1. TSMC Raises 2030 Global Semiconductor Market Forecast to $1.5 Trillion — AI Accelerator Wafer Demand to Surge 11x from 2022 to 2026
• Core Source
“The global semiconductor market will grow to over $1.5 trillion by 2030. This is a significant upward revision from the previous estimate of $1 trillion.”
“By 2030, AI-related semiconductors are expected to account for 55% of the total semiconductor market.”
“TSMC management projected that demand for AI accelerator wafers will increase 11 times from 2022 to 2026.”
“The company plans to grow capacity for next-generation 2nm and A16 chips at a CAGR of 70% from 2026 to 2028. CoWoS advanced packaging capacity is forecast to expand at a CAGR of more than 80% from 2022 to 2027.”
“The company has approved capital expenditures of $31.28 billion for this purpose.”
• Expected Impact
TSMC raising its 2030 semiconductor market forecast by 50%—from $1 trillion to $1.5 trillion—is not merely a numerical adjustment. It is the core player in the supply chain officially confirming that AI infrastructure investment will structurally persist over the long term.
The projection that AI and HPC will account for 55% of the entire semiconductor market by 2030 signals that the nature of the semiconductor market itself is being fundamentally restructured—away from smartphones and PCs and toward AI data centers. The figure showing AI accelerator wafer demand surging 11x from 2022 to 2026 demonstrates that the pace of demand explosion is incomparable to any previous technology cycle.
TSMC’s response is equally aggressive. Plans to break ground on nine phases of wafer fabs and advanced packaging facilities in 2026 alone, with $31.28 billion in approved capex, reflect TSMC’s own judgment that demand from AI chip customers including Nvidia will continue to outpace supply through at least 2028. The plans to grow 2nm and A16 process capacity at a CAGR of 70% and CoWoS packaging capacity at a CAGR of over 80% from 2022 to 2027 reinforce this view.
For investors, the core significance of this announcement is that TSMC has officially declared AI demand to be structural growth rather than a transient cycle. TSMC’s process, packaging, and global fab expansion are intertwined with the long-term capex plans of hyperscaler customers such as Nvidia, AMD, Google, and Microsoft—which simultaneously raises order visibility for key supply chain companies including ASML, SK Hynix, and Lam Research.
2. Cisco (CSCO) Raises Annual AI Infrastructure Order Target from $5B to $9B — Hyperscaler Orders Surge Over 100%, Stock Up +16.5% After Hours
• Core Source
“Cisco reported that AI infrastructure orders from hyperscalers reached $5.3 billion, and the company expects annual orders of $9 billion.”
“Hyperscaler orders were up more than 100% year-over-year. Even excluding AI-related orders, orders grew 19% year-over-year.”
“Cisco management raised its AI order guidance from $5 billion to $9 billion.”
“Cisco raised its FY2026 revenue guidance from the prior range of $61.2–$61.7 billion to $62.8–$63.0 billion.”
“We have historically applied a P/E multiple of 20–21x to Cisco. Given the AI tailwind confirmed in this earnings report, the P/E multiple could expand to 25x.”
• Expected Impact
Cisco had long been categorized as a “legacy networking company,” receiving little growth premium. However, this quarter’s results—where AI-related hyperscaler orders surged over 100% year-over-year and the annual order target was raised 80% from $5 billion to $9 billion—signal that Cisco’s investment narrative is fundamentally changing.
The core logic is that as AI data centers grow more sophisticated, the speed and bandwidth requirements for internal networking infrastructure grow exponentially. Ultra-high-speed Ethernet fabrics connecting tens of thousands of Nvidia GPUs, 800G-and-above optical transceivers, and CPO (Co-Packaged Optics) switches are areas where Cisco has invested heavily. The fact that even excluding AI-related orders, general orders grew 19% year-over-year shows that AI tailwinds are not confined to a handful of large contracts but are spreading broadly.
Morgan Stanley’s decision to revise its P/E target from 20–21x to 25x signals that the market is beginning to re-rate Cisco from a traditional networking company to an AI infrastructure company. The 16.5% after-hours surge reflects the speed of that recognition shift. Alongside peers Marvell and Arista Networks, resolving the “networking bottleneck” in AI data centers is poised to emerge as a core investment theme for the coming years.
3. Nebius (NBIS) Secures $27B AI Computing Contract with Meta — Revenue +684% YoY, ARR Guidance $7–9B
• Core Source
“Total group revenue was $399 million, up 684% year-over-year and 70.5% quarter-over-quarter.”
“Revenue from its core business, Nebius AI, was $390 million, representing extraordinary growth of 841% year-over-year and 82% quarter-over-quarter.”
“This contract is a 5-year deal totaling $27 billion, divided into two parts. $12 billion: a firm dedicated computing capacity contract beginning delivery in early 2027. $15 billion: optional capacity that Nebius can allocate to Meta or sell to general AI cloud customers at market prices.”
“Including a $2 billion equity investment from NVIDIA, Nebius has secured a total of $6.3 billion in capital.”
“Annual Recurring Revenue (ARR): $7–9B.”
• Expected Impact
Nebius, spun out of the cloud division of the former Yandex, was virtually unknown until recently. Yet with quarterly revenue surging 684% year-over-year and its core AI cloud business growing 841%, it has proven that a market exists where independent AI infrastructure players can survive and thrive beyond the AWS-Google-Microsoft hyperscaler triad.
Structurally more significant is the design of the $27 billion Meta contract. The $12 billion tranche is a firm dedicated computing contract beginning in early 2027, while the remaining $15 billion is structured so that Nebius can sell to other customers at higher market prices when demand is strong—and Meta has committed to purchasing the full volume as a backstop even if demand softens. This contract structure caps downside risk while keeping upside potential fully open, and additionally grants Nebius the ability to raise low-cost capital through asset-backed financing (ABF). Nvidia’s $2 billion equity investment is not merely a financial stake—it translates into priority access to next-generation Rubin chips and “NVIDIA Exemplary Cloud” status, a strategic alliance in every sense.
The ARR guidance of $7–9 billion and the target of securing over 4GW of contracted power by year-end demonstrate that in a supply-constrained AI infrastructure market, the operator that builds scale first gains pricing power as well—a real-world case study in action.
4. Alibaba Sacrifices Operating Margin to Zero in Pursuit of AI Investment — Cloud External Revenue +40%, 10x Computing Capacity Needed by 2033
• Core Source
“Alibaba management: By 2033, we will need 10 times the computing capacity we needed in 2022.”
“Alibaba executive: Due to our rapid investment in expanding computing capacity, we will exceed the originally announced RMB 380 billion.”
“Cloud external revenue growth accelerating to +40%, with a path for MaaS ARR to exceed RMB 30 billion by year-end, up from over RMB 8 billion.”
“AI-related revenue has recorded triple-digit growth for 11 consecutive quarters, with an annualized run rate of RMB 36 billion.”
“The cloud is currently supply-constrained—management disclosed there are no idle cards on servers.”
• Expected Impact
Alibaba driving its operating margin effectively to zero this quarter is not failure—it is a strategic choice. AI infrastructure investment exceeding RMB 380 billion and management’s declaration that 10x the 2022 computing capacity will be needed by 2033 mean that China’s largest cloud operator has already confirmed demand and is marshaling full resources to secure supply. The disclosure that “there are no idle cards on servers”—meaning all GPUs are already fully utilized—directly confirms that current demand is outstripping supply capacity.
AI-related revenue has grown triple digits for 11 consecutive quarters, and MaaS (Model-as-a-Service) ARR has grown 10x in six months and is now on track to exceed RMB 30 billion by year-end. Cloud external revenue growing +40% shows that the external enterprise customer base is expanding rapidly, suggesting Alibaba Cloud is positioning itself as China’s AWS rather than merely internal infrastructure.
Despite U.S. GPU export restrictions, the scale of Alibaba’s continued investment also implies that the domestic supply of Chinese semiconductors (Huawei Ascend, etc.) and ASICs is advancing in practice—a pathway for China’s semiconductor ecosystem to reduce its dependence on Nvidia over the medium to long term.
5. Nvidia H200 Export Approved for 10 Chinese Companies — China AI Market Opportunity Estimated at $50B, Previous Market Share Was 95%
• Core Source
“The U.S. Commerce Department has approved approximately 10 Chinese companies including Alibaba, Tencent, ByteDance, and JD.com to purchase Nvidia H200 chips.”
“Each approved customer may purchase up to 75,000 chips under U.S. license conditions.”
“Prior to large-scale U.S. export restrictions on semiconductor technology to China, Nvidia held a 95% share of China’s advanced semiconductor market. At that time, the Chinese market accounted for 13% of Nvidia’s revenue.”
“Nvidia estimates the Chinese AI market represents a $50 billion opportunity in 2026.”
“Demand for H200 chips in China is very strong, with estimated orders exceeding 2 million units.”
• Expected Impact
This export approval is the first concrete signal of easing in the U.S.-China semiconductor conflict. Before restrictions, Nvidia held 95% of China’s advanced semiconductor market, with China representing 13% of total Nvidia revenue. Estimated orders already exceeding 2 million units suggest that demand has grown even larger than before restrictions were imposed.
A per-customer cap of 75,000 chips applies, but the combined purchase potential of the ten approved customers alone could reach hundreds of thousands of units. With Nvidia estimating China’s AI market at a $50 billion opportunity in 2026, actual shipment resumption could push Nvidia’s revenues well above current consensus estimates.
That said, practical hurdles remain—Chinese government approval and other procedural steps mean actual shipments will take time. When exports resume in earnest, the benefits would extend beyond Nvidia to the broader supply chain: SK Hynix, which supplies the HBM essential to H200, and TSMC, which handles CoWoS packaging, would both stand to gain meaningfully.
6. SoftBank Announces 10GW Ohio AI Campus — Total OpenAI Investment to Reach $64.6B, Stargate 1.5GW Texas Data Center Under Construction
• Core Source
“In terms of power, the capacity is 10 gigawatts. The energy source is gas. The mega data centers required for AI need this enormous level of power.”
“In data center terms, 10 gigawatts exceeds the combined installed data center power capacity of the UK, Japan, and South Korea.”
“Construction of the data center in Milam County, Texas is progressing very well. That data center has a capacity of 1.5GW and is under a long-term contract with OpenAI for more than 15 years.”
“Once the investment commitments are fully completed in October of this year, our total investment in OpenAI will reach $64.6 billion. At that point, our ownership stake will be approximately 13%.”
• Expected Impact
SoftBank’s announced 10GW Ohio campus is unprecedented in scale as a single AI infrastructure project. The comparison to exceeding the combined data center power capacity of the UK, Japan, and South Korea is not simply a numerical statement—it represents a paradigm shift in resource allocation. The computing demand required to advance AI models is growing at a pace that far outstrips the imagination of conventional data center infrastructure, and SoftBank is backing that view with direct capital.
The already-in-progress Stargate 1.5GW Texas data center, locked into a 15-year-plus contract with OpenAI, demonstrates that the payback horizon and contractual structure for AI infrastructure investment have already become institutionalized. A total OpenAI investment of $64.6 billion and an approximately 13% ownership stake signal that SoftBank is claiming the role of builder of the AI infrastructure ecosystem, not merely a financial investor.
From an investor perspective, if a project of this scale is executed, the beneficiary universe extends well beyond AI semiconductors—spanning power equipment (gas turbines, transformers, switchgear), data center construction materials, cooling infrastructure, and optical communications equipment. As J.P. Morgan noted, gas turbines are already sold out through 2029, confirming that this demand is already materializing.
7. Humanoid Robots Enter Year One of Commercialization — Figure 03 Autonomously Processes 18,000 Packages in 14.5 Hours, 2030 Shipment Forecast 1.2 Million Units
• Core Source
“One unit operated for approximately 14.5 hours / sorted 18,000 packages. Two robots can cover 29 hours of work.”
“In 6 hours: 7,836 packages processed — 1,306 packages per hour / 21 packages per minute.”
“Humanoid robots moving beyond R&D into actual deployment in manufacturing and logistics. Real-world deployments at JAL Haneda Airport, BMW factories, Amazon fulfillment centers signal entry into commercialization trajectory. Agility Robot surpassed cumulative production of 10,000 units in March of this year, demonstrating China’s overwhelming manufacturing pace.”
“2026 shipment forecast of 90,000 units, with supply chain and AI models as the key competitive variables. Actuators account for 40–60% of unit cost, with China’s grip on parts supply chains acting as a structural advantage. VC investment in robotics grew 3x from 2023 to 2025, expanding to an annual scale of $40.7 billion.”
“Real factory environment data is priced at $136 per hour—tens of thousands of times more expensive than internet data—and with over 100 factories globally, there is virtually no competitor except Tesla.”
• Expected Impact
Figure 03 autonomously sorting 18,000 packages over 14.5 hours with autonomous shift handoffs between two units is not simply a demonstration. At 1,306 packages per hour, performance is now comparable to skilled logistics workers—and combined with falling unit prices as mass production ramps up, this pushes ROI into viable territory. The simultaneous appearance of real-world deployments at JAL, BMW, Amazon, and others confirms this transition is an industry-wide shift, not an experiment by one company.
Market size projections are also rapidly materializing. Shipment forecasts of 90,000 units in 2026 and 1.2 million units by 2030 imply a 13x expansion in just four years. VC robotics investment growing 3x from 2023 to 2025 to reach an annual scale of $40.7 billion shows that capital is already taking this market seriously.
On the competitive landscape, the fact that actuators account for 40–60% of unit cost and China’s parts supply chains hold structural advantages suggests that the hardware battle in robotics will be decided as much by supply chain control as by AI model performance. At the same time, the emergence of a new revenue model—where companies like Hyundai Motor Group can sell real factory training data at $136 per hour—signals that the humanoid ecosystem is opening across three simultaneous axes: hardware, software, and data.
Comment [0]