The Constraint Moved Again
Dylan Patel said the quiet part out loud this week: GPUs are no longer the biggest bottleneck in AI. CPUs are. Agentic loops (Codex running for seven hours, calling databases, scraping, spinning simulations) have eaten every spare cycle in the cloud. Microsoft sold its spare CPUs to Anthropic and OpenAI. Amazon tripled CPU installs year over year. GitHub has been flapping for weeks.
Look at SanDisk. Up 607% in six months. Storage. The part of the stack nobody wanted to talk about in 2024 is the part printing money in 2026.
This is the pattern. The constraint is always moving. And the person who sees the next bottleneck first gets paid, while everyone else is still optimizing the layer that stopped mattering.
Theory of Constraints, for Everything
Eli Goldratt's The Goal made this legible for factories. Every system has one bottleneck. Throughput of the whole system equals throughput of the tightest link. Any improvement anywhere except the bottleneck is fake progress.
Most people read it and think "supply chain." The actual scope is much wider. It applies to:
- Compute. GPU → CPU → storage → power → real estate → cooling → interconnect. The constraint migrates every twelve months and the winners migrate with it.
- Your business. Awareness → trust → offer → delivery → retention. Fixing the wrong link is theater.
- Your life. Sleep → nervous system → relationships → clarity → output. If your sleep is broken, no productivity app matters.
- Your mission. Vision → team → distribution → capital → operational cadence. Pouring capital on a weak team does not unlock the mission. It exposes it faster.
The constraint is the thing that, if it moved one inch, would unlock more value than everything else combined. Almost always, it is not what you are working on right now.
Why Most People Miss the Shift
Two reasons.
First, they fixate on the loud layer. GPUs are on every tech magazine cover. CPUs are not. The constraint is almost never the loud layer, because the loud layer is where everyone has already piled in. The loud layer is yesterday's constraint, now over-served and trading at commodity margins. Look for the layer that got called "boring" in the last cycle.
Second, they confuse optimization with leverage. Optimizing a non-constraint makes the non-constraint better. That is useless. It does not raise throughput. It just makes the bottleneck more obvious. Most people spend their careers optimizing non-constraints and wondering why nothing compounds.
The synthesis move that creates asymmetric edge is naming the new constraint before the market does.
How to Spot the Next Constraint
You do not need insider access. You need to read the signals most people dismiss as noise.
- Instability is a tell. When GitHub starts flapping, something upstream is saturated. When latency spikes across providers no one has ever heard of, something is getting squeezed.
- Price of the boring component. Memory, storage, power substations, transformer lead times, fiber builds. When the boring thing spikes, that is the new choke.
- Quiet moves by sharp operators. Watch where the best founders are hiring, where the best funds are allocating the "weird" part of their portfolio. They are positioning before the tweet cycle catches up.
- The reverse is also true. When something that used to be scarce becomes abundant, the constraint moved. Do not keep earning yesterday's premium.
What This Means for Where You Place Yourself
If you are building: put yourself next to the bottleneck, not downstream of it. Downstream of a bottleneck you wait in line. Adjacent to the bottleneck you get paid. This is why harness engineering has become the craft in 2026: the models are abundant, the code around the models is the constraint. Go to where the work is hard.
If you are investing your time or your capital: the highest return is on the layer that is about to become scarce. Last cycle it was GPUs. This cycle it is compute orchestration, power, storage, and the operators who can actually wield this stack. Next cycle it will be something you cannot see clearly yet. Train your eye.
If you are building a life: the constraint is usually not the thing you complain about. It is the thing you avoid. Sleep, bodywork, the hard conversation, the friend you have been meaning to call, the hour of prayer you keep skipping. Unlock that and everything downstream gets cheaper.
The Spiritual Layer
Discernment is the meta-skill of constraint-spotting. Seeing what others do not see. Naming what others have not named. The Holy Spirit is the world's most underpriced advisor on where the real bottleneck is, and most people never ask.
The prayer I keep praying: Lord, show me the constraint. Not the loud thing. The tight thing. The one link that, if it moved, would unlock the rest.
When He answers, the answer is almost always something I was avoiding.
The constraint is always moving. Your job is not to optimize the old one. Your job is to find the new one before the market does, and place yourself next to it.