Your newsletter digest — May 28, 2026
Today's digest covers two beats from your inbox: the AI infrastructure race—from SpaceX's orbit-or-bust compute thesis to the fracturing GPU era—and what product leaders need to rethink on AI freemium and monetization, plus a curated reading list for builders.
Every morning, the newsletters that landed in your inbox get folded into one read. Today's digest spans two beats: how the AI compute stack is reshaping every part of the software and hardware industry, and what product leaders should actually be doing differently as a result.
AI infrastructure
The SpaceX IPO and why data centers might end up in orbit
Ben Thompson's latest essay 1 opens with a confession about Uber Black and Model Y before landing on a more serious question: can SpaceX's $2 trillion IPO valuation ever make sense? His answer is nuanced. The current financials are absurd — $18.7 billion in revenue, $4.9 billion in losses, and a $26.5 trillion AI TAM claim that's laughably imprecise. But the underlying thesis — that demand for compute could eventually exhaust every buildable terrestrial site, making orbit the only remaining option — is at least plausible.
Three key points:
- Space racks are more feasible than they look. A single Starlink-sized satellite already has the rough physical dimensions of an NVL72 GPU rack. Power and heat dissipation remain the engineering hurdles, but neither is categorically unsolvable.
- The right workload is agentic inference, not training. Agents running overnight jobs without a human in the loop don't need low latency — they need cheap, abundant compute. Slower chips in orbit would be tolerable; the economics could work.
- Zoning is the stick. Thompson notes that communities now have veto power over terrestrial data centers in a way they never had over factory closures. As opposition grows, the cost advantage of earth-based compute shrinks — creating a genuine opening for the alternative.
Loading content card…
The inference shift: why GPU dominance is not inevitable
Earlier in May, Thompson published a longer structural piece 2 arguing that the AI compute story is splitting into three distinct markets: training (Nvidia-dominated, unlikely to change), answer inference (where speed wins and chips like Cerebras have a case), and agentic inference (the biggest long-term market, where memory hierarchy beats raw compute throughput).
Three key points:
- Agentic workloads tolerate latency. When an agent runs a multi-hour task with no human waiting, latency stops being the constraint. What matters is capacity — large context, persistent state, cheap DRAM — not peak token speed.
- AWS's decade of "inferior" chip investment is paying off. Trainium was mediocre for years. It's now decent, and the trajectory is favorable. AWS's Bedrock runs on it silently, just as Graviton powers RDS without customers choosing it — the same playbook, replayed for AI.
- Cerebras has a real but narrow window. Its wafer-scale chip has 6,000× the memory bandwidth of an H100, which makes it fast for answer inference. But the moment a workload needs more memory than fits on-chip — a larger model, a growing KV cache — the economics collapse.
Product and growth
Why your AI freemium strategy is probably wrong
Vikas Kansal writing in Lenny's Newsletter 3 makes the straightforward observation that every AI user interaction costs GPU compute — which means "add free users to reach product-market fit" destroys unit economics in a way it never did for SaaS.
Three key points:
- Gate intensity, not intelligence. Rather than paywalling smarter models, gate volume. Google's AI tiers give each level higher usage limits and larger context windows — prepaid pricing that maps costs to actual compute usage.
- Sell outcomes, not answers. Intercom Fin charges per resolved support ticket, not per API call. Google's autonomous browsing agent that completes web tasks is paywalled to higher tiers — customers are buying eliminated labor, not model access.
- Reserve the compute-heavy modalities. Cinematic video generation, real-time simulation, persistent 3D environments — these naturally belong behind a premium wall. Consumers already expect them to cost more; the paywall doesn't feel punitive.
Loading content card…
36 books every product builder should read
Lenny Rachitsky's May 26 piece 4 is a curated reading list organized by the specific gap it fills rather than by prestige or hype. Most picks are over 10 years old — he explicitly rejects novelty as a selection criterion.
Three key points:
- Communication first. Nobody Wants to Read Your Sht*, On Writing Well, and Storyworthy ranked ahead of any strategy or PM frameworks. The argument: as AI lowers the cost of building, distribution becomes the scarce resource — and distribution starts with writing.
- Strategy books that actually teach strategy. Good Strategy/Bad Strategy and Playing to Win made the list; most other strategy books didn't. Rumelt's framework for diagnosing bad strategy (fluff, failure to face the problem, mistaking goals for strategy) remains the clearest test.
- Management classics over management fads. High Output Management (1983), Radical Candor, The Making of a Manager — nothing on AI management or remote work hacks. The message: timeless > timely.
One thread to watch
Thompson's data center piece and Kansal's AI monetization piece converge on the same underlying pressure: compute is expensive and finite, which forces everyone — from SpaceX investors to product managers — to think harder about what workloads justify the cost. The companies that figure out how to charge for outcomes (not inputs), run the right workloads in the right places, and treat compute as a scarce resource rather than a commodity are the ones building durable businesses right now.
Sources: Stratechery by Ben Thompson · Lenny's Newsletter
Add more perspectives or context around this Drop.