尽管算力基础设施的需求持续供不应求，但英特尔与博通两家企业的首席执行官发出警告，资金获取能力或将成为人工智能发展的一大主要约束。博通 CEO 陈福阳称，全行业都在加紧建设基础设施，支撑复杂度越来越高的 AI 大模型。相较于半导体、存储芯片或是电力短缺，融资问题可能会成为 AI 扩张更大的阻碍。
“我认为最大的制约因素或许是资本。” 陈福阳周一于新加坡星展亚洲未来前瞻峰会上发言表示。
英特尔 CEO 陈立武也对融资风险表达了相同的担忧。他指出，资本成本持续走高，同时市场也心存疑问：大举投入 AI 的企业，最终能否创造足够回报。
“如果说未来存在挑战，那就是资本以及资本成本；从更长期角度看，市场希望看到实实在在的投资回报率（ROI）。” 他说道。
两位高管的发言，揭示出 AI 行业正在浮现的一大难题。为应对算力需求暴涨，行业已经向数据中心、先进芯片、电力基础设施投入海量资金。
当前借贷成本抬升，可能会给重金押注 AI 基建的科技企业的融资带来麻烦。债券收益率走高，将抬高数据中心建设、芯片采购的资金成本；而这类项目往往需要数年时间才能产生收益，项目的商业可行性或将承受压力。
陈福阳表示，过去两年行业接连遭遇各类供给瓶颈，但随着产能扩张，部分约束条件有望得到缓解。
“过去两年，各种瓶颈层出不穷，但我们总能逐一克服。如今的存储短缺，一年之后或许就不复存在。”
他预判，晶圆紧缺局面有望在 2028‑2029 年得到缓和；同时随着新建基础设施陆续投产，电力供给情况也会改善。
虽然担忧融资环境，但两位高管依旧看好 AI 算力基础设施的短期需求。
陈福阳称：“算力基础设施的市场需求已经超出行业供给能力。” 他补充，这种供需失衡状态或将持续至 2029 年，甚至延续到 2030 年。
英特尔的陈立武同样认为，受 AI 应用普及、算力需求拉动，未来两到三年行业不太会出现显著的增长失速。
但两位高管同时提到：数年之后行业能否维持高增长，取决于企业落地 AI 的速度，以及 AI 技术能否创造实实在在的经济效益。
博通的陈福阳提醒：倘若高级 AI 模型落地不及预期，或是模型性能提升进入平台期，算力基础设施的需求就有可能降温。
而陈立武则着重强调：企业越来越需要通过效率提升、全新商业模式，来证明 AI 投资可以带来看得见摸得着的回报。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
While the demand for computing infrastructure continues to fall short of demand, the CEOs of Intel and Chase warned that access to finance might be a major constraint to artificial intelligence development. Chase CEO Chen Fook Yang, the industry is stepping up its infrastructure to support AI LLM, which is becoming increasingly complex. Financing problems may be a greater barrier to AI expansion than semiconductor, storage chip or power shortages.
"I think perhaps the biggest constraint is capital." Chen Fookyang addressed the Singapore Star Show Asia Future Vision Summit on Monday.
Intel CEO Chen Li-Wu shares concerns about financing risks. He noted that capital costs continued to rise, while the market questioned whether a large number of AI-invented enterprises would ultimately generate sufficient returns.
“If there are challenges in the future, it is capital as well as capital costs; in the longer term, markets want to see real returns on investment (ROI). He said.
The presentations by two executives revealed a major problem that the AI industry is experiencing. In response to the surge in computing demand, the industry has invested significant amounts in data centres, advanced chips and power infrastructure.
The current increase in borrowing costs may cause difficulties in financing AI-based technology enterprises. A higher rate of return on bonds would raise the financial cost of data centre construction and chip purchases; such projects often took several years to generate benefits and the commercial viability of the projects might be under pressure.
According to Chen Fookyang, the industry has experienced successive supply bottlenecks over the past two years, but some of the constraints are expected to be eased as production capacity expands.
“Over the past two years, bottlenecks have emerged, but we can always overcome them one by one. Today's storage shortage may cease to exist one year later.”
He predicted that the shortfall would be eased in 2028-2029, while the supply of electricity would improve as the new infrastructure became operational.
Despite concerns about the financing environment, the two executives still appreciate the short-term needs of AI ' s computing infrastructure.
Chen Fookyang states: “The market demand for computing infrastructure has exceeded the supply capacity of the industry. He added that this imbalance in supply and demand would continue until 2029 or even 2030.
Intel's Chen Liwu shares the view that, driven by the demand for universal access to AI, there will not be a significant slowdown in the industry over the next two to three years.
However, the two executives also mentioned that the sustainability of high growth in the industry after several years would depend on the speed at which firms landed in AI and whether AI technology produced real economic benefits.
JT: If the advanced AI model falls short of expectations, or if model performance increases into the platform period, the need for computing infrastructure is likely to cool down.
Chen Ji-Wu stressed the growing need for businesses to demonstrate that AI investments can deliver tangible returns through efficiency gains and new business models.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.