Ben Thompson on tech strategy, Intel's structural failure, and why xAI would be a great acquisition target without X

Summary

Ben Thompson on Stratechery positions himself in a deliberate gap between product-focused tech journalism and Wall Street's financial analysis, covering strategy, culture, and business model dynamics. His subscription model, launched roughly two years after Stripe's 2011 billing product at $10 per month (now $15), prioritizes a large, low-ARPU base over a small high-paying clientele. The structure means no single subscriber holds meaningful leverage over his editorial output, which he views as a structural independence advantage rather than a moral posture.

Thompson publicly corrected his bullish call on Apple Intelligence, which he made following Apple's announcement last year. He argued at the time that the platform would give Apple pricing power over model makers. He now considers that wrong, writing this week that Apple will have to pay up for AI capabilities rather than dictate terms.

AI Talent Economics

The AI talent war represents a structural shift in employer bargaining power that Thompson frames as a meaningful bear case for tech margins during this period. For decades, tech companies benefited from opaque internal labor markets where critical employees were hard to identify, difficult to poach, and their skills were not easily transferable across firms. AI research breaks all three conditions simultaneously: model quality is measurable through benchmarks, the skills are transferable across Google, Anthropic, OpenAI, and Meta because they are all building toward the same objective, and the pool of capable researchers remains scarce.

Mark Zuckerberg paying reported $100 million bonuses is rational from Meta's position, Thompson argues, because AI is pure upside for a social media company with no existing business to cannibalize. But those compensation signals cascade across the entire industry, raising costs for every competitor. Tim Cook's $75 million compensation creates an implicit internal ceiling at Apple that makes competing for top AI researchers structurally difficult, with some researchers two levels below Zuckerberg now reportedly earning more than Cook.

Google's Structural Resilience

Despite being the classic innovator's dilemma case study, Google has outperformed Thompson's post-ChatGPT expectations. The core argument is cultural: Google was founded to be an answer engine, and that mission is more aligned with large language models than with the 10 blue links and ad auction model that became its business. The "I'm Feeling Lucky" button, Thompson notes, persisted long after it became functionally useless, illustrating how deeply the original mission is embedded in the company's identity.

Google's search product response, using AI overviews and AI Mode as a testing layer before full deployment, mirrors what it did roughly 12 years ago when it overhauled search results pages to incorporate local and shopping results against vertical search competition. The search-to-AI funnel strategy is coherent, he argues, even if the company failed to launch a conversational AI product before OpenAI and produced no credible response for the first six to nine months after ChatGPT launched in late 2022.

Intel as a Structural Failure

Thompson's analysis of Intel dates to a 2013 Stratechery piece arguing the company was structurally doomed unless it built a foundry business serving external customers. The root cause was Intel's decades-long manufacturing-first, performance-maximizing culture, shaped by a foundational 1980s decision championed by Pat Gelsinger to stay on the x86 CISC architecture rather than switch to RISC, betting that manufacturing improvements would outpace any theoretical efficiency gains from switching.

That bet was correct and drove the PC era but embedded a performance bias that made Intel unable to compete in mobile, where efficiency was the dominant variable. The story that Paul Otellini turned down the original iPhone contract is, according to Tony Fadell, incorrect. Intel's ARM chips at the time were still optimized for performance rather than power efficiency, making them non-competitive regardless.

Missing mobile was fatal in slow motion. Fab costs that were roughly $500 million per facility in 2013 have risen to approximately $20 billion today, making volume essential to spread fixed costs. TSMC captured mobile volume, built scale, and is now the only manufacturer capable of meeting AI chip demand at scale. Intel's stock ran strongly for eight to nine years after Thompson's warning because cloud server chips masked the structural deterioration, but the AI chip cycle exposed it completely.

A foundry split is not a clear solution. Intel needs internal volume to justify its manufacturing base, and AMD's decade-ago separation from GlobalFoundries resulted in years of difficult negotiations and weak performance before AMD moved production to TSMC and rebuilt its chip design operation from scratch. Intel's foundry ambitions are also culturally misaligned: foundry is a customer-service business where manufacturing adapts to client chip designs, the opposite of Intel's historical model where design teams accommodated manufacturing constraints. Thompson's assessment is that Intel had it changed course in 2013, it would be positioned to capitalize on AI demand today. Instead, the company is a managed decline, extracting cash flow while the strategic window has closed.

The Microsoft Azure parallel is instructive. Azure did not dominate cloud, with AWS holding that position, but building the infrastructure capability positioned Microsoft to move quickly when AI demand emerged. Azure is now a significant AI infrastructure player. Intel had the equivalent opportunity and did not take it.