On July 2, 2026, several originally separate leads converged at the same time: SoftBank announced the establishment of SB Neo Inc. in the United States, planning to operate the Neocloud business, laying out approximately 10 gigawatts of AI infrastructure, extending the global AI layout that it developed through Arm and others; on the same day, NVIDIA open-sourced the Nemotron-Labs-TwoTower discrete text diffusion architecture, accelerating text generation by about 2.4 times with a dual-tower parallel model of approximately 30B parameters, using algorithmic efficiency to hedge against computing power costs. The other end of the story, however, was cold—pre-market trading saw AXT Inc. down about 5%, Applied Optoelectronics and Fabrinet down over 4%, Micron Technology and Corning down over 3%, SanDisk down over 4%, Western Digital and Micron down over 3%, Seagate Technology down nearly 3%, and the optical communication and storage sectors continued to decline following the prior day's adjustment, as the secondary market tightened its valuation expectations for the hardware chain needed to "move sand and water" for AI infrastructure. The U.S. non-farm employment data for June, due to be released at 20:30, was expected to show only around 110,000 new jobs, in addition to the slowdown signs already revealed by the "little non-farm" reports, and the U.S. warning to Iran regarding the Strait of Hormuz—this globally critical oil passage is seen as the first test of Iran's commitment to fulfill its obligations. Energy and inflation uncertainties quickly surged, and under such macroeconomic and geopolitical conditions, one side saw a surge in AI computing power investment driven by 10 gigawatts of digital ambition, while the other side witnessed a continuous adjustment in optical communication and storage stock prices. The market began to hesitate: can this new round of AI infrastructure expansion deliver the returns it has been entrusted with amid potential economic cooling and cyclical fluctuations?
SoftBank's 10 Gigawatt Cloud Plan: A Gamble on AI Computing Power
On the same day that the optical communication and storage sectors faced pressure, SoftBank chose to push its chips toward a farther end. On July 2, 2026, it announced the establishment of SB Neo Inc. in the United States, specifically to operate a business called "Neocloud": this company is planned to be officially founded in July 2026, with SoftBank holding 51% and SoftBank Group holding 49%, aiming to build approximately 10 gigawatts of AI infrastructure in the U.S. and expected to launch related services within SoftBank's fiscal year 2027 according to a single source. The 10-gigawatt level is an extremely high-capacity data center computing resource arrangement on a global scale, yet the time window is compressed into a mere few years. As SoftBank begins to see a cooling in the secondary market for AI hardware, it publicly reveals a classic "counter-cyclical expansion" strategy.
This is not an isolated rush. SoftBank has long embedded itself deeply in the AI chip ecosystem through companies like Arm, and now extends downstream to Neocloud, attempting to tie upstream architecture, chip capability, and downstream cloud computing power into a single chain, seeking to transform from being the one who designs instruction sets into the one who controls the gateway to computing power. This ambition for integration from chip to cloud forms a closed loop on paper: Arm outputs technology standards, SB Neo Inc. undertakes computing power facilities, Neocloud provides services externally, and SoftBank stands at the central position of the entire value chain. However, the real challenge lies in whether such a substantial investment in computing power can achieve sufficient utilization and cash returns within the anticipated timeframe, or if it will be forced to endure an expensive test of patience amid cyclical fluctuations, as market sentiment begins to become cautious, hardware stock prices adjust, and macro and geopolitical risks compound.
TwoTower Open Source Acceleration: How Algorithms Reshape Hardware Demand
Almost simultaneously with SoftBank's physical construction of 10 gigawatts of data centers, NVIDIA delivered a markedly different response on the algorithmic level. The Nemotron-Labs-TwoTower employs a discrete text diffusion architecture, splitting capabilities that were originally squeezed onto the same path into two "towers": one focused on "one-way context understanding," responsible for interpreting and unfolding long text along the time axis; the other undertakes "bi-directional parallel error correction," synchronously repairing and correcting outputs from both front and back directions within an already unfolded context. The traditional problem with parallel generation is that once outputs are broken into multiple segments for simultaneous generation, the model's grasp of the overall semantics can become blunt, resulting in the paradox of "increasing speed while degrading cognition." TwoTower resolves this through decoupling, allowing understanding and error correction to perform their respective roles and reportedly enhancing text generation speeds to approximately 2.4 times at a scale of around 30B parameters without sacrificing generation quality.
This 2.4 times is not just a string of impressive numbers, but rather repositions a cost structure table in front of computing power investors. With the same chip and the same data center, if the throughput enhancement of the algorithmic layer allows two times the tasks to be run without increasing power consumption, then the marginal value of adding a watt of electricity or a piece of high-energy hardware will be reassessed. NVIDIA's decision to open source the TwoTower architecture, while not yet disclosing specific licensing and training cost details, means that infrastructure builders like SoftBank, when selecting a technology route for SB Neo Inc., must no longer simply bet on "how much hardware to buy," but must assess how much similar efficiency dividends algorithms may release in the coming years, as each instance of such lossless acceleration will rewrite their expectations for Neocloud's utilization rates, depreciation cycles, and return rhythms on the balance sheet.
Optical Communication and Storage Stocks Decline: AI Concept Stocks Preemptively Cooling Off
While SoftBank discusses "10-gigawatt AI infrastructure" in its press release, another screen in pre-market trading quietly discounts this grand narrative. In the optical communication sector, AXT Inc. fell about 5% in pre-market, Applied Optoelectronics and Fabrinet down over 4%, Micron Technology and Corning down over 3%; on the storage chain, SanDisk fell over 4%, Western Digital and Micron down over 3%, Seagate Technology down nearly 3%. This familiar list of "AI hardware beneficiaries" did not suddenly turn around but extended the prior day's adjustment, as the calls for computing power expansion grew louder while stock prices collectively chose to take a step back.
Current public information has not provided specific negative events corresponding to this round of declines; financial reports, orders, or regulatory factors have not pointed clearly in that direction, but the prices themselves have begun to release an emotion: the market is starting to seriously question whether this "upward curve of AI hardware," written into all PPTs, can indeed pay off as expected in terms of revenue and profits on financial statements. The high-end components from optical communication manufacturers and high-spec products from storage companies have almost automatically been categorized into the "computing power dividend" basket over the past few months, but now they are collectively under pressure in pre-market trading, in stark contrast to giants like SoftBank that are still increasing infrastructure investments. On one end, entities like SB Neo Inc. are preparing to expand data centers, electricity, and networks; on the other, publicly listed companies providing optical modules and storage are preemptively entering defense mode, signaling a significant risk for the current AI computing theme characterized by disparate primary funding increases and a cooling secondary market.
Non-Farm and Hormuz Warning: Dual Pressures from Macro and Geopolitics
On the same night SoftBank announced SB Neo Inc. and its 10-gigawatt AI infrastructure plan, the U.S. June non-farm employment report was scheduled to be unveiled at 20:30, with the market uniformly expecting around 110,000 new jobs. Prior reports known as “little non-farm”—represented by ADP employment data—had already shown a significant slowdown, exceeding market expectations of "the economy cooling gradually from high levels," and rapidly igniting discussions on the future path of interest rate cuts: a potential decline in rates theoretically benefits long-duration tech assets, but any surprise in non-farm data could instantly alter this narrative. More subtly, at the point where real data has yet to be released, investors must manage risk between expectations and the unknown, making high-beta and high-energy AI-related sectors the prime candidates for pre-market position reductions and exposure control.
Alongside this "pending" employment report is the rising geopolitical tension in the Strait of Hormuz. As one of the world's most crucial oil transport passages, this narrow waterway bears the fragile balance of market expectations for oil prices and inflation. The U.S. openly warned Iran not to alter the status quo in the Strait of Hormuz and views this as the first test of Iran fulfilling its commitments under the agreement, essentially dropping a geopolitical bomb that has yet to detonate in the energy market. For AI infrastructure frequently planning for 10-gigawatt data centers, the oil prices and inflation expectations behind electricity costs are no longer abstract variables but hard constraints that directly determine project cash flow discounting and capital expenditure return rates: one side considers potential incoming low rates, heightening valuation sensitivity for future growth stories, while the other confronts potentially rising energy costs under Hormuz risks, compressing profit margins. Under the overlay of macro and geopolitical factors, the market begins to view AI computing power as a heavy asset gamble that needs revaluation rather than merely a one-way growth theme driven by sentiment.
Is It a Bubble or a New Cycle: AI Infrastructure Navigating Uncertainty
From SoftBank’s $10-billion investment in U.S. computing power via SB Neo, aiming to launch Neocloud services around fiscal year 2027, to NVIDIA’s open-sourcing of the TwoTower architecture, achieving about 2.4 times acceleration in text generation without increasing parameter size, and the continuous decline in optical communication and storage sector stocks in pre-market on the same day, these seemingly disconnected leads actually outline multiple signals of AI infrastructure transitioning from "story" to "arithmetic": one end displays optimism about future computing power demand and efficiency dividends, while the other end quickly re-evaluates current hardware demand, profit certainty, and valuation levels. In an environment where non-farm data awaits disclosure and the Strait of Hormuz situation continues to evolve, the risk premium for high energy, high capital expenditure projects is being continually reassessed, causing capital to oscillate between "locking in long-term computing power dividends" and "avoiding short-term pullback pressures." Moving forward, whether AI technology efficiency can continue to evolve, whether hardware manufacturers like NVIDIA can meet orders and capital expenditure commitments through performance, and whether policy and energy cost curves align with expansion will jointly determine whether this round of AI infrastructure is proven to be a bubble stacked by sentiment or evolves into a new cycle driven by technology, energy, and cash flow constraints.
Join our community to discuss and become stronger together!
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
Disclaimer: This article represents only the personal views of the author and does not represent the position and views of this platform. This article is for information sharing only and does not constitute any investment advice to anyone. Any disputes between users and authors are unrelated to this platform. If the articles or images on the webpage involve infringement, please provide relevant proof of rights and identity documents and send an email to support@aicoin.com. The relevant staff of this platform will conduct an investigation.