Apple’s AI Restraint Turns Defensive Edge, But Valuation Stretches

Apple’s cautious AI investment stance has turned from a perceived weakness into a market advantage as investors tire of Big Tech’s heavy AI spending. The stock rebounded sharply in the second half of 2025, lifting Apple to a ~$4.1T valuation and a top S&P weight, with investors betting it can monetize AI through devices and services without massive capex. Yet the shares are expensive, Berkshire trimmed its position, and technicians see a possible pullback despite a supportive long-term trend.

Key Points

• Apple’s restrained AI strategy has become a strength as investors question massive AI capex across Big Tech.
• The stock fell 18% through June 2025 but then rallied 35%, outpacing AI favorites; Apple’s market cap is about $4.1T and its S&P 500 weight is now second-largest.
• Apple is seen as an “anti-AI” defensive play that can still benefit from AI via devices and services when adoption broadens.
• Valuation has stretched to ~33x forward earnings (well above its long-term average), raising concerns about overpaying for safety.
• Berkshire trimmed its Apple stake by 15% in Q3; technicians warn of a possible near-term pullback despite a bullish long-term trend.

Sentiment

The community is skeptical and divided. Many commenters feel the 'success' claim is misleading because the primary implementation uses the screenshot as a background image rather than genuinely recreating the site. There is genuine appreciation for the autoformalization insights and the Playwright approach, but the overall tone leans critical — particularly around the implementation quality and the 'one-shot' framing. A meta-debate about anti-AI bias on HN runs through the thread but doesn't resolve cleanly.

In Agreement

• The Playwright + screenshot pixel diff approach is genuinely clever and validates the idea that proper tooling enables AI to accomplish tasks that naive prompting cannot.
• The autoformalization lesson is valuable — 'Optimizers will optimize your objective function, not your intended objective function' — and is an important caution for anyone using LLMs for agentic tasks.
• Claude with proper configuration (Nori) required only minimal human attention (~5 minutes), demonstrating the efficiency gains from well-structured AI tooling.
• The index_tiled.html version (using original assets without the screenshot background) correctly positions elements and is as close to the original as scaling/compression artifacts allow.
• If you know how to prompt and have the right tools, this type of visual layout task is genuinely achievable in one session.

Opposed

• The main hosted version (index.html) uses the original screenshot as a CSS background image with invisible anchors overlaid — it didn't actually recreate the site, it just pasted the screenshot as wallpaper.
• The 'one-shot' framing is misleading because the effective prompt was only possible after learning from the previous day's failed attempt, making it an iterative process across multiple sessions.
• A developer familiar with 90s-style table-based HTML could have done this manually in roughly the same time, making the exercise of limited practical value.
• The recreation lacks responsive design, proper zoom behavior, and window-resize handling — it only matches at exact screenshot dimensions, not a faithful recreation of the original's design intent.
• The post reads as advertising for the author's Nori tool rather than a substantive demonstration of AI capability.
• The original Space Jam website creator commented that they were confused why Claude was needed for such basic HTML/CSS.