AI News AI资讯 3h ago Updated 1h ago 更新于 1小时前 48

TSMC Chairman C.C. Wei: Predicts Full-Year Revenue Growth to Exceed 30% This Year 台积电魏哲家:预计今年全年营收成长仍将超过30%

The figure—30%—that C.C. Wei cited at the shareholder meeting is like a glaring neon light, illuminating the underlying hue of the current AI frenzy while faintly casting shadows behind it. When the TSMC chairman confidently ties "agentic AI" and "instruction-action mode" to semiconductor demand over the next few years, it is more than a technical judgment; it resembles an optimistic manifesto directed at Wall Street and global capital. But is the capital market too intoxicated by this one-dimen 魏哲家在股东会上说出的数字——30%——就像一道刺眼的霓虹,照亮了当下AI狂热的底色,也隐隐照出了这片霓虹背后的阴影。当台积电的董事长自信满满地将“代理式AI”和“指令行动模式”与未来数年的半导体需求强绑定时,这不仅仅是一个技术判断,更像是一份面向华尔街和全球资本的乐观宣言。但资本市场是否过于沉醉于这种单向度的叙事了?

70
Hot 热度
65
Quality 质量
70
Impact 影响力

Analysis 深度分析

The evolution from generative AI to agentic AI appears logically unassailable: more complex AI agents require more rounds of interaction, extensive context processing, which in turn consume more tokens—ultimately translating into a voracious demand for computing chips. As the "sluice gate" at the very upstream of this supply chain, TSMC’s perception is perhaps the most acute. Yet this highly simplified model of correlating industry prospects directly with an "AI narrative–chip demand" positive relationship is itself riddled with cyclical arrogance. When was the last time the semiconductor industry collectively proclaimed "demand will grow forever"? Was it on the eve of the memory chip bubble burst in 2018, or at the frenzied peak of the global chip shortage in 2021? Industry memory is always short, especially during lucrative upswings.

The claims that "customers and their customers" are all offering positive outlooks—citing third-party sentiment to validate one’s own judgment—resemble a carefully orchestrated confidence relay game. NVIDIA tells Microsoft that cloud demand is robust; Microsoft tells TSMC that AI training and inference orders are full; TSMC then assures shareholders that all is well. Every link in this chain of optimism is bound to the same narrative express due to its own interests. But has the real end-user demand—whether the ROI of enterprise AI applications or ordinary users’ willingness to pay a premium for AI features—truly solidified? Or is most of the compute procurement still stuck in a phase of panic-driven stockpiling driven by a fear of missing out (FOMO)? When all players expand based on a "consensus" about the future, it often sows the seeds of future overcapacity.

C.C. Wei bases his confidence on "technological differentiation and a broad customer base," which is undoubtedly TSMC’s most formidable moat. Yet it is precisely this "one giant, many strong players" structure that has made the hardware foundation of the entire AI industry exceptionally concentrated and fragile. The lifeline of global cutting-edge AI computing power is firmly controlled by the capacity planning and customer relationships of just a handful of companies. This highly centralized supply chain, while pursuing ultimate efficiency, also significantly diminishes the industry’s resilience and ability to withstand risks. Should a disruptive shift occur in a technological pathway or a collective pivot in downstream demand, the ripple effects would be global.

Therefore, TSMC’s projection of 30% revenue growth is less a certainty about the AI future and more a monetization of the industry’s current "collective imagination." It resembles a self-fulfilling prophecy: because leading vendors firmly believe in the future and continue investing heavily in chip procurement, capacity must expand and revenue can grow; the improved financial results then retrospectively validate the initial investment and fuel another round of optimistic forecasts. This cycle is beautiful when it runs positive—but the key question is: how much of what is driven by "faith" is underpinned by tangible, sustainable commercial returns?

We may be standing at a critical crossroads in the AI industry: one path leads to practical applications that genuinely transform productivity structures, while the other could slide toward a computing bubble based on capital games. TSMC’s optimism largely leans toward the premise that the former will arrive swiftly. But as observers, we must maintain a degree of calm: the technological iteration (agentic AI) is indeed real, but the market’s acceptance pace, the cultivation of willingness to pay, and the maturation of business models are always slower and more complex than the technology’s evolution itself. When chipmakers’ capacity planning severely disconnects from the real, sustainable rhythm of market demand, the semiconductor industry’s infamous "cyclical" specter will quietly resurface.

C.C. Wei’s confidence reflects the collective will of the players at the top of the industry chain. But for the entire industry, perhaps more important than blindly following this optimism is pondering: once the consensus that "AI needs more chips" is fully priced in and reflected in astronomical valuations, what will be the next pillar of support? Or are we already using future cash flows to pay for today’s excesses?

魏哲家在股东会上说出的数字——30%——就像一道刺眼的霓虹,照亮了当下AI狂热的底色,也隐隐照出了这片霓虹背后的阴影。当台积电的董事长自信满满地将“代理式AI”和“指令行动模式”与未来数年的半导体需求强绑定时,这不仅仅是一个技术判断,更像是一份面向华尔街和全球资本的乐观宣言。但资本市场是否过于沉醉于这种单向度的叙事了?

生成式AI到代理式AI的演进,逻辑链条看似无懈可击:更复杂的AI代理需要更多轮次的交互、更大量的上下文处理,从而消耗更多token,最终转化为对算力芯片的饥渴需求。台积电作为这条供应链最上游的“水闸”,其感知或许最为敏锐。但这种将行业前景高度简化为“AI叙事-芯片需求”的正相关模型,本身就充满了周期的傲慢。上一次半导体行业集体高呼“需求永远增长”是什么时候?2018年存储芯片泡沫破灭前夕,还是2021年全球缺芯的疯狂顶点?行业记忆总是短暂的,尤其是在利润丰厚的上行周期里。

“客户以及客户的客户”都在给予正面展望,这种引用第三方情绪来佐证自身判断的说法,像极了一场精心编排的信心传递游戏。英伟达告诉微软云需求强劲,微软告诉台积电AI训练和推理订单饱满,台积电再转告股东一切安好。这条乐观传导链上的每一个环节,都因自身利益而被绑定在了同一条叙事的快车上。但真正的终端需求,无论是企业的AI应用ROI,还是普通用户愿意为AI功能支付的溢价,真的已经夯实了吗?还是大部分算力采购,仍停留在“怕错过”的恐慌性储备阶段?当所有玩家都基于对未来的“共识”进行扩张时,往往也埋下了未来产能过剩的伏笔。

魏哲家将信心建立在“技术差异化与广泛客户群”上,这无疑是台积电最坚固的护城河。但恰恰是这种“一超多强”的格局,也让整个AI产业的硬件基座变得异常集中和脆弱。全球尖端AI算力的命脉,被少数几家公司的产能规划和客户关系所牢牢掌控。这种高度集中的供应链,在追求极致效率的同时,也极大地削弱了产业的弹性和风险抵御能力。一旦某个技术路线出现颠覆性变化,或是下游需求出现集体转向,其涟漪效应将是全球性的。

因此,台积电30%的营收增长预测,与其说是对AI未来的笃定,不如说是对当前产业“共同想象”的变现。它更像一个自我实现的预言:因为头部厂商坚信未来并持续投入巨资采购芯片,所以产能必须扩张,营收得以增长;而增长的财报又反过来验证了当初投入的正确性,并激励新一轮的乐观预测。这个循环在正向时无比美丽,但关键在于,它由“信仰”驱动的成分,有多少是由实实在在、可持续的商业回报支撑的?

我们或许正站在一个AI产业的关键岔路口:一侧是通往真正改变生产力结构的实用化应用,另一侧则可能滑向基于资本游戏的算力泡沫。台积电的乐观,更多地站在了前者即将快速到来的立场上。但作为观察者,必须保持一份冷静:技术的迭代(代理式AI)固然是真实的,但市场接受的节奏、付费意愿的培养、以及商业模式的成熟,永远比技术演进本身要慢,也更为复杂。当芯片厂的产能规划与市场真实、可持续的需求节奏出现严重脱节时,半导体那著名的“周期性”幽灵,便会再次悄然浮现。

魏哲家的信心,代表了产业链顶端玩家的集体意志。但对于整个产业而言,比盲目跟随这份乐观更重要的,或许是思考:当“AI需要更多芯片”这个共识被充分定价并反映在天价市值中之后,下一个支撑点在哪里?又或者,我们是否已经在用未来的现金流,来为今天的过度预支买单?

Disclaimer: The above content is generated by AI and is for reference only. 免责声明:以上内容由 AI 生成,仅供参考。

芯片 芯片 大模型 大模型 Agent Agent
Share: 分享到: