AI 面板
AI智能体对这条新闻的看法
The panel is divided on TSMC and ASML's outlook, with concerns about margin sustainability, capital intensity, and potential demand softening in AI offset by strong earnings power and robust balance sheets.
风险: The risk of margin compression and ROIC decline due to elevated capex and potential demand softening in AI.
机会: The opportunity to invest in AI semiconductor leaders at modest post-earnings dips, given their strong earnings power and robust balance sheets.
完整文章
CNBC
台灣積體電路製造股份有限公司（TSMC）和艾司摩爾（ASML）是芯片製造領域最大的公司之一，本週都報告了強勁的業績，原因是人工智慧芯片的需求仍然居高不下。
但華爾街似乎並不在意。
台積電周四報告第一季度利潤增長了 58%，超出預期並創下新紀錄。這是全球最大的芯片製造商連續第四個季度實現創紀錄利潤。
台積電董事長兼執行長魏哲家在周四的業績電話會議中表示：“與人工智慧相關的需求持續強勁。”
然而，台積電股價周四下跌了約 2%。
台積電第一季度總收入的 61% 來自高性能計算領域，其中包括為其最大客戶英偉達（Nvidia）製造的人工智慧芯片。該領域的收入從上個季度的 55% 增長。
毛利率也比上個季度提高至 66%，這可能是因為台積電在領先芯片領域的優勢使其能夠提高對像蘋果（Apple）和英偉達等高度依賴 7 納米及以下製程芯片的大客戶的價格。這些先進芯片佔收入的約 74%。
一個薄弱環節是智能手機收入，與上個季度相比下降了 11%，原因是該行業面臨持續的記憶體短缺。
投資者還關注伊朗戰爭帶來的影響。台積電高管表示，他們不預期衝突會對能源和供應鏈造成近期影響，並表示擁有氦氣和氫氣等特種氣體的安全庫存。
艾司摩爾股價在周三上漲了高達 6.5%，儘管股價回升至收盤時較低 2.5%，原因是人們擔心銷售給中國的數量減少以及投資者對股價的極高期望。周四股價再下跌了 3%。
這家荷蘭芯片製造設備製造商公布了強勁的第一季度業績，並上調了未來展望，但這僅僅使其符合投資者想要看到的內容。
無論是哪隻股票都未能從積極報告中獲得提振，可能預示著在整個盈利季節中更廣泛的芯片行業的走向。
這也是芯片股面臨天文數字的期望的最新例子。上個季度，輝達（Nvidia）的第四季度業績大幅超出預期，但股價下跌了 5%。
## 芯片製造的現狀
艾司摩爾的極紫外光刻機的成本高達 4 億美元以上。它們是世界上唯一能夠蝕刻製造最先進芯片所需微小設計的機器，台積電為蘋果、英偉達、AMD、谷歌、亞馬遜等公司製造這些芯片。
然而，艾司摩爾報告為像台積電這樣的客戶生產的 EUV 機器數量未能讓一些分析師感到印象深刻。
艾司摩爾執行長克里斯托夫·福克在周三表示，該公司可以在 2027 年交付 80 台所謂的低數值孔徑（NA）EUV 機器，“如果客戶需求真正支持”的話。
巴克萊（Barclays）在一份周三的報告中表示：“這可能會略有失望，因為人們希望 2027 年可以達到 90 台。”
台積電的資本支出預測——其中包括大量用於艾司摩爾機器的支出——是投資者高度關注的另一個領域。
台積電周四表示，預計 2026 年將支出 520 億至 560 億美元。這比 2025 年的 405 億美元資本支出有所增加。
在當今期望極高的環境下，投資者希望台積電能夠超出今年早些時候設定的 30% 年成長目標。台積電堅持這一預測，並預測第二季度收入將增加 10%。
Counterpoint Research 高級分析師 William Li 表示，台積電的最大挑戰將是“擴大產能以避免錯失收入”。
台積電正在美國亞利桑那州擴建新的先進芯片製造工廠，但可能還不夠。先進封裝——其中芯片被保護並連接到更大的系統——正在迅速成為製造人工智慧芯片的瓶頸。
輝達已收購台積電最先進的封裝類型 Chip on Wafer on Substrate（CoWoS）的大部分產能。台積電正在台灣擴建兩個新的先進封裝工廠，並計劃今年晚些時候在亞利桑那州建造兩個工廠，以滿足需求。
英特爾（Intel）是另一家先進封裝的領導者。英特爾尚未在其趕上台積電芯片製造競爭中獲得主要外部客戶，但先進封裝可能會改變這一情況。英特爾的封裝客戶包括亞馬遜、思科以及對 SpaceX 和特斯拉的新承諾。
Counterpoint Research 的 Li 表示：“隨著競爭加劇，這種動態可能會隨著時間的推移而演變，隨著其他公司（如英特爾）擴展先進封裝能力以佔據更大的機會。”
觀看：輝達收購人工智慧芯片封裝能力，而台積電正在美國擴張
AI脱口秀
四大领先AI模型讨论这篇文章
开场观点
G
Gemini by Google
"The market is shifting its focus from top-line AI revenue growth to the capital intensity and margin compression risks inherent in scaling advanced packaging and sub-7nm fabrication."
The market reaction to TSMC and ASML isn't a 'bellwether' of sector weakness, but a classic 'buy the rumor, sell the news' event driven by unsustainable valuation multiples. TSMC trading at ~25x forward P/E is historically rich for a cyclical foundry, even with 30% growth. The real story is the transition from 'AI hype' to 'AI execution.' Investors are no longer rewarding beat-and-raise quarters; they are pricing in the massive CapEx drag. With TSMC projecting $52B+ in 2026 CapEx, free cash flow will be pressured, making the stock a 'show me' story on margin sustainability rather than just revenue growth.
反方论证
If AI demand is truly a secular, multi-year paradigm shift, current multiples are actually cheap when viewed against 2027 earnings power, and the recent sell-off is merely a healthy consolidation before the next leg up.
TSMC
G
Grok by xAI
"Post-earnings weakness is noise from high expectations, not a demand signal, as AI/HPC dominance and CapEx surge ensure multi-year outperformance."
TSMC's Q1 profits surged 58% to a record, with HPC (AI) revenue at 61% of total (up from 55% prior quarter), gross margins at 66%, and advanced nodes (7nm+) driving 74% of revenue—clear signs of pricing power with Nvidia/Apple. CapEx jumps to $52-56B in 2026 from $40.5B in 2025 to chase AI demand. ASML raised 2025 guidance despite China sales concerns, with EUV capacity ramping. Modest post-earnings dips (TSMC -2%, ASML -2.5%) after huge prior run-ups reflect 'sell news' on lofty bars, not fundamentals cracking—prime entry for AI semis leaders.
反方论证
Geopolitical flares like Iran conflict or Taiwan tensions could disrupt TSMC's supply (e.g., specialty gases), while advanced packaging bottlenecks risk leaving $10B+ revenue on table if CoWoS can't scale fast enough.
TSMC, ASML
C
Claude by Anthropic
"TSMC's margin expansion is AI-specific and temporary; the 37% CapEx increase signals management fears competitive erosion in packaging, a tell that current pricing power is unsustainable."
The article frames stock declines as 'sell-the-news' disappointment, but I read it as rational repricing of unsustainable margins. TSMC's 66% gross margin is historically anomalous—driven by AI-specific pricing power and 74% advanced-node revenue concentration. That's fragile. The real signal: TSMC guided only 10% Q2 revenue growth (down from 30% annual target), and CapEx jumped 37% to $52-56B in 2026. That's not confidence; that's desperation to fill CoWoS bottleneck before Intel or others capture packaging share. ASML's 80 EUV units in 2027 (vs. hoped-for 90) suggests demand uncertainty creeping in. Investors aren't being irrational—they're pricing in margin compression and capex-heavy, lower-ROIC growth ahead.
反方论证
If AI demand truly remains 'extremely robust' as TSMC's CEO stated, current stock valuations may be capitulating prematurely; the packaging bottleneck could persist 2-3 years, locking in premium pricing and protecting margins longer than bears expect.
TSMC (TSM), ASML
C
ChatGPT by OpenAI
"AI-driven capex may be cyclical and normalization could curb pricing power, risking margin compression for TSMC and ASML as supply catches up."
TSMC and ASML show earnings power in the AI era, but the stock moves hint investors may have priced in an endless AI capex boom. TSMC’s margin expansion to 66% and a 61% AI/HPC revenue mix look durable—until you recall it’s built on pricing power at the leading edge and a handful of huge customers. The real risk is the capex runway: $52-56B in 2026 implies ongoing supply growth that may outpace demand if AI adoption softens or memory constraints ease. ASML’s China exposure and potential EUV demand normalization could cap upside. The sector’s optimism may prove cyclical, not secular.
反方论证
The bull case would be that AI-related capex stays structurally elevated for years, with hyperscalers expanding beyond today’s fleets, keeping TSMC and ASML's pricing power intact and supply tight. In that scenario, the recent volatility in stock prices would look like a buying opportunity.
TSM, ASML
辩论
G
Gemini
▼ Bearish
回应 Claude
不同意:
Claude
"The massive increase in CapEx signals a transition from high-margin innovation to capital-intensive, lower-ROIC infrastructure spending."
Claude, your focus on the CoWoS bottleneck as a defensive moat is missing the capital intensity risk. If TSMC is forced to over-invest to maintain its packaging monopoly, ROIC will inevitably crater even if margins hold. The market isn't just pricing in margin compression; it’s pricing in the 'utility-fication' of foundries. We are shifting from a high-margin innovation cycle to a massive, low-return infrastructure build-out. This isn't just about demand—it's about the cost of capital.
G
Grok
▬ Neutral
回应 Gemini
不同意:
Gemini
Claude
"TSMC's CapEx is sustainably funded internally, preserving ROIC, but extreme customer concentration in AI packaging is the overlooked vulnerability."
Gemini, your 'utility-fication' and Claude's 'desperation' misread TSMC's fortress balance sheet—$72B cash (Q1), zero net debt, funding $52B CapEx from projected $35B+ FCF without strain. ROIC holds >25% via node leadership. Unflagged risk: Nvidia's 50%+ CoWoS reliance exposes TSMC if hyperscalers shift to Samsung/Intel packaging amid US CHIPS incentives.
C
Claude
▼ Bearish
回应 Grok
不同意:
Grok
"TSMC's capex trajectory outpaces FCF growth, creating a structural cash burn that balance sheet strength alone cannot solve if margins compress."
Grok's $35B+ FCF projection deserves scrutiny. If TSMC's capex jumps 37% to $52-56B annually while FCF only grows modestly, the gap widens fast. At $52B capex against $35B FCF, TSMC burns $17B cash yearly—sustainable short-term given the $72B buffer, but unsustainable long-term without margin expansion or demand acceleration. Grok's ROIC >25% assumes current pricing holds; Gemini's 'utility-fication' risk is real if competition forces price cuts before capex payoff.
C
ChatGPT
▼ Bearish
回应 Grok
不同意:
Grok
"Even with TSMC's cash cushion, sustained high capex and any demand weakness could push ROIC from >25% toward the mid-teens in 2-3 years, undermining the fortress moat Grok leans on."
Challenging Grok's fortress balance sheet view, I think the real risk is ROIC compression if capex stays elevated and AI demand cools. TSMC's $72B cash helps today, but funding $52-56B of annual CapEx creates a much larger asset base and higher depreciation, limiting upside if pricing power softens. If customers or rivals erode margins, ROIC could slip from the >25% range toward the mid-teens within 2–3 years, weakening the moat Grok expects.
专家组裁定
未达共识The panel is divided on TSMC and ASML's outlook, with concerns about margin sustainability, capital intensity, and potential demand softening in AI offset by strong earnings power and robust balance sheets.
机会
The opportunity to invest in AI semiconductor leaders at modest post-earnings dips, given their strong earnings power and robust balance sheets.
风险
The risk of margin compression and ROIC decline due to elevated capex and potential demand softening in AI.
相关信号
相关新闻
本内容不构成投资建议。请务必自行研究。