--- title: "Why 2026 will be the year AI hype collides with reality" type: "News" locale: "zh-HK" url: "https://longbridge.com/zh-HK/news/268858744.md" description: "The article discusses the speculative assumptions and challenges in AI investments, driven by US-China rivalry. It highlights escalating costs, stock valuations, and energy bottlenecks. Morgan Stanley projects $3 trillion spent on data centers by 2028. China's AI investment could hit 700 billion yuan this year. AI adoption is slower than expected, with uneven progress. The article questions whether AI promises are myths and explores circular business deals among AI companies, highlighting potential stranded assets and competing AI paradigms." datetime: "2025-12-07T12:35:36.000Z" locales: - [zh-CN](https://longbridge.com/zh-CN/news/268858744.md) - [en](https://longbridge.com/en/news/268858744.md) - [zh-HK](https://longbridge.com/zh-HK/news/268858744.md) --- > 支持的語言: [简体中文](https://longbridge.com/zh-CN/news/268858744.md) | [English](https://longbridge.com/en/news/268858744.md) # Why 2026 will be the year AI hype collides with reality Gales of creative destruction will sweep away folly and naivety next year as they batter what appears to be the largest boom in modern history – the scramble to dominate artificial intelligence (AI) driven by US-China rivalry. The market debut of China’s Moore Threads last week, its shares more than quintupling on the first day, underscores the feverish momentum.\\nSpeculative assumptions guiding trillions of US dollars in AI investments are colliding with real-world obstacles. Escalating costs, stratospheric stock valuations, tenuous collaborations and energy bottlenecks are compounding the inevitable challenges when new technologies struggle for profitability. Many are worried the bubble may be bursting.\\nMorgan Stanley projects that the cumulative amount spent worldwide on data centres could exceed US$3 trillion by year-end 2028. China’s AI investment could hit 700 billion yuan (US$99 billion) this year, 48 per cent more than last year, according to Bank of America, with the government supplying US$56 billion.\\nSuch vast capital expenditures will only pay off if demand for AI services grows fast enough to absorb the costs. Bain forecasts that companies will need US$2 trillion in annual revenue to profitably fund the computing power required to meet AI demand by 2030, but the world is still US$800 billion short of keeping pace with demand.\\nChina’s AI industry could add more than 11 trillion yuan to the country’s gross domestic product by 2035, about 4 to 5 per cent of total output, according to the China Telecom Research Institute. Even that pace is likely to fall short of the revenue needed to recoup investment costs.\\nAI adoption is proving slower and more uneven than expected. Apollo Academy has reported a decline in AI adoption among US firms, drawing on the US Census Bureau’s fortnightly survey of more than a million firms. The Saint Louis Federal Reserve found only a modest increase in work-related generative AI use, from 33.3 per cent to 37.4 per cent of US workers, in the 12 months to August, as non-work uses rose more sharply. Chinese companies are likely experiencing similar patterns.\\n\\n\\nWhy? “AI’s emergent capabilities form what’s often called a ‘jagged frontier,’ where it excels at some tasks while struggling with others,” professors Will Drover and Laura Huang write. “As its capabilities progress, AI is racing to fully automate some tasks while slowing to a crawl or stalling entirely for others.” Businesses face a volatile mix of sudden leaps and hard limits, of functions that can be reliably dependent on AI and others that will remain stubbornly human-based. This unpredictability complicates investment plans.\\nIn China, the AI agent penetration rate is much lower, according to China International Capital Corporation, 17.7 per cent versus 40 per cent in the US, reflecting “relatively weak” digital infrastructure and tight corporate budgets. The “AI+” plan issued in August aims to reach a penetration rate of above 70 per cent across various industries by 2027 and 90 per cent by 2030. That’s an enormous gap to fill for a slowing economy buffeted by challenges: a collapse in real estate, significant rural-urban education gaps, trade tensions, high public debt, diminishing returns on investment and entrenched deflation.\\nIn the coming months we will learn whether the speed and scope of adoption worldwide continue to falter. That will indicate AI’s success ahead and whether its promises, such as vast gains in productivity, are myths.\\n\\n\\nNext year will also see whether circular business deals – companies lending money to others to buy their products – among AI titans endure or implode. Nvidia announced a deal to invest billions in OpenAI, which has reached US$500 billion in value. In return, OpenAI is expected to purchase a significant amount of Nvidia’s AI chips.\\nOpenAI gained the option to buy a piece of Advanced Micro Devices in exchange for agreeing to purchasing its chips. OpenAI plans to spend upwards of US$1 trillion that it doesn’t yet have on infrastructure build-outs. The company burns capital with such intensity that its operating losses are expected to soar to US$74 billion in 2028 alone. Those investments could become stranded assets chasing revenues that never materialise.\\nEven worse, such investments may be pouring money down the wrong hole. Two paradigms are competing to set frameworks for AI: “large language” and “world” models.\\nLarge language models train on vast data sets to generate humanlike responses. These models power the likes of ChatGPT, Gemini and Claude.\\nIn “world” models, systems create internal representations of how the world works to reason, plan and simulate future outcomes, moving AI beyond pattern recognition towards agency and decision-making. Advancements could overshadow LLMs, leaving AI megaliths stuck with worthless tools.\\n\\n\\nPhysical constraints – such as energy consumption, chip design, heat dissipation, copper supplies and data centre infrastructure – are emerging as hard limits on the pace, cost and geographic spread of AI development, regardless of software breakthroughs. For AI to scale sustainably, breakthroughs are needed in materials science, cooling systems, energy supplies, mining processes and computational efficiencies. These will be just as critical as algorithmic advances. A flurry of advancements next year would augur well for AI’s quantum leaps forward, while fits and starts will test investors’ patience.\\nThe mantra of “make it, don’t fake it” is growing louder. The reckoning ahead promises to reprice expectations, force economic trade-offs and call out circular deals. The point at which the flickers of doubt become amplified into an avalanche will be investors’ top concern next year.\\n ### 相關股票 - [Morgan Stanley (MS.US)](https://longbridge.com/zh-HK/quote/MS.US.md) - [C3.AI (AI.US)](https://longbridge.com/zh-HK/quote/AI.US.md) ## 相關資訊與研究 - [08:03 ETCensys Raises $70 Million in Strategic Funding to Expand Its Internet Intelligence Platform](https://longbridge.com/zh-HK/news/281182734.md) - [Morgan Stanley China A Share Fund, Inc. Announces Preliminary Results of Tender Offer | MS Stock News](https://longbridge.com/zh-HK/news/281370291.md) - [Exclusive-E*Trade in talks to lead SpaceX IPO share sale to small investors in US, sources say](https://longbridge.com/zh-HK/news/281057646.md) - [BAWAG Group AG (0B2) Receives a Buy from Morgan Stanley](https://longbridge.com/zh-HK/news/280842472.md) - [08:56 ETAdversa AI Wins "Most Innovative Agentic AI Security" Platform at Global InfoSec Awards During RSA Conference 2026](https://longbridge.com/zh-HK/news/281372271.md)