U.S.-China Tech Rivalry Enters the 'Dual-Track Hard-Tech Era': CMC Expansion Meets AI Compute ETF Surge
Geopolitical Pressure and Capital Logic in Sync: The U.S.-China Tech Rivalry Enters the “Dual-Track Era of Hard Tech”
In mid-June 2024, global tech capital markets received a highly charged combination of signals: The U.S. Department of Defense added 37 Chinese enterprises to its “Chinese Military Companies” (CMC) List—covering critical segments including optical communications, AI chips, and high-end servers. Almost simultaneously, ProShares—the largest U.S. leveraged ETF issuer—formally filed with the SEC applications for multiple 2× leveraged ETFs targeting Chinese tech stocks, with underlying assets explicitly naming key AI-computing infrastructure players such as InnoLight, Accelink, TFN Communications, and Cambricon. A regulatory ban and a securities filing collided head-on within the same temporal-spatial frame—not merely posing a binary choice between “decoupling” or “investment,” but signaling a substantive leap into the “Dual-Track Era of Hard Tech.” Here, geopolitics erects technological firewalls through administrative authority, while international capital accelerates bets on China’s foundational growth momentum in computing infrastructure. Though seemingly contradictory, both forces converge on one undeniable reality: optical modules, domestic GPU substitution, and AI server clusters have become dual fulcrums—simultaneously central to global strategic competition and resource allocation.
U.S. List Expansion: From “Entity List” to Precision Pressure via “Military Affiliation”
Although the full list of newly added CMC entities remains undisclosed, Reuters and Bloomberg tracking reveals that several firms specializing in 800G/1.6T optical module packaging & testing, high-speed interconnect chip design, and liquid-cooled AI server integration are included. Unlike the traditional Entity List—which focuses primarily on export controls—the CMC List’s core impact lies in its investment ban: Under Section 1237 of the National Defense Authorization Act (NDAA), U.S. persons are prohibited from holding or trading securities issued by listed companies—and this prohibition applies retroactively. Consequently, not only will newly launched ETFs be barred from including these stocks, but existing QFII and Northbound funds also face compliance-driven divestment pressure. More alarmingly, the U.S. is significantly advancing the threshold for determining “military end-use”: rather than solely examining whether end-customers are military entities, it now probes deep into technical specifications—including bandwidth density, energy efficiency ratios, and anti-interference metrics—to assess potential military applicability. For instance, an optical module vendor was added to the list because its products support ultra-low-latency data return protocols used in phased-array radar systems. This “capability-as-risk” logic systematically subjects China’s civilian upgrades in AI computing infrastructure to the framework of national security review.
Leveraged ETF Filings: Capital’s Pricing Confirmation of “China’s Irreplaceable Computing Infrastructure”
In dramatic contrast to administrative bans stands ProShares’ proactive positioning. Its proposed ETFs explicitly target the “China AI Computing Value Chain,” with constituent stocks heavily concentrated in three segments: optical modules (InnoLight, Accelink), optical components (TFN Communications), and AI chips (Cambricon). Choosing a 2× leverage structure is no short-term speculation—it reflects an institutionalized response to the sector’s steep growth trajectory. IDC data shows that investment in China’s intelligent computing centers surged 68% year-on-year in 2024, with annual compound growth in optical interconnect bandwidth demand reaching 42%; Cambricon’s MLU370 AI chips have achieved over 35% penetration in training clusters at leading internet firms; and InnoLight’s 800G product yield has stabilized above 92%, with mass-production cycles compressed to just eight weeks. At its core, the leveraged ETF embodies international capital’s deep confidence in China’s structural cost advantage and delivery certainty, forged through breakthroughs across three layers: the physical layer (optical components), the protocol layer (CXL interconnect), and the compute layer (heterogeneous chips). While U.S. AI server vendors remain bottlenecked by NVIDIA GPU supply constraints, Chinese manufacturers are delivering more cost-effective computing solutions globally—powered by integrated “optical modules + domestic AI chips + liquid-cooling architecture.”
Valuation Restructuring Amid Dual-Track Tension: Policy Premium, FX Clarity, and Refinancing Constraints
Investors cannot ignore how this dual-track reality is reshaping valuation logic. First, the policy risk premium has shifted from implicit to explicit: CMC-listed firms must now absorb additional compliance costs (e.g., establishing independent supply-chain audit systems) and endure tighter financing channels (with overseas bond issuance yields rising 150–200 basis points). Second, FX impact clarity has become a critical variable—while markets previously feared RMB depreciation eroding overseas revenues, InnoLight and others disclosed that their 800G products are priced in USD but settled in RMB, rendering actual FX gains/losses manageable (Source [6]). Third, the Shanghai Stock Exchange’s Sci-Tech Innovation Board (STAR Market) is deepening its “hard-tech” orientation (Source [16]), compelling firms to standardize R&D capitalization practices; meanwhile, tightening refinancing regulations (Source [13]) are curbing models reliant solely on equity issuances to sustain high-growth expectations. Under these three constraints, valuation anchors are shifting—from “story-driven” narratives toward “cash-flow realization timing”: optical module makers must demonstrate stable gross margins amid the 800G-to-1.6T transition; AI chip firms must validate sustained order flows from large-model developers; and server vendors must prove scalable deployment of liquid-cooling solutions in data centers achieving PUE < 1.1.
Pathways to Breakthrough in Hard-Tech Sectors: Prioritizing Both Technical Depth and Financial Resilience
Faced with dual-track pressure, Chinese hard-tech enterprises are forging differentiated survival strategies. Technically, “vertical integration” has become mainstream: TFN Communications acquired an upstream ceramic ferrule manufacturer, raising its in-house optical component self-sufficiency rate to 65%; Cambricon jointly established an AI chip compatibility lab with Sugon, shortening customer onboarding cycles. Financially, “cross-border capital instrument innovation” is accelerating: some firms are piloting “green tech bonds,” linking bond coupon rates to energy-saving KPIs at AI computing centers to attract ESG investors; others are trialing “pre-sold computing capacity pledge financing,” using three-year service contracts for intelligent computing centers as credit enhancement. These initiatives reveal that true breakthrough does not lie in avoiding rivalry—but in transforming geopolitical pressure into dual forging: of technical depth and financial resilience. When optical module yield becomes the bedrock of supply-chain security; when domestic chip compatibility becomes a decisive factor in customer procurement; and when liquid-cooled server PUE data becomes a pricing determinant for financing costs—China’s hard-tech industrial chain completes its pivotal transition: from being defined by others to defining the rules.
The ultimate outcome of this contest may not be zero-sum victory or defeat—but rather a complex co-evolution of technological sovereignty and capital sovereignty. As names on sanctions lists and ETF ticker symbols are repeatedly searched on the same trading day, what we witness is not merely conflict, but the very gestation of a restructured global computing order: whoever masters light-speed interconnectivity at the physical layer, defines intelligent coordination at the protocol layer, and delivers affordable computing power at the compute layer—will wield foundational authority to shape digital civilization over the next decade.