On October 21, 2026, the figures regarding computing power were abruptly brought to light: ByteDance, Alibaba, and Tencent were reported to have collectively ordered over 3 million NVIDIA RTX PRO 5500, with a total order value exceeding 35 billion dollars. In the long term, there are plans for SpaceX and Tesla's joint venture, Terafab, to mass-produce chips in the third quarter of 2028, providing approximately 1 terawatt of computing power annually. The newly established Arda also just secured 230 million dollars in financing, with a valuation of about 1.5 billion dollars. Global AI infrastructure is rapidly accumulating in a “state-owned enterprise arms race + startup financing + giant joint ventures” tripartite structure. At the same time, Griffin, CEO of Citadel Securities, publicly stated that it is almost certain we are in an AI bubble, comparing the current situation to the eve of the internet bubble burst, pressing a jarring pause button on this frenzy; in the crypto world, the fear and greed index rose from 55 yesterday to 63 today, staying in the “greed” territory for two consecutive days. On one side is the infinitely magnified imagination of computing power, while on the other side is an almost overlooked on-chain black hole.

Compliance Computing Power Frenzy Under 35 Billion Graphics Card Orders

While the on-chain black hole continues to consume assets, the arms race for computing power is being rapidly filled by several giants offline. According to a single source, ByteDance, Alibaba, and Tencent have jointly locked in over 3 million NVIDIA RTX PRO 5500, each priced at about 11,400 dollars, raising the total order amount to over 35 billion dollars. This is not a scattershot equipment purchase, but a concentrated buy completed within the U.S. export red line: the RTX PRO 5500 is explicitly described as meeting export restrictions to China and primarily being used for AI inference tasks. Against the backdrop of high-performance training chips being blocked, this "compliant computing power" is packaged as the foundational building blocks for the next generation of data centers.

For NVIDIA, this super order, with an undisclosed delivery schedule and contract details, symbolizes its continued consolidation of global AI infrastructure discourse amidst regulatory layers. For these three Chinese tech giants, this is a bet of 35 billion dollars to "pre-bury" inference capabilities for the next few years in a constrained environment. While one end is still discussing IPOs, financing, and future wafer fabrication plans, the other end is signing millions of compliance graphics card spot orders. The focus of global AI infrastructure competition is shifting from abstract visions and stories to who can first move all available and legally circulating computing power into datacenters. The order disclosed on October 21, 2026, is itself a landmark node in the deeper waters of global AI arms and regulatory games.

Arda and Firmus: AI Infrastructure Financing Frenzy

If the major internet companies locking in millions of graphics cards represent an arms race on the consumption side of computing power, then the emergence of Arda brings the lens back to the production side. Founded by the former OpenAI research director and former head of product and engineering at Palantir, Bob McGrew, this AI manufacturing startup announced the completion of 230 million dollars in financing, with a valuation of around 1.5 billion dollars, led by Altimeter Capital, Accel, and Khosla Ventures. The combination of the founders' backgrounds and leading investors itself is a "top-tier venture capital-recognized AI hardware story" showing that capital is not satisfied with betting on models or applications, but is willing to take on the uncertainties of manufacturing and infrastructure in advance, as long as the narrative is imaginative enough.

Unlike Arda's entry from the supply chain side, Firmus stands at the other end of cloud computing power. In early October this year, this Australian AI cloud service provider suddenly withdrew its IPO plan, still untested by public market valuation, and market sentiment remained in flux. Yet at almost the same phase, news emerged that NVIDIA was negotiating potential investments with it, possibly amounting to hundreds of millions of dollars, though the final transaction and specific scale remain unconfirmed. With Arda securing high-valued private financing on the manufacturing side, and Firmus still welcoming direct negotiations with chip giants after canceling its IPO on the cloud side, a capital chain focused on "future computing power" is rapidly stretching from factory to data center, providing the necessary background for discussing whether this round of AI investment heat is drifting away from real demand.

Terafab's 1 Terawatt Long-Term Bet

If Arda and Firmus represent the realistic battleground for this round of AI infrastructure in the coming years, then the answer provided by the Musk system deliberately extends the timeline—Terafab, a chip joint venture established by SpaceX and Tesla, is explicitly positioned as a computing power factory serving AI and robotics. Its focus is not on the immediate expansion of cloud services, but on how to continuously "generate electricity" for the next generation of AI and robotics.

According to official plans, Terafab has directly set its milestone for the third quarter of 2028: that is its planned node for achieving chip mass production, aiming to provide approximately 1 terawatt of computing power annually for AI and robotics. This long-term bet measured in "terawatts" is in itself a reverse statement on the current AI craze—not chasing this year's valuation rankings, but betting that the entire industry will still need higher-density dedicated computing power years down the line. However, parallel to this grand narrative is Musk's own "reserved calibration" regarding the timeline: he admits that there is a significant margin of error in this progress, and the specific investment scale, chip types, process technology, and production capacity details of Terafab have not been disclosed so far, deliberately keeping a buffer zone of uncertainty over a backdrop of optimistic expectations, making Terafab feel more like a long-term gamble rather than a well-calculated engineering order.

Griffin's Bubble Warning: Will AI Repeat the Fate of Tech Stocks?

Beyond the long-term gamble of Terafab, traditional financial capital's seasoned players have begun to issue cooler annotations on this round of AI fervor. According to event summaries, Citadel Securities CEO Ken Griffin recently stated that it is nearly certain we are in an AI bubble and compared the current situation to the phase before the internet bubble burst—valuations rising rapidly, but business models and profit paths have not been equally validated in fundamentals. This judgment does not name any specific company or asset but raises a macro risk reminder for the entire sector: the market may be paying in advance for cash flows and technological breakthroughs that have yet to materialize.

If you juxtapose this warning with the capital movements of the past few weeks, its sharpness becomes more evident: ByteDance, Alibaba, and Tencent ordered over 3 million RTX PRO 5500 at a price exceeding 35 billion dollars; Arda secured 230 million dollars in funding while still in its early stages, translating to a valuation of about 1.5 billion dollars; Firmus just withdrew its IPO, and news emerged that NVIDIA might make a significant investment. Upstream computing power and infrastructure are being priced as "cash cows" for the next twenty years, yet uncertainty has not been cleared. For the crypto market, the question arises: on top of these high-valued infrastructures, are the on-chain assets deeply bound to the AI narrative also being lifted into the bubble zone? Once upstream valuations readjust, can they escape a repricing cycle driven by the same story?

Greed Index 63 and 1.18 Billion Stolen Amidst Crypto AI Frenzy

At the same time that upstream AI infrastructure is being bid as a "future cash flow machine," sentiment in the crypto market is also clearly leaning towards risk-taking. The fear and greed index shows yesterday at 55, rising to 63 today, classified for two days in the "greed" range, with sentiment sliding from cautiously optimistic to risk-ignoring excitement. Against the backdrop of AI narratives permeating everything, this indicates that the market is more willing to pay for stories, especially those that bundle computing power, models, and on-chain assets together, rather than calculating underlying safety and risk compensation calmly.

However, the greed of sentiment is layered over a group of cold, hard safety numbers. According to Blockaid, from January to August 2026, it tracked 60 attack incidents linked to on-chain or on-chain interactions, involving approximately 1.18 billion dollars; victims have only recovered about 9.8 million dollars, with a recovery rate of less than 1%. The materials did not provide specific addresses, protocols, or methods, but this recovery rate is enough to indicate that, once the safety line on the current crypto infrastructure is breached, losses are almost irreversible. As the AI boom and capital frenzy extend into the crypto field, this reality of high frequency, high loss, and low recovery is quietly eroding the sustainability of the "crypto AI" narrative, for any new narrative built on ongoing greed must first answer a question: in a system where errors are nearly irretrievable, who will bear the final bill when the bubble is pierced?

Subsequent Observations Under the Shadow of Computing Power Arms Race and Bubble

On one end, ByteDance, Alibaba, and Tencent have locked in over 3 million RTX PRO 5500 graphics cards worth over 35 billion dollars; Arda has secured 230 million dollars in financing; and Terafab is planning long-term production capacity to reach 1 terawatt of computing power by 2028, all pointing to the acceleration of AI infrastructure investment in 2026. On the other end, Griffin's almost confirmation of "being in an AI bubble" and the less than 1% recovery rate from 60 attack incidents between January and August 2026 expose the fragile underpinnings of this expansion within on-chain data. Moving forward, what is truly important for readers to observe is not the abstract "next round of market movements," but these specific links: whether the 35 billion graphics card orders can be executed on time or adjusted; whether the hundreds of millions of dollars transaction between NVIDIA and Firmus will ultimately materialize and reshape the cloud computing narrative; whether Terafab will face delays or advancements in its mass production timeline; and how the fear and greed index, along with Blockaid's security data, will continue to reflect the market's delicate balance between greed and risk tolerance. For participants in the crypto market, only by viewing these on-chain security events and bubble risks as preconditions on the same investment chain, before embracing the "AI + crypto" narrative, can they truly understand the irreversible risk curve they are undertaking in this long interplay of computing power arms race and emotional frenzy.

Join our community, let’s discuss and grow stronger together!
Exclusive Hyperliquid benefits for AiCoin: https://app.hyperliquid.xyz/join/AICOIN88
Exclusive Aster benefits for AiCoin: 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
On-chain Twitter for AiCoin: 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。