Recent analysis from Wall Street firm Wedbush Securities indicates that lithography giant ASML Holding NV (ASML.US) raising its 2027 performance guidance beyond expectations is highly favorable for the bullish outlook of the AI compute infrastructure theme, which is closely tied to the most advanced process logic chips like CPUs/GPUs and DRAM memory chips. The latest perspectives from Wedbush and other institutions, coupled with the strong earnings and optimistic future outlooks from ASML Holding NV and Taiwan Semiconductor Manufacturing (TSM), provide a significant boost to the AI compute theme, which had recently faced intense selling pressure due to overcrowded and highly leveraged long positions.
From Wedbush's optimistic stance on AI compute, to the explosive earnings and robust forecasts from key upstream capacity and supply players ASML Holding NV and Taiwan Semiconductor Manufacturing, to the latest analysis from research firm SemiAnalysis showing Anthropic's transition from long-term losses to a phase of soaring overall profitability, a crucial signal is being sent to global equity markets. The AI compute supply chain is gradually shifting from the "super-cycle of AI capital expenditure driven by training large models" to a new phase of "exponential expansion in AI inference compute demand driven by the scaled application of agents." These latest signals strongly counter the recent pessimistic narrative of "compute overcapacity," which contributed to sharp declines in the AI semiconductor sector.
Wedbush's core assessment is that ASML Holding NV's upward revision of its 2027 outlook and its consideration of further increasing 2028 EUV capacity align perfectly with the seemingly insatiable demand for advanced process logic chips (i.e., sub-5nm CPUs/GPUs/TPUs) and high-end data center DRAM, driven by AI data center construction. Increased shipments of ASML Holding NV's lithography equipment in 2027 are expected to translate into initial wafer manufacturing output by late 2027 to early 2028, particularly for high-end DRAM/HBM memory chip capacity. However, the persistently growing AI-related capital expenditures from chip manufacturing giants like Taiwan Semiconductor Manufacturing and global tech giants like Microsoft and Meta Platforms, Inc. (META) mean it remains uncertain when supply will truly catch up with demand.
"We believe that ASML Holding NV management's unexpected increase in their 2027 outlook, along with the potential for a significant rise in 2028 EUV lithography equipment output, aligns closely with the robust demand growth pace we are observing for data center-focused advanced process logic chips and high-end DRAM, aimed at supporting the ongoing global AI compute infrastructure build-out," wrote Wedbush senior analyst Matt Bryson in a recent note to clients.
"The timeline for increased shipments also aligns well with our baseline view that the rise in AI compute-related capital expenditure in 2027 will ultimately drive increased production of advanced process logic chips and DRAM, especially HBM/DRAM memory chips, from late 2027 to early 2028. However, given that capital expenditure continues to grow, we remain uncertain about when supply will finally catch up with the expanding demand curve," stated the Wedbush equity analysis team led by Bryson.
Beyond the 2027 guidance, ASML Holding NV management now expects full-year 2026 total net sales (i.e., total revenue forecast) to be between €43 billion and €45 billion, significantly higher than the previous forecast range of €36 billion to €40 billion and also surpassing the Wall Street consensus estimate of €39.3 billion.
Four Key Supply Chain Signals Challenge the "AI Compute Overcapacity" Thesis
Earnings data shows ASML Holding NV's Q2 revenue reached €9.326 billion, exceeding market expectations of €8.8 billion. Net profit was €2.918 billion, also surpassing the expected €2.62 billion, with a gross margin of 54%. More critically, the company significantly raised its 2026 revenue guidance from €36-40 billion to €43-45 billion, increased its gross margin guidance from 51%-53% to 54%-56%, and expects Q3 revenue to further rise to €11-12 billion.
ASML Holding NV management described first-half orders as "exceptionally strong," with customers accelerating capacity builds for advanced process logic and memory chips. Consequently, the company plans to increase production capacity for both low-NA EUV and immersion DUV lithography systems by approximately 30% by 2027 and is studying a further ~30% increase in 2028. This indicates that major foundry customers like Taiwan Semiconductor Manufacturing and Intel are voting for robust AI compute demand in 2027-2028 with multi-year equipment commitments, not just verbal forecasts.
The rare, substantial capacity expansion by ASML Holding NV signals that advanced logic chip manufacturers like Taiwan Semiconductor Manufacturing and Intel are preparing significantly larger wafer capacity for the growing demand for data center server-grade CPUs, GPUs, and custom AI ASIC accelerators like TPUs. Meanwhile, the persistently tight supply situation in the memory chip segment, highlighted by Samsung's latest earnings and SK Hynix's CEO on the day of its US ADR listing, further validates the continued strong demand for complete AI servers from another core link in the AI compute supply chain.
More crucially, ASML Holding NV management plans to increase the annual production capacity of approximately 65 low-NA EUV systems by 30% in 2027, with a study for another 30% in 2028, following a similar expansion path for immersion DUV systems. This is not short-term inventory replenishment but a global foundry lock-in of upstream capabilities for advanced logic, HBM/DRAM, and advanced packaging through multi-year equipment commitments.
Recent AI industry developments this week, including Meta Platforms, Inc.'s expansion of its Louisiana Hyperion data center campus investment from an initial $10 billion to over $50 billion, creating a supercluster exceeding 5GW of compute power, the procurement of 27,500 Nvidia Rubin chips for Japan's physical AI infrastructure project, and Nvidia CEO Jensen Huang's collaboration with Japanese industrial giants to advance physical AI, all point to a multi-centered expansion. This, combined with the strong, better-than-expected earnings and increasingly optimistic outlook for AI compute demand from Taiwan Semiconductor Manufacturing—the exclusive manufacturer of Nvidia's AI chips—demonstrates that the global AI compute investment cycle is far from over. The compute expansion is upgrading from a US hyperscale cloud provider-led arms race to a multi-center resonance of "cloud AI, sovereign AI, and physical AI."
Taiwan Semiconductor Manufacturing's earnings and outlook validate that this demand is not merely confined to equipment orders. Q2 revenue in US dollars was $40.2 billion, a year-on-year increase of 33.7%. Net profit reached NT$706.56 billion, surging 77.4% year-on-year and significantly exceeding the market expectation of NT$632.6 billion. The High-Performance Computing segment accounted for 66% of revenue, and 7nm and more advanced processes constituted 77% of wafer business revenue.
More significantly, Taiwan Semiconductor Manufacturing raised its Q3 revenue guidance further to $44.6-$45.8 billion. The company also substantially increased its 2026 capital expenditure forecast from $52-$56 billion to $60-$64 billion, raised its full-year US dollar revenue growth guidance to slightly above 40%, and announced an additional $100 billion US investment, bringing its total US commitment to approximately $265 billion.
As the ultimate capacity planner for core AI chip clients like Nvidia and the manufacturing chain fulfiller for AI compute demand, Taiwan Semiconductor Manufacturing is signaling through profits, utilization rates, and capital expenditure that visibility for AI compute infrastructure demand surrounding advanced process logic chips, cutting-edge 2nm processes, and advanced packaging is still rising, not indicating a global slump in AI compute demand.
ASML Holding NV and Taiwan Semiconductor Manufacturing are not mechanically accepting customer forecasts; they are approving capacity expansions after scrutinizing data center build-outs and end-demand. Therefore, this latest set of guidance provides strong counter-evidence to the assertion of "systemic compute overcapacity already present."
Demand is also continuing to expand outward. The Japanese government-supported Noetra project plans to procure 27,500 Nvidia Rubin chips to build physical AI infrastructure, with construction scheduled to start in April 2027 and operation in June 2028. Meta Platforms, Inc. has expanded its Louisiana Hyperion campus from the initially announced $10 billion project in 2024 to a supercluster exceeding $50 billion and 5 gigawatts of compute power. These four key supply chain signals undoubtedly challenge the "AI compute overcapacity" thesis.
Anthropic's Profit Trajectory Leap Signals AI Compute Bull Market Enters "Token Compounding Era"
OpenAI and its long-time rival in the AI applications space, Anthropic PBC, have been racing to develop more advanced AI agents to streamline workflows across broader domains. Previously, both companies achieved significant success with AI development tools capable of automatically writing, debugging, and deploying code. Earlier this year, Anthropic launched a similar product called Claude Cowork, aiming to attract a wider user base into the unprecedented wave of AI agents.
Both OpenAI and Anthropic have confidentially filed for public listings. Reports suggest Anthropic could potentially list on the US stock market as early as this fall, while OpenAI is considering a listing next year.
Latest analysis from research firm SemiAnalysis reveals that Anthropic is reshaping the AI commercialization landscape with profitability and growth rates far exceeding its competitors. Leveraging a high-margin, API-centric business model, Anthropic has become a leader in the B2B AI market.
A deep-dive report from SemiAnalysis projects that Anthropic will achieve $1 billion in GAAP EBIT in the third quarter of 2026, corresponding to a roughly 6% margin. Concurrently, its Annual Recurring Revenue (ARR) has surged from $9 billion at the end of 2025 to over $60 billion currently. The firm predicts that if Anthropic maintains a Net New ARR (NNARR) pace of approximately $15 billion per month, its ARR could reach $300 billion by the end of 2027, corresponding to a $6 trillion enterprise value, potentially making it the world's most valuable company.
Anthropic's inflection point stems from the explosive adoption of Claude Code. Statistics exclusively compiled by SemiAnalysis show that Claude Code currently accounts for over 7% of all code commits on GitHub, directly driving the company's monthly NNARR to surge from $3 billion in January to $11 billion in March in the first quarter. Furthermore, SemiAnalysis estimates show that Anthropic's current comprehensive gross margin has risen to the mid-60% range, compared to negative 94% in 2024, with the API business gross margin exceeding 80%.
The grand investment narrative of global capital "seeking silicon-based inflation, weakening carbon-based" this year is essentially capital shifting from traditional manufacturing, automotive, consumer, real estate, and energy—"carbon-based assets" reliant on population, resources, and linear economic growth—towards the high-end silicon-based manufacturing chain related to AI compute infrastructure.
Therefore, the impending launch of GPT-5.6 with ChatGPT Work and Anthropic's commercialization data jointly reinforce a core investment judgment: the unprecedented demand cycle for AI compute infrastructure is not over; it is transitioning from the AI model training-driven phase to the AI inference application-driven phase. The true super-cycle for AI compute infrastructure may stem from the global, large-scale deployment of AI agents as a new generation of digital employees. This also implies that the current correction in the AI semiconductor sector is a healthy adjustment, not a bear market crash driven by "compute overcapacity."
Latest research from the well-known firm Exponential View indicates the AI industry is upgrading from a capital expenditure cycle primarily dependent on frontier model training to a dual-wheel cycle of "continued training expansion, with inference becoming the main incremental engine." Exponential View's latest bottom-up calculations, which exclude double-counting in the supply chain, show that end-user revenue related to generative AI over the past 12 months has reached $110 billion, with a recent monthly annualized run rate exceeding $175 billion—a growth rate about three times that of the internet and mobile waves at comparable stages. More crucially, the AI compute infrastructure-related revenue attributable to hyperscale cloud providers is now roughly sufficient to cover the depreciation costs of new computing assets.
Exponential View's exclusive model also shows that a 10% decrease in Token price leads to a 12%-18% increase in usage, implying that a reduction in unit inference cost does not necessarily suppress total revenue but may instead expand overall compute expenditure through demand elasticity. In other words, the AI compute infrastructure clusters being vigorously built by large AI tech giants like Microsoft, Meta Platforms, Inc., and Google are transitioning from an "expectation trade" of "build first, find revenue later" to an economic closed-loop validated by real Token consumption and enterprise payments.
Comments