Apple Bets Against the AI’s ‘Illusion of Thinking’

In the summer of this year, four researchers from Apple, published a paper titled. “The Illusion of Thinking: Understanding the Strengths and Limitations of Reasoning Models via the Lens of Problem Complexity.” The findings of the study ran rather contrary to the hype around AI’s ‘reasoning model’. The researchers found that “frontier LRMs face a complete accuracy collapse beyond certain complexities. Moreover, they exhibit a counter-intuitive scaling limit: their reasoning effort increases with problem complexity up to a point, then declines despite having an adequate token budget.”

This perhaps explains why Apple has been a ‘deliberate laggard’ in the AI race between the Magnificent Seven, rather it is following a ‘second-mover’ strategy to not just stay in the race, but out compete its rivals.

Apple's Methodical AI Strategy

Apple's perceived lag in AI is attributed to its commitment to shipping features only when they meet exacting standards for user experience and privacy. The paper exemplifies this philosophy by rigorously stress-testing reasoning models rather than rushing to market with unproven technology. While competitors advanced quickly with LRMs like o1 and Claude 3.7 Sonnet, Apple researchers methodically demonstrated these models fail at exact computation, struggle with algorithmic reasoning, and show inconsistent performance across complexity scales.

The company is prioritizing capital discipline, integration into its ecosystem, and privacy over an allout infrastructure arms race. This approach may help it avoid the most extreme risks of today’s AI capex boom, but it also carries a real risk of Apple falling behind if rivals convert massive spending into compelling consumer products and platforms faster than expected.

The AI capex arms race

The Magnificent 7—especially Microsoft, Alphabet, Amazon and Meta—have entered an unprecedented AI infrastructure spending cycle, committing well over US$200–300 billion annually across data centers, GPUs, and networking over 2024–2025. Estimates for 2025 alone suggest these firms could collectively exceed US$300 billion in AIrelated capex, with Microsoft guiding around US$80 billion, Alphabet around US$75 billion, and Meta US$60–65 billion, much of it explicitly justified as investment in AI. Financial analysts also flag the likelihood of “capex indigestion,” where investors must absorb years of heavy infrastructure spend before clear profit pools emerge.​

Apple’s comparatively restrained spending

Apple is the clear outlier on AI infrastructure intensity. In fiscal 2024 it spent about US$9.4 billion on capital expenditure, roughly 2% of revenue—far below the tens of billions each that Microsoft, Alphabet, Amazon and Meta are directing into AI. Even with analyst expectations that Apple capex will rise to roughly US$12–14 billion in its current fiscal period, it remains a fraction of hyperscaler levels in absolute terms and as a share of cash flow.​

Apple’s CFO has emphasized that the company is increasing AI investment but is doing so through a “hybrid” model, relying more on external cloud compute and classifying much AI cost as operating expense and R&D rather than locking into massive long-lived datacenter assets.

Strategic logic of being a second mover

Analysts point out that Apple has previously won by entering categories late—smartphones, tablets, smartwatches—but with superior integration, design, and user experience that converted a late start into dominant share.​

In AI, Apple’s strategy centers on; personal, ondevice and “private cloud” intelligence rather than cloudonly chatbots; deep integration of models into iOS, macOS and Siri to create differentiated device experiences, not standalone AI services chasing pure usage metrics; a focus on privacy and user trust as a moat, positioning “Apple Intelligence” as safer and more personal than datahungry competitors.​

By watching how generative AI business models, regulation, and usage patterns evolve, Apple can deploy capital more selectively, avoiding the risk of overcommitting to architectures, vendors, or revenue models that may not endure. This allows it to conserve balancesheet flexibility while others absorb the early technical and regulatory shocks.

The risk of falling seriously behind

The core strategic question is whether this discipline becomes complacent. Wall Street coverage increasingly notes concern that Apple has a narrowing window to clearly articulate and deliver an AI strategy compelling enough to prevent users from drifting toward rival ecosystems perceived as more capable or innovative. For much of 2025, Apple underperformed AIhypedriven peers before an “AI catchup” narrative began to take hold as it rolled out Apple Intelligence, highlighting how quickly sentiment can shift with perceived AI leadership.​