surge in computing power will be required to push the boundaries further. AMD's involvement aims to disrupt NVIDIA's monopoly and offer more cost-effective MI300X series chips.

Professional analyses suggest that this move will expedite OpenAI's “Stargate” super data center project, with the ambitious goal of deploying 6 gigawatts of computing power by 2030 — equivalent to the annual electricity consumption of a mid-sized U.S. city.

The harsh reality, however, is that this exacerbates the energy crisis. Data centers already account for 2-3% of global electricity consumption, and AI's expansion could drive this figure up to 8%. The irony lies in the fact that while these tech giants proclaim AI's potential to “save the world,” they are simultaneously constructing a power-hungry behemoth, as if to say, “Green AI? Let me satiate my hunger first.”

This phenomenon touches upon the core of technological ethics. Computing power is a double-edged sword, capable of driving medical breakthroughs (such as personalized drug design) but also potentially amplifying environmental injustice — with developed nations seizing resources while developing nations watch helplessly.

02 Bubble or Value: OpenAI's Valuation Skyrockets to $500 Billion

Also on October 6th, OpenAI completed an employee share sale transaction, surpassing SpaceX with a staggering $500 billion valuation to become the world's largest startup.

This milestone was achieved on the back of its H1 revenue reaching $4.3 billion (a 16% year-over-year increase), though it also exposed high spending levels, with computing expenses in the first half of the year hitting $2.5 billion.

Logically, this reflects investors' unbridled enthusiasm for generative AI. ChatGPT boasts a user base exceeding 200 million, and potential commercialization paths (such as enterprise integration) have made it a darling of Wall Street.

Professionally, this valuation is predicated on model iterations, such as the newly launched Agent Kit toolkit, which empowers developers to build, test, and deploy AI agents. Additionally, ChatGPT has introduced instant checkout functionality and third-party application integration (such as Spotify and Zillow), transforming the chatbot into an all-encompassing platform.

However, it must be acknowledged that if innovation stalls, the bubble could burst at any moment — the spending rate has already reached a staggering 58% of revenue.

03 Infrastructure Investment Frenzy: Meta's $14 Billion Pact with CoreWeave

On the eve of the National Day holiday, Meta inked a $14.2 billion computing agreement with CoreWeave (extending through 2031), involving NVIDIA GB300 chips for AI infrastructure expansion. Meanwhile, Oracle announced the purchase of $40 billion worth of Nvidia chips for OpenAI's U.S. data centers.

These deals signify a collective bet by major tech companies on AI hardware. It is anticipated that by 2025, Amazon, Meta, Microsoft, and Google's capital expenditures will reach a staggering $320 billion, primarily earmarked for data centers.

From a professional standpoint, this reflects the transition from experimental phases to full-scale industrialization. CoreWeave, as a GPU cloud provider, focuses on efficient inference, and the agreement includes outsourcing 10.5 gigawatts of electricity to partners such as Brookfield. The underlying logic is that scale effects reign supreme — whoever commands computing power will dominate the AI landscape.

However, the “old story” persists — energy bottlenecks are emerging. AI data centers could drive up global electricity demand by a staggering 40%.

These giants resemble ravenous dinosaurs, vying for the last morsel of “silicon meat” while turning a blind eye to the looming climate “meteor.” This echoes the paradox of capitalism — innovation fuels growth but at the expense of sustainability. If we do not transition to green energy, this “dramatic shift” will extend beyond technology to precipitate an ecological crisis.

04 Model Iteration Wave: Anthropic Claude 4.5 and DeepSeek V3.2

Within a week, Anthropic unveiled Claude 4.5 (Sonnet 4.5), emphasizing long-context reasoning and tool integration. Meanwhile, DeepSeek launched V3.2, claiming to slash long-context computing costs by over 50%.

These iterations reflect the evolution of AI from general-purpose models to efficient, specialized ones. Claude 4.5 excels in coding competitions, while DeepSeek is optimized for enterprise-level applications — cost remains the linchpin.

According to product guidance, Claude 4.5's “Agent SDK” customization feature enables users to craft personalized agents, enhancing productivity. DeepSeek's efficiency gains stem from architectural optimization, making it suitable for fields such as biological computing.

To some extent, this mirrors the broader trend of the current Sino-U.S. AI competition — the United States leads in the number of models, but China continues to close the gap in terms of quality.

Of course, amidst the fierce competition, there is often a great deal of “comical reality,” as some spicy (xinla, “spicy” or “bold”) third-party comments have noted. These models resemble “straight-A students” in academia, copying each other while feigning originality.

05 AI Video New Milestone: Sora 2, Transitioning from Research Prototype to Consumer-Grade Product

Earlier, on September 30th, OpenAI officially unveiled Sora 2, a major upgrade to its text-to-video generation model, marking the transition of AI in video creation from a research prototype to a consumer-grade product.

Sora 2 enhances the physical realism and controllability of videos and, for the first time, integrates synchronous audio generation. It was launched alongside an iOS social app named “Sora,” akin to TikTok, enabling users to generate, share, and remix videos.

OpenAI has not disclosed full training details but emphasizes enhancing training data through re-captioning and drawing on research such as Step-Video-T2V and DiTraj to improve temporal consistency and motion trajectories. Early user feedback indicates that Sora 2 achieves higher accuracy in first-time generation, albeit at a slower pace.

These advancements position Sora 2 ahead of competitors such as Google Veo 3 or Runway Gen-2, offering superior capabilities in multi-shot storytelling and world state maintenance.

Some argue that Sora 2 marks AI's progress toward a general-purpose world simulator, potentially empowering robots and enabling more complex simulations.

OpenAI plans to further accelerate Sora 2's commercialization through APIs and global promotion, reshaping social media (TikTok should be “trembling” in anticipation). However, high computing costs and ethical scrutiny remain formidable challenges — “deepfakes,” privacy erosion, and copyright disputes linger. Sam Altman also self-deprecatingly joked about the copyright risks of Sora 2's video tools at the launch event, quipping, “Hopefully, we won't get sued.”

In a sense, while we are creating a utopia, we are also inadvertently constructing our own Pandora's box.

Above all, the frenzy of capital, model advancements, and infrastructure expansion intertwine to form a complex network. The process of AI's concrete and steel progression accelerates even during our National Day holiday. One might even argue that our holiday is merely a pause in its relentless acceleration.

This serves as a poignant reminder to humanity that technology cannot evolve in isolation to the point of “losing control” (shikong). It must dance in harmony with humanity. Otherwise, by the next National Day holiday, it will be society itself that undergoes a dramatic shift.