Cramer Says Send the Data Centers to Mexico Instead
Jim Cramer floated a two-part workaround for data center operators blocked by American zoning boards, with Mexico as one destination — and warned the cheaper path usually beats the smarter one.
On August 26, 2026, CNBC's Jim Cramer proposed a two-part escape route for data center operators facing US zoning board resistance — one part of which is siting facilities in Mexico — and predicted the cheaper option would win out over the smarter one.
Jim Cramer's answer to the growing local revolt against data center construction in the United States is blunt: if American towns don't want them, put them somewhere that does. The CNBC host laid out a two-part escape plan for operators tangled in zoning fights, one half of which involves building in Mexico. His own prediction about which half prevails is the more interesting part — Cramer expects the cheaper option to beat the smarter one.
That framing, reported by 24/7 Wall St, lands on an industry that has spent two years discovering that the binding constraint on artificial intelligence capacity is not chips or capital. It is a county commission meeting.
Zoning boards became the choke point
For most of the cloud era, siting a data center was an engineering and tax-abatement exercise. Find cheap power, find fiber, negotiate an incentive package, break ground. The AI build-out changed the arithmetic. Facilities got bigger, their electricity draw got larger, and their water consumption for cooling became a line item that neighbors could see on their own utility bills.
The result is a permitting environment where individual projects can stall for reasons that have nothing to do with the balance sheet behind them. A hyperscaler can have the capital, the chips on order and the transmission interconnection queued, and still lose eighteen months to a rezoning hearing. When your competitor is racing to stand up compute, that delay is the whole game.
Cramer's implicit point is that capital doesn't sit still for local politics. It relocates. The question is only where.
Why Mexico shows up in the conversation
Mexico is the obvious candidate on a map. It is adjacent, it is inside the North American trade framework, and it has industrial corridors already built around cross-border manufacturing. Latency to US population centers from northern Mexico is a fraction of what it would be from Asia or Europe, which matters enormously for inference workloads that serve American users in real time — though far less for model training, which can happen anywhere with enough electricity.
That distinction is the crux of any offshoring argument. Training runs are batch jobs: they tolerate distance. Serving a chatbot response does not. An operator moving capacity south of the border is implicitly deciding which of those two workloads it is willing to relocate.
The harder questions are the ones the map doesn't answer. Grid capacity, generation mix, water availability in arid northern states, and the reliability of long-term power contracts all determine whether a site pencils out. And permitting resistance is not a uniquely American phenomenon — communities everywhere object to industrial facilities that consume large volumes of water and electricity without hiring many local people. Exporting a project does not automatically export the objection.
The cheaper option versus the smarter one
Cramer's forecast — cheaper wins over smarter — is the part investors should sit with. It is a familiar pattern in capital-intensive industries under time pressure. When the strategic clock is running, management teams optimize for the option that gets steel in the ground this quarter, not the one that produces the best asset in a decade.
Cramer's forecast — cheaper wins over smarter — is the part investors should sit with.
Applied to data centers, the cheap-versus-smart trade-off shows up in several places at once:
- Site quality versus site speed. A location with abundant firm power and a cooperative regulator may take longer to secure than one with adequate power and a permissive local government.
- Owned generation versus grid dependence. Building dedicated power is expensive and slow; plugging into an existing grid is fast and exposes the operator to future rate and curtailment risk.
- Domestic versus cross-border. A US site carries known legal and tax treatment. A foreign site introduces currency, sovereign and counterparty variables that a spreadsheet can price but a board may underweight when the alternative is losing the race.
The market's discipline on these choices is loose right now, which is precisely the concern. When capital is abundant and the narrative rewards announced capacity, the incentive is to announce.
What the tape says while the argument plays out
The broad market was not moving on any of this. As of the last trade at 15:27 GMT on August 26, 2026, the SPDR S&P 500 ETF Trust (NYSEARCA: SPY) was at $765.83, down 0.01% from the prior close of $765.91, inside a day range of $764.68 to $766.96. The Invesco QQQ Trust (NASDAQ: QQQ), the tech-heavy proxy where the AI infrastructure trade concentrates, sat at $710.11, off 0.09% against a $710.72 close, ranging $707.97 to $712.20. The SPDR Dow Jones Industrial Average ETF (NYSEARCA: DIA) was the softest of the three at $534.19, down 0.20% from $535.24.
In other words: a flat tape. Commentary about where the next tranche of compute gets built is a multi-year structural question, not an intraday catalyst, and the indexes are pricing it that way. That is worth noting, because siting decisions made under time pressure in 2026 will show up in returns on invested capital several years out, long after the headlines that prompted them have been forgotten.
What to watch from here
The tell will not be in rhetoric. It will be in disclosures. Watch for capital expenditure guidance that breaks out international siting, for interconnection agreements filed with foreign utilities, and for lease commitments from the third-party colocation operators that hyperscalers increasingly rent from rather than build themselves. Those specialists absorb the permitting risk on behalf of their tenants, which makes them the first place a shift in geography becomes visible.
Watch, too, for the political response. Data center construction has become a domestic jobs-and-investment talking point in the United States. Any visible migration of that capital across the southern border invites a policy answer — incentives to keep it home, or friction to make leaving costly. The zoning board is a local actor; the response to operators routing around it may not be.
Cramer's provocation is really a stress test of a claim the industry has made repeatedly: that American AI infrastructure must be built in America. If the permitting math keeps getting worse, that claim gets tested by the cheapest available alternative rather than the best one.
Key facts
- S&P 500 (SPY): $765.83, -0.01%, as of 15:27 GMT Aug 26, 2026
- Nasdaq 100 (QQQ): $710.11, -0.09%, as of 15:27 GMT Aug 26, 2026
- Dow 30 (DIA): $534.19, -0.20%, as of 15:27 GMT Aug 26, 2026
- Cramer's proposal: Two-part plan to bypass US zoning boards; one part sites data centers in Mexico
Frequently asked questions
What exactly did Jim Cramer propose?
Cramer laid out a two-part plan for data center operators to get around American zoning boards that are blocking or delaying new projects. One part of the plan involves locating facilities in Mexico. He also predicted that of the two options, the cheaper one would most likely win out over the smarter one.
Why are data centers facing local opposition in the United States?
AI-scale facilities draw large volumes of electricity and, for cooling, water, while employing relatively few people once operational. That combination makes them a target at county and municipal permitting hearings, where residents raise grid strain, water use and land impact. The result is that zoning approval, not capital or chips, often sets the construction timeline.
Would moving data centers to Mexico actually work?
Geographically it is plausible: northern Mexico is close enough to US population centers to keep network latency low, and it sits inside the North American trade framework. But the economics depend on grid capacity, generation mix, water availability and long-term power contract reliability. Local permitting resistance also is not unique to the United States.
Does the type of AI workload change the offshoring calculus?
Yes, substantially. Model training is a batch process that tolerates distance from end users, so it can be relocated with little performance penalty. Inference — serving live responses to users — is latency-sensitive and benefits from being physically close to the customer base. An operator moving capacity abroad is effectively choosing which workload it will relocate.
How did markets react on the day?
They did not move on the commentary. As of the last trade at 15:27 GMT on August 26, 2026, SPY was $765.83, down 0.01%; QQQ was $710.11, down 0.09%; and DIA was $534.19, down 0.20%. Data center siting is a multi-year structural question rather than an intraday market catalyst.
What signals would confirm capacity is actually moving abroad?
Look at disclosures rather than commentary: capital expenditure guidance that separates out international siting, interconnection agreements filed with foreign utilities, and lease commitments taken by third-party colocation operators. Those specialists absorb permitting risk for their tenants, so a geographic shift typically shows up in their filings first.
Sources
Photo: Adrien Olichon · Pexels Licence — source
