On June 16, 2026, all three major U.S. stock indices closed higher—The Dow Jones rose 0.92% to 51,671.03, setting a new all-time high. The Nasdaq soared 3.07% to 26,683.94, and the S&P 500 gained 1.65% to 7,554.29. The Philadelphia Semiconductor Index surged 5.45% in a single day to 14,099.62, also reaching a record close.
The driving force behind this rally is clear and consistent: AI computing power.
NVIDIA climbed 3.54% to $212.45, pushing its market cap past $5.145 trillion. Micron Technology jumped 10.84%. Western Digital soared 16.1%. SpaceX, on its second day of trading, rose another 19.6%, with its market cap surpassing $2.5 trillion. While market attention remains fixed on GPU and chip giants, the foundational infrastructure supporting this boom—fiber optic networks, cooling systems, and power supply—is growing at an even steeper pace.
These three "invisible tracks" form the physical backbone of AI computing infrastructure.
Fiber Optics: The "Neural Network" of AI Clusters, with Record Supply-Demand Gaps
One easily overlooked fact: AI computing clusters consume 5 to 10 times more fiber optics than traditional data centers. A single cluster with tens of thousands of GPU cards uses as much interconnect fiber as dozens of traditional data centers combined. NVIDIA’s commercial CPO switches have completed their first deliveries, and the Spectrum series will begin mass production ramp-up in the second half of 2026—optical interconnects are shifting from "optional technology" to "essential infrastructure."
The demand-side data is even more striking. According to CRU, global data center fiber optic cable demand will grow by an astonishing 75.9% year-over-year in 2025. Looking ahead to 2026, global data center fiber demand is expected to reach 91.6 million fiber-kilometers, up 32% year-over-year. In the AI sector alone, fiber demand will exceed 100 million fiber-kilometers, a 233% increase year-over-year. By 2030, global data center fiber demand is projected to hit 128 million fiber-kilometers, with AI-related demand surpassing 80 million fiber-kilometers.
However, rigid supply constraints are creating a historic supply-demand mismatch. The expansion cycle for fiber preforms (the raw material for fiber optics) is as long as 18 to 24 months, with significant technical barriers to certification. CRU forecasts total global fiber optic cable demand at around 577 million fiber-kilometers in 2026, but effective supply will only be about 397 million fiber-kilometers, resulting in a full-year gap of 180 million fiber-kilometers—a shortage rate of roughly 16.4%, far exceeding the previous peak of 4.8% during the 2017–2018 upcycle.
Naoki Okada, President of Japanese fiber optic leader Fujikura, recently revealed that the company is receiving orders from nearly all major U.S. hyperscale cloud providers. Supply for AI data center fiber products is "extremely tight," and price increases are underway, with some customers accepting higher prices. Some high-end specialty fiber orders are already booked out to 2028. UBS estimates that global fiber demand grew at an average annual rate of just 2% over the past five years, but as AI data center construction accelerates, industry demand growth could exceed 30% in the coming years, with data center-related fiber demand potentially reaching a compound annual growth rate of over 75%.
Against this backdrop, the fiber optic cable industry is undergoing a valuation reset from the "telecom cycle" to the "AI cycle"—with scarcity premiums and rising demand driving the sector.
Cooling: From "Optional" to "Essential"—Liquid Cooling Becomes a Computing Necessity
If fiber optics are the "neural network" of AI clusters, then cooling systems are the "life support" that keeps these networks from overheating and failing.
Exponential increases in rack power consumption are fundamentally transforming the cooling industry. Next-generation HBM-equipped AI servers, physical AI industrial robots, and supercomputing clusters now routinely exceed 15kW per rack. Traditional air cooling, limited by temperature, noise, and energy consumption, can no longer meet demand. New intelligent computing centers are now required to use cold plate or immersion liquid cooling solutions.
The data speaks for itself: According to a Zhongtai Securities report, NVIDIA’s GB200 racks have already hit the critical threshold for air cooling. The upcoming Vera Rubin NVL72 model, set to ship in the second half of 2026, is expected to reach 200 to 220kW per rack—far beyond the 40–60kW/rack cooling limit of air systems. Continuing with air cooling not only results in insufficient heat dissipation but also causes local hotspots and forced throttling, directly undermining computing stability and lifespan.
At the same time, leading global cloud providers’ custom ASIC chips are also exceeding air cooling’s limits. Guosheng Securities notes that Google’s TPU V7 cards now consume over 980W each, and Meta’s TVR server racks reach a power density of 170kW. New-generation computing equipment from Google, AWS, and Microsoft all feature fully liquid-cooled architectures.
Policy is also providing strong support. According to Founder Securities, regulations require that new and upgraded large or hyperscale data centers maintain a PUE (Power Usage Effectiveness) below 1.25, with projects in national computing hubs needing to stay under 1.2. Liquid cooling can dramatically cut cooling system energy use and has become a compliance requirement for computing infrastructure.
In terms of value, GF Securities estimates that the liquid cooling systems for GB200, GB300, and Vera Rubin architectures are worth approximately $40,000, $50,000, and $80,000–$100,000 per set, respectively. Liquid cooling is evolving from an "add-on component" to "core equipment."
On June 16, liquid-cooled server stocks saw strong performance: Chunzong Technology hit its daily limit, and Shenglan shares jumped over 10%. The market is pricing in the certainty of this sector’s growth.
Power: AI’s "Fuel" Bottleneck—Widening Supply Gap for Gas Turbines
Among these three tracks, power demand from AI data centers is the largest in scale, the longest in cycle, and the most structurally challenging.
Slow grid integration is ushering in the era of "self-owned power plants" for data centers. A CITIC Construction Investment report raised its North American AI power demand forecasts for 2026–2028 by 4%, 35%, and 42%, respectively, with a three-year compound annual growth rate of 73%. The capacity gap for gas turbines needed by data centers will grow from 4.3GW in 2025 to 27.2GW by 2030.
From 2026 to 2028, the global supply gap for gas turbines is expected to be 16GW, 16GW, and 19GW, respectively. Including other power generation needs, the total global power shortfall will reach 30GW, 42GW, and 59GW over the same period. Siemens Energy secured orders for 102 gas turbines in Q1 of its 2026 fiscal year, a quarterly record. In Q1 2026, about 25% of demand came directly from data center clients. Gas turbine orders are expected to peak in 2026, as developers lock in equipment for over 60GW of new gas-fired capacity. Siemens Energy’s backlog of data center-related orders has risen from 22GW to 24GW.
Given gas turbine delivery cycles of 1–3 years, current orders effectively lock in production capacity for the next several years. Meta has announced plans to build 10 gas-fired power plants to supply its Hyperion AI megacenter, with a total installed capacity of 7.5GW. Microsoft is negotiating a $7 billion gas power project. Major tech companies are shifting from "buying electricity" to "building power plants"—a structural change that will reshape pricing power across the global power equipment supply chain.
Potential Targets: Identifying Structural Opportunities Across the Value Chain
Based on the industry logic above, the following areas merit attention:
In fiber optics, leading companies with self-sufficient preform capacity (such as Hengtong Optic-Electric) have greater pricing power and production flexibility as supply-demand gaps widen. Lumen Technologies is actively pivoting to build dedicated fiber networks for the AI economy, targeting 58 million intercity fiber miles by 2031.
In liquid cooling, NVIDIA’s Vera Rubin NVL72 racks reach 200–220kW, and the value of liquid cooling systems continues to climb. Infrastructure providers like Vertiv have launched integrated system solutions for high-density AI computing.
In power equipment, the gas turbine supply chain (Siemens Energy, General Electric, etc.) stands to benefit directly from the trend of data centers building their own power plants. The persistent supply-demand gap means equipment providers will continue to gain bargaining power.
Gate Stock Trading: One-Click Access to Core AI Infrastructure Assets
For investors looking to participate in these sectors, traditional routes come with multiple hurdles: cross-border remittance, brokerage account setup, currency exchange, and juggling multiple accounts—a complex and time-consuming process. Gate is changing the game.
On June 1, 2026, Gate officially launched real stock trading services, becoming one of the first crypto exchanges to offer direct access to the U.S. stock market within a crypto platform. As of June 2026, Gate TradFi has listed over 10,000 real stocks and ETFs, covering the NYSE, Nasdaq, NYSE Arca, NYSE American, and BATS—five major exchanges. On June 11, Gate further rolled out Hong Kong stock trading, allowing users to trade over 1,500 Hong Kong-listed stocks directly.
Gate stock trading offers three core advantages:
First, USDT settlement. Users don’t need to exchange currencies, wire funds internationally, or open additional brokerage accounts. Simply use the USDT in your Gate account to buy U.S. or Hong Kong stocks in one click. This eliminates the cumbersome process of "selling crypto → withdrawing fiat → cross-border remittance → broker funding."
Second, ultra-low entry barriers. Fractional share trading starts at just 0.01 shares, and you can invest in U.S. stocks with as little as $1. This is especially friendly for investors looking to dip their toes into the AI infrastructure sector with small amounts of capital.
Third, compliance and security. All stock trades are executed by Alpaca, a licensed U.S. broker-dealer with clearing qualifications. Real assets are independently custodied through the DTC system and are fully protected by SIPC coverage.
How do you trade stocks on Gate?
Simply update your Gate App to v8.23.5 or later, go to the TradFi section, and select "U.S. Stocks" or "Hong Kong Stocks" to browse and trade supported stocks and ETFs. The entire process takes place within the crypto-native account system—no need to leave the Gate ecosystem.
Additionally, on June 16, 2026, at 06:00 UTC, Gate’s contract stock section launched live perpetual contracts for eight tickers: TWLO, ROK, CGNX, IVV, SOXX, SMH, SPYM, and VOO, all settled in USDT. This gives investors more options to gain leveraged exposure to AI infrastructure-related assets.
Conclusion
The AI computing power race is shifting from a "front-end war" at the chip level to a "back-end battle" in infrastructure. The supply-demand gap in fiber optics, the technological leap in liquid cooling, and the capacity bottleneck in power supply—these three invisible tracks form the physical foundation of the AI era and are becoming the core assets for capital repricing.
From an industry perspective, all three sectors share the traits of rigid demand, supply-side constraints, and extended growth cycles. From a market perspective, the explosive rally in U.S. storage chip and AI-related stocks on June 16 is a clear validation of this logic.
Through its real stock trading and contract stock offerings, Gate is bridging crypto ecosystem liquidity with core assets of traditional capital markets. For investors focused on the long-term value of AI infrastructure, this provides a new channel to access global core assets without leaving the crypto ecosystem. As computing power becomes the new means of production, investing in the "infrastructure" of computing itself may offer greater compounding returns than betting on any single chip company—and fiber optics, cooling, and power are where that compounding curve begins.
