Skepticism About Agents and Capex Abounds, but not Doubts About Importance of 'Beating' China in AI

Capex Skepticism Mounts Amid December Market Volatility

As any frequent reader of these Pulse reports could testify, media coverage of artificial intelligence infrastructure spending has consistently represented both sides of the narrative for much of 2025. It promotes the overwhelming belief that AI and data center spending will continue to pile up AND the overwhelming belief that it may never pay off. The latter appears to be the flavor that will carry us into 2026. Perscient's semantic signature tracking skepticism about AI capital expenditure as a risky bet with uncertain payoffs reached its highest recorded level this week, with a z-score of 3.45, up 0.17 from the prior week.

The Oracle Corporation debacle earlier this month served as a stark preview of what happens when investor enthusiasm collides with financial reality. The enterprise software giant shed roughly $80 billion in market capitalization in a matter of days after its AI infrastructure plans revealed approximately $100 billion in debt with bonds trading at spreads more commonly associated with junk-rated issuers. BCA Research has warned that markets may be approaching what they term a "Metaverse Moment," a phase where investors begin actively punishing companies for raising capital expenditures rather than rewarding them for ambition.

Yet the scale of committed spending – and the stories we are telling about it - continue to expand. Perscient's semantic signature measuring media emphasis on the sheer magnitude of AI infrastructure investment remained elevated at a z-score of 2.20. What began as projections of $250 billion in AI-related capital expenditure for 2025 has now ballooned past $405 billion, with Goldman Sachs analysts projecting that AI companies may invest more than $500 billion in 2026. The hyperscaler buildout shows few signs of slowing despite mounting questions about returns.

The comparison to previous technology bubbles persists in media coverage. Perscient's semantic signature tracking parallels between current AI spending and the fiber optic overbuilding of the dot-com era remained at 2.92, indicating that journalists and analysts continue to draw uncomfortable historical analogies. Mackenzie Investments published analysis exploring whether the world's race to deploy $7 trillion toward AI infrastructure might exceed near-term demand, echoing concerns that defined the telecom bust of 2000-2001.

The tension between bullish capital projections and skeptical interpretations defines the current moment. Morgan Stanley has noted that the AI capex boom powering U.S. equity gains for nearly three years may be entering its later phases, with major AI cloud providers experiencing declining free cash flow growth. Meanwhile, Menlo Ventures' market data tells a more optimistic story, showing enterprise AI spending surging from $1.7 billion to $37 billion since 2023 and now capturing 6 percent of the global SaaS market. Language reflecting growing resistance to big AI capital projects rose to a z-score of 1.37, up 0.10 from the prior week, suggesting that opposition framings are gaining modest traction even as spending accelerates.

Enterprise Skepticism Remains Despite Agentic AI Advances

As skepticism about infrastructure spending has intensified, so too has skepticism about the practicality of autonomous agents capable of executing multi-step tasks without constant human supervision. Perscient's semantic signature tracking media focus on agentic behavior as a pathway to LLM improvement declined by 0.25 from the prior week. Outside of narrative land, however, agentic uses for AI continue to be a topic of debate for large-scale enterprise deployments.

What distinguishes the current moment from previous years of agent hype is that practical implementations have actually arrived at scale. KPMG's second quarter 2025 survey found that 33 percent of organizations had deployed AI agents, representing a threefold increase from the 11 percent reported in the prior survey period. Deloitte predicts that by 2027, half of companies using generative AI will have launched agentic AI initiatives, with some applications seeing actual workflow adoption this year.

The capability improvements underpinning this adoption are substantial. Anthropic's latest  Sonnet releases now reliably complete tasks that would take humans approximately one hour, while frontier models from OpenAI and Google achieve near-perfect success on tasks requiring less than four minutes of human effort. The November and December model releases represent a transition from question-answering systems to autonomous agents capable of planning, executing, and iterating on complex multi-step projects. Microsoft has described this shift as moving from systems of record to systems of action, where AI agents go beyond support to help interpret signals, uncover patterns, and initiate actions.

The market projections reflect this momentum. The global agentic AI tools market is experiencing a compound annual growth rate of approximately 56.1 percent, reaching an estimated $10.41 billion in 2025. MarketsandMarkets forecasts expansion from $7.06 billion in 2025 to $93.20 billion by 2032. Gartner projects that by the end of 2026, 40 percent of enterprise applications will include task-specific AI agents, compared to less than 5 percent in 2025.

These deployment figures connect directly to questions about workforce transformation. Perscient's semantic signature tracking language describing AI's transformative potential for white-collar work remained at 0.30. A Stanford and Carnegie Mellon study found that while AI agents alone operate 88.3 percent faster and cost 90 to 96 percent less than human workers, their output quality lags behind. Hybrid human-AI workflows boosted overall performance by 68.7 percent, suggesting that the near-term future involves augmentation rather than replacement.

US AI Exceptionalism Framing Faces Growing Scrutiny

The slow and skeptical enterprise adoption of agentic AI and the similarly skeptical acceptance of the new normal of capex ad infinitum are part of the backdrop to a broader framing of artificial intelligence as central to American competitive advantage. Perscient's semantic signature tracking language positioning AI as a pillar of US exceptionalism rose to 1.26, up 0.38 from the prior week, reflecting heightened media coverage positioning the technology as foundational to national power.

A December White House executive order signed by President Trump stated that "United States leadership in Artificial Intelligence will promote United States national and economic security and dominance across many domains," explicitly noting the country is "in a race with adversaries for supremacy." The order, which lacks the force of law, also creates a taskforce whose sole responsibility will be challenging state-level AI regulations, signaling an administration priority on removing domestic barriers to AI development.

Standard Chartered analysts wrote that "the U.S. is best placed to benefit from any AI-driven productivity boost," suggesting the technology could support the economy, the dollar, and American exceptionalism. J.P. Morgan similarly noted that "U.S. exceptionalism appears to be the consensus view in 2025, underpinned by strong fundamental factors including tech innovation and the broadening AI cycle." Time magazine's decision to name the "Architects of AI" as its 2025 Person of the Year reflects how thoroughly the technology has permeated American life and discourse.

Yet the exceptionalism framing faces growing challenges. Gavekal Research's Louis-Vincent Gave has argued that "the days of US exceptionalism are fading," contending that "the US-China trade war is over and that for all intents and purposes, China won." Standard Chartered itself cautioned that "if AI and other technologies sputter, the U.S. would likely be just another G10 economy with mediocre demographics, elevated external debt and a poor fiscal position."

The competitive dynamics remain contested in media coverage. Perscient's semantic signature tracking language associated with predictions of American AI victory remained at 0.65, up 0.10 from the prior week, while the corresponding signature for Chinese victory stayed at 0.70. David Sacks, the incoming Trump AI czar, has begun to worry some in the technology industry, with lobbyists fearing he could be derailing the industry's agenda rather than advancing it. Language emphasizing the imperative that America must win the AI race remained flat at 0.54, while Cold War framings of AI competition held steady at 0.35, indicating stable but persistent geopolitical framing of technological development.

Archived Pulse

December 16, 2025

• Hyperscale Infrastructure Spending Defies Persistent Skepticism
• AI Productivity Gains Show Early Signs of Reaching Corporate Bottom Lines
• US-China AI Competition Intensifies as Both Nations Pursue Divergent Strategies

December 09, 2025

• OpenAI's "Code Red" Signals Intensifying Competition in the AI Race
• AI Capex Skepticism Grows as Hyperscale Builds Face Mounting Questions
• AI's Role in Loneliness Draws Intensifying Scrutiny

December 01, 2025

• Investor Skepticism Mounts Over an AI and Data Center CapEx…Bubble?
• Anthropic Emerges as Competitive Force as AI Race Narrative Intensifies
• The Death of Consulting Narrative Just Won’t Die

November 24, 2025

• Mounting Skepticism Over AI Infrastructure Spending Reaches New Heights
• AI Race Leadership Narrative Shifts as Google Gains Ground
• AI Mental Health Applications Draw Heightened Scrutiny

November 17, 2025

• Hyperscale Infrastructure Investment Narratives Accelerates Despite Growing Scrutiny
• Power Generation Capacity an Increasing Focus of Infrastructure Narratives
• Consulting Industry Faces Transformation from AI Adoption

November 10, 2025

• Hyperscale AI Infrastructure Spending Reaches Historic Highs
• AI Race Geopolitics Intensifies Between US and China
• Mental Health Risks from AI Chatbots Intensify Regulatory Scrutiny