The Gulf - buying the complement you already own#
Contents
The Gulf states are the purest test of this document's central economic claim, because they are deliberately buying the thing Game 3 identifies as valuable and they have an unusual amount of it already.
Why the position is genuinely strong#
Reread the inelastic-complement list from a Gulf perspective:
| Complement | Position |
|---|---|
| Energy | Abundant, cheap, and domestically controlled - the constraint that binds everyone else first |
| Land with power | Effectively unlimited, and adjacent to the generation |
| Permitting speed | Measured in months where the US measures in years - the constraint that actually binds |
| Capital | Sovereign, patient, and not subject to commercial discipline |
| Strategic non-alignment | Able to transact with both blocs, and courted by each |
Four of the five are the exact inputs the US is short of. The scarce input in AI infrastructure from ~2026 is not talent or capital - it is powered, permitted land, delivered quickly. That is the Gulf's endowment, and it is not a coincidence that capital has flowed accordingly.
The sovereign-capital point from Capital matters here more than anywhere: a state buying strategic optionality does not stop when the NPV goes negative. In a correction, commercially-financed capacity gets culled and sovereign-financed capacity does not. The post-correction map is more Gulf-weighted than the pre-correction one.
The conversion problem#
Owning capacity is not the same as having an economy, and this is where the strategy is genuinely untested.
Datacenters employ very few people. A gigawatt campus is a multi-billion-dollar asset with a workforce in the hundreds. As a diversification strategy for economies whose stated goal is employing a young, growing national population, compute capacity is close to the worst possible asset class on the employment metric - it is capital-intensive, labor-light, and skill-concentrated.
The bet is therefore not really about the datacenters. It is that hosting the capacity anchors an ecosystem - research institutions, applied AI firms, and the domestic absorption that the hub page calls the third axis. That is the same bet many regions have made on anchor industries, and the base rate on it is mixed at best.
the Gulf converts capacity into strategic position and financial return with high probability, and into broad domestic economic transformation with much lower probability - call it ~30% by 2035. The capacity bet is sound; the diversification bet is the one carrying the risk, and they are being reported as the same bet.
The dependency that isn't discussed#
Gulf compute capacity runs on imported accelerators, subject to export licensing by a third party. That is a real and acknowledged constraint, and it has shaped the deals.
The less-discussed version: the capacity is also dependent on demand it does not control. A datacenter is worth what its compute sells for, and that price is set by the model developers and their customers - none of whom are local. If inference costs collapse faster than demand grows in a particular period, the asset is stranded regardless of how well it was built.
This is the standard commodity-infrastructure risk and the Gulf has run it before with hydrocarbons. The difference is that oil in the ground does not depreciate on a five-year schedule and an accelerator does. Trading a non-depreciating reserve asset for a rapidly-depreciating one is a real strategic change, and it is being made largely on the argument that the alternative - holding the reserve asset into a decarbonizing world - is worse.
That argument may well be right. It is worth noticing that it is an argument about the hydrocarbon outlook rather than about the AI outlook.
The version where it works#
The strongest case for the strategy is not the ecosystem bet at all, and it is worth stating properly because the region's own communications underplay it. Compute is a way of exporting energy. A gas molecule sold abroad is worth the delivered commodity price minus liquefaction and shipping; the same molecule burned locally to serve inference is worth whatever the token is worth, with the margin captured domestically and no pipeline politics attached. Framed that way the Gulf is not diversifying away from hydrocarbons at all, it is moving up the value chain of the thing it already has, which is a far more conventional and far more achievable industrial strategy than becoming a research hub.
The constraint on that framing is workload geography. Exporting energy as tokens requires demand that tolerates the latency and the jurisdiction: training runs and batch inference do, interactive consumer products serving distant users mostly do not, and regulated workloads in health, finance, and government are frequently prohibited from leaving their home jurisdiction regardless of latency. So the addressable share of global compute demand is real but bounded, and it is precisely the share with the least pricing power, since batch training capacity is the most fungible product in the stack. Being the low-cost supplier of the most commoditized layer is a durable business, not a change of national position, which is roughly what the assessment above already says.
Failure mode: if inference demand grows faster than efficiency lowers cost per token, and it stays economical to serve globally distributed users from a small number of very cheap-power locations, then the bounded-workload objection is wrong and the region's capacity becomes systemically important rather than merely large. → Inference economics
The labor contradiction inside the ecosystem bet#
The diversification story requires a domestic ecosystem, an ecosystem requires researchers and engineers, and the region's answer to every previous skills shortage has been to import the labor on temporary terms. That worked for construction and services because those workers were interchangeable and the arrangement's unattractiveness did not affect supply much. It works less well here: the scarce input has options and opinions about where it works, and the specific thing it tends to want - long-horizon residence, institutional permanence, portable social status - is the thing the political settlement has been most reluctant to grant. Several states have moved on exactly this in recent years with long-term residence schemes, which is evidence they understand the problem. Whether it converts is the open question, and it is a question about political economy rather than about capital.
The employment arithmetic makes the same point from the other side. If the datacenters themselves employ hundreds and the ecosystem is staffed by imported specialists, then the strategy's contribution to national employment runs entirely through second-order effects, which are the hardest thing to forecast and the easiest thing to announce.
What to watch#
| Signal | Reading |
|---|---|
| Share of regional capacity contracted to foreign hyperscalers versus serving domestic workloads | Whether capacity is an export good or an ecosystem anchor |
| Long-term residence and equity-ownership grants to technical staff | Whether the labor contradiction is being resolved |
| Domestic AI firms with revenue from outside the region | The ecosystem bet converting, or not |
| Accelerator export-licence terms and their renewal cadence | The dependency that is acknowledged |
| Sovereign-fund capacity commitments during a compute drawdown | The counter-cyclical claim above, tested |
Why it matters to everyone else#
Three consequences that reach beyond the region:
- It breaks the compute-governance model. Game 2 treats accelerator tracking as the only plausible verification lever. Large capacity in jurisdictions outside the primary regulatory blocs, financed by states rather than firms, is exactly the configuration the lever handles worst. → C3
- It confirms the decoupling of capacity from capability. Energy predicted that countries able to build power fast gain structural advantage independent of research talent. The Gulf is that prediction happening in real time and is the cleanest available evidence for it.
- It changes what a correction looks like. Sovereign capacity persists through a drawdown. The 2030s compute map is therefore more state-adjacent and more geographically dispersed than a straight extrapolation of 2026 market share implies.
The labor contradiction is the binding long-run risk#
Capacity can be built with foreign EPC and imported operators for a decade. An ecosystem cannot: it needs long-horizon residents, universities that produce more than guest workers, and firms whose customers are not only the sovereign. The signals table above (residence grants, domestic firms with external revenue) is therefore not a nice-to-have social indicator - it is the difference between "power export with GPUs" and "third pole of the industry." Money and gas buy the first; only institutions buy the second. A Ground pass that only counts MW of announced campuses is scoring the easy half.
Failure mode of this page: treating every campus ribbon-cutting as confirmation of the ecosystem bet. Contracted export capacity to hyperscalers can grow while the ecosystem bet fails quietly.
Sovereign capex through a drawdown is the counter-cyclical claim to test. If Gulf capacity commitments hold when commercial labs cut, the post-correction map is more state-adjacent as capital argues. Watch commitments during a credit event, not only during boom announcements.
Related: Energy · Capital on sovereign financing · Game 2 - Nations · US and China · Uncertainty 2