博通首席执行官黄仁勋将许多AI投资者比作对自己在押注什么知之甚少的赌徒。他周一在新加坡举行的DBS亚洲未来前瞻大会上发表了上述言论，彭博社报道。
Broadcom chief executive Hock Tan compared many AI investors to gamblers who have little idea what they are betting on. He spoke on Monday at the DBS Asia Future Forward Conference in Singapore, Bloomberg reported.
“许多投资者根本不知道自己在投资什么，”黄仁勋说。他们“设定了极端的期望，结果发现这几乎就像他们身处赌场。”黄仁勋与英特尔首席执行官林孝俊同台，参加了一个关于下一万亿美元AI投资将从何而来的小组讨论。DBS首席执行官陈淑善担任主持。TikTok首席执行官周受资在同一场会议上更早发言。
“A lot of investors have no clue what they are investing in,” Tan said. They “set very extreme expectations only to find out that it’s almost like they are in a casino.”Tan shared the stage with Intel CEO Lip-Bu Tan, in a panel on where the next trillion dollars in AI will come from. DBS chief executive Tan Su Shan moderated. TikTok’s Shou Chew spoke earlier at the same conference.
“浪费金钱”在代币上黄仁勋表示，大多数公司尚未找到能够盈利的AI用途，《商业时报》报道。“人们只是在浪费金钱，花费代币，试图理解（AI），”他说。
“Wasting money” on tokens Hock Tan said most companies have not yet found AI uses that pay off, The Business Times reported. “People are just wasting money, spending tokens, trying to understand (AI),” he said.
他将其比作游戏厅里一个孩子，十分钟内花掉10美元却一无所获。“这就是企业今天正在发生的事情，”他说。
He compared it to a child at an arcade who spends $10 in ten minutes and learns nothing. “That’s what’s happening in enterprises today,” he said.
他表示，OpenAI在9月份的年化收入为500亿美元，“对于一家初创公司来说，这本身并不算糟糕”。OpenAI仍在亏损，因为它在训练上投入巨额资金，他说。因此，这一数字目前还无法说明长期盈利前景。
He said OpenAI’s annualised revenue of $50bn in September was “by itself, not that bad” for a startup. OpenAI is still losing money because it spends heavily on training, he said. So the figure says little yet about long-term profits.
资本成为瓶颈两位首席执行官均表示，资金可能成为制约AI发展的更大因素，而非芯片、内存或电力，CNBC报道。“我认为最大的制约因素可能是资本，”黄仁勋说。
Capital as the bottleneck Both CEOs said money could become a bigger constraint on AI than chips, memory or power, CNBC reported. “What I think might be the biggest constraint is capital,” Hock Tan said.
AI的未来价值在于前沿模型构建者，而非“淘金者的铲子和镐”般的硬件供应商，黄仁勋说。这些模型构建者将争夺与当下基础设施公司相同的融资来源。“如果前端有任何挑战，那就是资本和资本成本，”林孝俊说。他补充道，从长远来看，人们希望看到投资回报。
AI’s future value lies with frontier model builders, rather than with the “picks and shovels” hardware suppliers, Hock Tan said. Those model builders will chase the same sources of financing as today’s infrastructure companies. “If there’s any challenge in the front, it’s capital and the cost of capital,” Lip-Bu Tan said. In the longer run, he added, people want to see a return on their investment.
即便如此，对计算基础设施的需求仍然超过供应，霍克·谭表示。这种情况可能会持续到2029年，甚至可能延续到2030年。他预计晶圆短缺将在2028年或2029年得到缓解。利普-Bu谭表示，他在未来两到三年内看不到显著的放缓。
Even so, demand for computing infrastructure still outstrips supply, Hock Tan said. That could last through 2029, possibly into 2030. He expects wafer shortages to ease by 2028 or 2029. Lip-Bu Tan said he saw no significant slowdown in the next two to three years.
博通自身正在筹集600亿美元的债务，以帮助Anthropic及其他客户支付AI芯片和基础设施的费用。
Broadcom itself is raising $60bn in debt to help Anthropic and other customers pay for AI chips and infrastructure.
利普-Bu谭还表示，中国的前沿模型如DeepSeek、Kimi和GLM是“足够好的”。他说，它们较低的令牌成本推动了“非常快速”的采用，路透社报道。
Lip-Bu Tan also said Chinese frontier models such as DeepSeek, Kimi and GLM are “good enough”. Their lower token costs have driven “very fast” adoption, he said, Reuters reported.
马斯克和市场“赌场里不许哭泣”，埃隆·马斯克在X上写道，回应了一条关于谭言论的帖子。
Musk and the markets“No crying in the casino,” Elon Musk wrote on X, replying to a post about Tan’s remarks.
周二，亚洲的芯片股票下跌。韩国的Kospi指数在交易过程中下跌超过7%。截至午盘，三星电子下跌9%，SK海力士下跌10%，商业内幕报道。
Chip stocks fell in Asia on Tuesday. South Korea’s Kospi index dropped more than 7% during trading. Samsung Electronics fell 9% and SK Hynix 10% by midday, Business Insider reported.
投资者也在权衡中国存储器制造商长鑫存储（CXMT）的进展。该公司表示，它在一种能够将更多数据存储在存储芯片上的技术上取得了进展。
Investors were also weighing progress by Chinese memory maker CXMT. The company said it had made progress on technology that packs more data onto memory chips.