1. H200 Rental Price Surges 56% in 3 Trading Days, Surpassing B200 — China’s AI Self-Sufficiency Strategy Unchanged Despite Purchase Approval for 10 Chinese Firms
• Core Source
“H200 rental prices surged 56% in just 3 trading days, surpassing the price of the B200”
“According to media reports, the United States has approved the sale of NVIDIA H200 chips to 10 Chinese companies including Alibaba, Tencent, and ByteDance, with each company allowed to purchase up to 75,000 H200 chips”
“The total allowance of 750,000 H200 units is in line with industry expectations, and since 750,000 chips are insufficient to meet the strong demand from scaling, it is unlikely that Chinese CSPs will abandon their aggressive AI capex plans, and all state-funded data centers will still purchase only domestically produced AI accelerators”
• Expected Impact
The crux of this issue lies in two contradictory signals emerging simultaneously. The U.S. approval for 10 Chinese companies to purchase H200 chips appears on the surface to provide breathing room for China’s AI compute capacity expansion. However, as Citi’s analysis makes clear, the total allowance of 750,000 units is structurally insufficient to meet the explosive AI capex demand from Chinese Big Tech.
It is precisely this perception of supply shortage that directly caused H200 rental prices to surge 56% in just three trading days, surpassing even the rental price of the next-generation B200. The phenomenon of an older product’s price exceeding that of its successor in an environment where demand overwhelms supply starkly illustrates how severe the GPU shortage has become. In particular, since data centers funded by Chinese state capital are required by policy to purchase only domestically produced chips, demand for Chinese AI accelerators such as Huawei’s Ascend will continue to expand regardless of whether H200 purchases are approved.
Consequently, this approval measure is simultaneously acting in two directions: triggering a short-term rental price spike in the GPU market while accelerating China’s semiconductor self-sufficiency efforts, rather than meaningfully alleviating the global GPU supply-demand imbalance. For NVIDIA, there is some hope of partial recovery of Chinese revenue, but structurally, it also buys China time to build its own AI chip ecosystem. From an investor perspective, it is worth noting the short-term profitability improvement of GPU rental and cloud infrastructure operators, alongside the structural growth potential of Huawei Ascend’s supply chain and China’s AI infrastructure value chain.
2. Enterprise AI Cost Crisis — Anthropic’s Shift to Usage-Based Pricing Doubles or Triples Corporate Costs, Observability Market Emerges as New Beneficiary
• Core Source
“As Anthropic transitions its enterprise Claude pricing from a flat monthly fee to a usage-based model, companies with high AI usage may see their cost burden increase by 2–3x”
“ServiceNow and Uber mentioned that they exhausted their annual Anthropic budgets within just a few months, and cases have emerged where AI agent malfunctions caused excessive token usage in a single day”
“In Datadog’s Q1 results, growth in large customers and demand for AI-related monitoring had a positive impact”
“Datadog is expanding into areas such as LLM observability, AI agent monitoring, token analytics, and inference visibility”
• Expected Impact
Anthropic’s transition from flat-rate to usage-based pricing is a structural change long anticipated in the enterprise AI market. In an environment where AI agents perform complex computations repeatedly and token consumption grows exponentially, unlimited flat-rate pricing was an unsustainable model for AI service providers. The fact that OpenAI and GitHub Copilot are moving in a similar direction confirms this is an industry-wide trend.
The biggest problem for enterprises is cost unpredictability. Cases like ServiceNow and Uber exhausting their annual budgets within months, or incidents where AI agent malfunctions generate excessive costs in a single day, demonstrate that as AI adoption deepens, cost control is emerging as a critical new management challenge. The root cause of this problem is that Anthropic does not provide sufficiently granular usage telemetry data, making it difficult for companies to identify which workflows are driving token spikes.
The observability solutions market is filling this gap. The fact that Datadog has already confirmed tangible benefits in its Q1 results by expanding into LLM monitoring, AI agent tracking, and token analytics proves the real growth of this market. As AI evolves from a simple software feature into an independent operational subject requiring monitoring, control, and optimization, the importance and scale of the infrastructure layer serving this function will grow structurally in proportion to the expansion of AI usage.
3. AI Optical Communications Demand Explodes, Datacom Market to Triple to $47 Billion by 2028 — Chinese Optical Players Fully Booked Through 2028, 1.6T Emerges as New Standard
• Core Source
“The overall market is estimated to expand from $19 billion in 2025 to $47 billion in 2028 at a CAGR of approximately 35%, with growth led by the 1.6T data rate, followed by 800G”
“1.6T is projected to expand to $21 billion at a CAGR of approximately 200%, while 800G is forecast to reach $21 billion at a CAGR of approximately 26%”
“Driven by AI data center and global compute infrastructure investment expansion, China’s optical communications industry has entered a super-boom phase. In particular, demand for 800G and 1.6T optical modules, optical fiber, and optical cable has surged, with orders at major Chinese manufacturers already booked through 2028”
“The XPU optical component attach rate continues to rise, from over 2.5x in 2023 to 4.0x in 2025, approaching 4.5x in 2027, with approximately 6.0x per GPU in NVIDIA deployment environments”
• Expected Impact
AI data centers require a fundamentally different network architecture from conventional cloud servers. In AI clusters where tens of thousands of GPUs must be connected with ultra-low latency, optical communications — transmitting data as light rather than electrical signals — is the only practical solution. The figure showing that the optical component attach rate per GPU is rising from 2.5x in 2023 to 4.5x by 2027, and reaching as high as 6.0x per GPU in NVIDIA environments, reveals that AI investment growth drives optical communications demand in a nonlinear, amplified fashion.
As a result, the datacom optical communications market is projected to expand from $19 billion in 2025 to $47 billion in 2028 — roughly 2.5x growth in just three years. In particular, 1.6T optical modules are expected to grow at an explosive 200% CAGR, while lower-spec products such as 400G and below are projected to shrink during the same period. The center of gravity in the market is shifting entirely.
The fact that Chinese optical communications manufacturers already have their order books filled through 2028 signals that a structural bottleneck has formed where supply cannot keep pace with demand. It also creates an ironic situation where U.S. Big Tech’s AI capex flows directly into Chinese optical communications manufacturers’ revenues. The entire value chain — optical modules, optical chips, optical fiber, and optical materials — is in a synchronized boom, and both U.S. optical component companies such as Coherent and Lumentum, and Chinese optical communications firms, are jointly benefiting from this demand.
4. CXMT: From “Cash Incinerator” to “Money Printer” — Q1 Net Income +1,688% YoY, Global DRAM Market Share Reaches 7.67%
• Core Source
“China’s largest memory semiconductor company, ChangXin Memory Technologies (CXMT), logged a 1,688% surge in net profit for the January–March quarter compared to a year earlier”
“Revenue surged more than 719% year-on-year, driven by the global AI boom and memory semiconductor supply shortages”
“Q1 revenue was 50.8 billion yuan, +719.13% YoY. Net income was 33.012 billion yuan, +1,268.45% YoY. The company guided for H1 revenue of 110–120 billion yuan and net income of 50–57 billion yuan. The company’s global DRAM market share has already risen to 7.67%”
“There are views in the industry that the structure centered on Samsung Electronics, SK Hynix, and Micron is gradually beginning to crack”
• Expected Impact
CXMT’s earnings reversal is not simply a story about an individual company. The fact that CXMT — long dismissed as a “cash incinerator” — has posted Q1 revenue of +719% YoY and net income of +1,688% YoY by riding the AI and memory supercycle is the first time China’s semiconductor self-sufficiency has been demonstrated in concrete numbers to have moved beyond catch-up into the realm of genuine competition. This represents the first data point showing cracks forming in the Samsung-SK Hynix-Micron DRAM oligopoly.
Two forces are driving this. First, amid a global DDR5 and server memory shortage, Chinese companies unable to use Samsung, SK Hynix, or Micron due to U.S. sanctions have shifted en masse to CXMT. Second, CXMT’s own yield improvements and transition toward higher-value products have aligned favorably, enabling the company to secure a meaningful 7.67% global market share.
From an investor perspective, this requires a nuanced reading. In the short term, domestic Chinese demand from the de-Americanization of supply chains will continue to propel CXMT’s results. However, since the current explosive profitability is heavily dependent on surging memory prices, there are risks tied to post-peak cycle volatility. For the incumbent trio, CXMT’s encroachment on the Chinese market could act as a long-term headwind to pricing power — a competitive issue that warrants monitoring.
5. Figure AI Humanoid Beats Human in Live Packaging Competition — 2.79 Sec vs. 2.83 Sec: The Robot That Never Stops Is Transforming the Warehouse Floor
• Core Source
“Human: 2.83 seconds/package” “Robot: 2.79 seconds/package” “(The human took lunch and bathroom breaks during the competition)”
“Demonstrated productivity comparable to humans in repetitive tasks in actual industrial settings”
“The key is not speed but sustained work capability”
“Logistics, picking, and packaging tasks, while appearing simple, carry significant upper-body fatigue and musculoskeletal burden”
• Expected Impact
The true significance of this live-streamed competition is not the numbers themselves. More important than the robot’s 2.79 seconds beating the human’s 2.83 seconds is the fact that the human needed lunch and bathroom breaks — the robot did not stop. In industrial settings, what determines productivity in repetitive tasks is not peak speed, but the ability to sustain that same speed for 8 or 24 hours without interruption.
This demonstration is the first occasion on which a humanoid robot has publicly and quantitatively proven — in real time and against a live human opponent — that it has crossed the practical threshold for commercial deployment in logistics, picking, and packaging environments. Until now, humanoid technology demonstrations were largely confined to showcases in controlled settings, but a live-streamed real-time comparison against a human carries a fundamentally different level of credibility. The debate around the economics of robot adoption in logistics and manufacturing can now shift from “is it possible?” to “how quickly will it scale?”
The beneficiary structure is clear. In the near term, humanoid developers like Figure AI stand to gain; in the medium term, suppliers of critical components such as actuators, sensors, and control software; and in the long term, logistics automation platform operators deploying these robots at scale. Mitsubishi Research Institute projects that humanoids will be present in 10% of Japanese households by 2050, with prices expected to fall from the current approximately 18 million yen to between 1 and 5 million yen — suggesting that a path to mass-market adoption is becoming visible.
6. SpaceX IPO: The Largest Public Offering in History Takes the Stage — June 12 Nasdaq Listing, $75 Billion Target Raise, Valuation Up to $2 Trillion
• Core Source
“SpaceX plans to raise approximately $75 billion through this IPO, with a target valuation of $1.7 to $2 trillion”
“If SpaceX succeeds in raising $75 billion as targeted, this would represent approximately 2.5 times the scale of the previous record holder, Saudi Arabia’s Aramco”
“SpaceX IPO valuation of $1.75 trillion or more (EV/Sales of 58x), serving as a rerating catalyst for the entire space sector through index inclusion and space-themed ETF fund flows”
“In the upside scenario, SpaceX’s space data center would achieve a 10-year cumulative cost of $27 billion, representing a decisive cost advantage over ground-based alternatives ($45–90 billion)”
“SpaceX’s IPO allocation for retail investors is reported to be 30% of the public offering. Considering that traditionally only approximately 10%–15% has been allocated to retail investors, retail access to SpaceX has been significantly expanded”
• Expected Impact
The scale of the SpaceX IPO is without precedent. A $75 billion raise represents 2.5 times the size of the Saudi Aramco IPO, and an EV/Sales multiple of 58x cannot be explained by the standards of a conventional manufacturing or space company. The premium the market is ascribing to SpaceX is not about its launch business — it is a forward-looking valuation of the entire future business portfolio, including Starlink, space data centers, and the Golden Dome defense program.
The space data center thesis is backed by specific numbers. While ground-based data centers carry a 10-year cumulative cost of $45–90 billion, space-based alternatives are projected to come in at approximately $27 billion — less than half. With ground-based data centers facing a triple bottleneck of power, permitting, and cost, space emerging as a structural alternative signals the next chapter of AI infrastructure investment.
The decision to raise the retail investor allocation from the traditional 10–15% to 30% is also noteworthy. This signals that Musk has designed this IPO as a broad-based public participation event rather than an exclusive institutional affair. Combined with the strategy of leveraging the Nasdaq Fast Entry criteria for early index inclusion, a large influx of passive capital immediately following the listing is anticipated. The projection that SpaceX will enter the global top 7 by market capitalization upon listing, surpassing Tesla, is poised to serve as a catalyst for a broad revaluation of the entire space sector.
요약되어 보기좋습니다~~!!