prefshares

Nike's ex-CEO just explained how his own strategy backfired. He measured the org on digital growth. So the org optimized: it held Nike's scarcest inventory back from retail partners to feed the online channel. Wholesale read it as abandonment. Competitors took the shelf space. Part of the digital growth was never demand. It was supply, steered to wherever the scoreboard pointed.

We're extending Claude Fable 5 access on all paid plans, as well as keeping Claude Code’s weekly rate limits 50% higher, through July 19.

Extreme poverty worldwide has never been as low as it is today, and it's never fallen as fast as in recent decades. We need more facts and less ideology...

The US economy is now dependent on AI spending: AI investment now accounts for more than 25% of US GDP growth, the largest contribution on record. This includes spending on software, IT equipment, R&D, and data centers. In other words, for every $4 of US economic growth today, over $1 is coming from AI investment. This comes as AI spending is up to a record ~8% of US GDP. By comparison, spending on IT equipment, software, and R&D peaked at ~6.5% of GDP during the 2000 Dot-Com bubble. US economic growth is now all about AI.

Normalize salaries as a percentage of company earnings

The economics behind AI infrastructure may be even more attractive than the economics of training frontier models themselves. This analysis estimates that SpaceX’s compute contracts generate roughly $15 billion of annual revenue from Anthropic and another $11 billion from Google, producing EBIT margins between 70% and 83%, ROIC ranging from 70% to well above 120%, and capital payback periods of less than eighteen months. Even allowing for estimation error, the conclusion is difficult to ignore: in a supply-constrained market, leasing AI compute may be one of the highest-return businesses in technology. The implications extend well beyond SpaceX. Meta’s latest Muse Spark 1.1 benchmark demonstrates that its AI models are now approaching the frontier. Muse Spark 1.1 scores 51 on the Artificial Analysis Intelligence Index, effectively tying GPT-5.4, GPT-5.6 Luna and GLM-5.2 while remaining among the most token-efficient and lowest-cost models in its performance tier. The intelligence gap between frontier labs continues to narrow, while inference costs continue to fall. That combination changes the economics of AI. As models become increasingly commoditized, competitive advantage shifts away from model intelligence alone and toward the infrastructure required to train and serve them. The real bottleneck is no longer algorithms. It is compute. If external developers are willing to pay economics similar to the SpaceX contracts, Meta’s enormous GPU fleet becomes more than an internal expense. It becomes an asset capable of generating infrastructure-like cash flows. Leasing even a portion of unused capacity at comparable pricing could produce tens of billions of dollars of incremental revenue with exceptionally high operating leverage, while simultaneously improving returns on the hundreds of billions of dollars already committed to AI capital expenditure. This is why Mark Zuckerberg recently described the SpaceX compute model as “quite interesting.” The strategic value lies in optionality. Meta can continue using its infrastructure to train increasingly competitive frontier models while retaining the flexibility to monetize excess capacity whenever external demand exceeds supply. Those two businesses are complementary rather than mutually exclusive. Internal AI development strengthens long-term competitiveness, while compute leasing generates immediate cash returns that help finance the next generation of infrastructure. This also reinforces a broader shift occurring across the AI industry. For much of the past three years, investors focused almost exclusively on benchmark leadership. Increasingly, however, AI is becoming an infrastructure business. The companies controlling GPUs, networking, power, cooling and data centers may ultimately capture as much economic value as the companies building the models themselves. The AI race is therefore evolving from a software competition into a capital allocation competition. As frontier models converge in capability, the scarce resource becomes compute capacity rather than intelligence. Companies that own and efficiently monetize that infrastructure may end up earning the most durable returns. In that sense, Meta’s greatest AI asset may not be Muse Spark itself. It may be the massive compute platform sitting behind it. Long Meta.