Primary Barriers to Entry (Relative Severity)
Estimated Share of Announced Projects That Will Not Be Built
Source: Kevin O’Leary statements (MarketBeat interview, October 2026)
Political & Regulatory Risk Timeline (2026–2030)
Data Center Barriers to Entry: Why Power, Permits, and Politics Will Decide the Winners Through 2030
Kevin O’Leary has been unusually blunt about the AI infrastructure boom. In a MarketBeat interview, the Shark Tank investor—who is himself developing large-scale data center projects—warned that 60–65% of announced data centers will never get built. The reason is not a sudden collapse in demand (he calls compute demand “insatiable” and roughly 3× current supply into at least late 2027). It is a combination of hard physical and regulatory barriers plus politics.
These constraints will shape the industry for the next four years, through the remainder of the current U.S. political cycle, the 2028 presidential election, and corresponding provincial and municipal decisions in places like Ontario. Sites that clear the barriers early will become scarce and valuable. Most announcements will not.
The Real Barriers to Entry
O’Leary’s list of what is required to succeed is straightforward and unforgiving. Missing any major item creates what he calls a “fatal flaw.”
- Land with the right attributes — preferably non-agricultural, within reasonable driving distance of a community for labor, but not so close that noise or visual impact triggers strong opposition (often cited in the 3–4 mile range).
- Water rights — even with modern closed-loop or air-cooled designs that use far less water than older facilities, legal access must still be secured.
- Dark fiber and low-latency connectivity — tenants (especially hyperscalers) require it; without it the site is incomplete.
- Power — the dominant constraint. Developers must typically bring their own generation (often natural-gas turbines behind the meter) rather than relying on a constrained grid that raises residential rates. Low-cost power (ideally well below 6¢/kWh) is a major differentiator. Pre-development work alone can run $20–30 million per gigawatt. Large ~1.4 GW projects have been described in the $16 billion range.
Ihttps://youtu.be/oLi2I-49PdU?si=t6c5ZtxeRzNsO0ak
Kevin O’Leary (Mr. Wonderful) argues that AI/data center demand is “insatiable” (roughly 3× current supply for compute capacity into at least Q4 2027 from hyperscalers, government, and private users), but 60–65% of announced projects will never get built due to fatal flaws or politics.Yahoo Finance
He frames data centers as a real-estate/infrastructure play (like specialized climate-controlled storage) where developers supply the powered shell, cooling, and path to operations; tenants (especially hyperscalers) bring their own racks/compute and decide the exact use (AI training/inference, cloud storage, etc.).
Key factors for successfully building and attracting tenants
Power is the primary bottleneck and value driver
• Developers must typically bring their own power (often natural-gas turbines / behind-the-meter generation) rather than relying on the grid, because grid interconnection is slow, politically sensitive, and can raise local rates.
• Lowest-cost power is a major competitive advantage (examples he cites: ~1¢/kWh in places like Saudi Arabia; sub-6¢ in hydro/nuclear-rich areas such as Norway, Finland, or Alberta with stranded gas).
• Tenants want reliable, low-latency power at scale; modern projects target multi-hundred-MW to multi-GW sizes (he references ~1.4 GW class projects).
• Power + dark fiber + path to permit is what makes a site valuable once many announced projects fail.YouTube
Infrastructure and site requirements (“fatal flaws” that kill projects)
• Land + water rights (modern closed-loop/air-cooled designs use far less water than older ones—he compares it to a golf course).
• Dark fiber / low-latency connectivity.
• EPA/air permits, state and federal approvals, and turbines/equipment delivery aligned with the timeline.
• Pre-development work alone can cost $20–30 million per gigawatt for engineering, permitting, and agreements. A large ~1.4 GW project is described as roughly $16 billion total ($8B equity / $8B debt range).
• Timeline pressure: tenants want a credible path to permit in ~18–24 months and operations around late 2027; longer delays destroy IRRs and financing.
Politics, community, and permitting risk (the biggest near-term killer)
• O’Leary expects 60–65% of announced projects to fail or be shut down in the next ~24 months from fatal resource gaps or political reasons.
• Midterms and local elections create incentives for moratoriums or opposition; many people oppose data centers near residences (70% in some framing), though support rises when projects create high-paying jobs ($150k examples) and tax revenue (he cites cases that could represent ~30% of a local budget).
• Location strategy: 3–4 miles from communities (close enough for jobs/taxes, far enough to reduce noise/visual/NIMBY issues), avoid prime agricultural land, and diversify across jurisdictions (he mentions exposure in multiple countries including U.S., Canada/Alberta, Norway, Finland).
• Community benefits and “bring your own power” approaches help win political and local support.
Attracting tenants (hyperscalers and others)
• Provide a ready path to power + shell + cooling + fiber; tenants roll in their equipment and decide the workload.
• Low latency, reliable scale, and political certainty matter more than pure announcements.
• Demand is broad (AI compute, cloud, storage, government), so permitted, powered sites become scarce and valuable once most announced capacity doesn’t materialize.
• Tenants carefully diligence local politics, governor/county dynamics, and exact power timelines.
Investment framing
He treats successful projects like commercial real estate/utilities: buy at higher cap rates, hold for cash flow and land/value appreciation, with target returns in the high-teens to low-20s range on a leveraged basis. Diversification across public vehicles (for liquidity) and private developments, plus picks-and-shovels exposure (power, turbines, etc.), is part of his approach. He remains bullish on the power + compute verticals for the next several years even amid stock-market volatility in the sector.Marketbeat
Bottom line from the interview: Scarcity of deliverable power + permitted sites with fiber and community acceptance is the real constraint, not demand. Projects that clear the power, water, fiber, permitting, and political hurdles can attract tenants; the majority of announcements .
Do these conditions for success exist in Ottawa region?
Ottawa meets ONLY some conditions but falls short on the most critical ones (especially low-cost power and low political friction) that O’Leary highlights for large-scale success.
Here’s how the Ottawa region stacks up against the key success factors from the interview:
Power (biggest constraint)
Ontario’s power is relatively clean (nuclear + hydro) and reliable, but it is not low-cost by the standards O’Leary prefers (sub-6¢/kWh regions like Alberta stranded gas, Norway, or Finland).
• Industrial/large-user effective rates are typically well above the cheapest jurisdictions once Global Adjustment, delivery, and other charges are included.
• The province’s 2026 Data Centre Playbook explicitly moves new data centres (>1 MW) into a higher rate class, removes access to cost-saving programs like the Industrial Conservation Initiative, and requires them to pay the full cost of their electricity plus any grid upgrades so residential rates are not affected.Gowlingwlg
• Hydro Ottawa is already under “extraordinary pressure”: large-load connection requests (heavily data-centre driven) recently totaled ~1,075 MW — roughly 86% of the city’s average system load — with most wanting connection in the next 1–2 years. Transformer and equipment lead times have stretched to 2–4 years.Cbc
• Own-generation / behind-the-meter is encouraged, but the overall environment prioritizes protecting ratepayers over offering cheap power to data centres.
Verdict: Manageable for smaller or government projects that can absorb higher power costs or bring generation. Poor match for hyperscale AI projects competing on pure power economics.
Permitting, Politics, and Community
Elevated and rising risk — closer to the “political reason” failure mode O’Leary warns about.
• City of Ottawa is developing a formal assessment framework (expected back to council around early 2027). Until then, hyperscale projects (≥50 MW) trigger notification so council can consider interim controls. Some councillors and mayoral candidates have expressed support for pauses or stricter limits.Cbc
• Provincial Playbook adds a dual review (technical grid impact + strategic priorities around digital sovereignty, local economic benefits, and community/environmental outcomes).
• A private member’s bill (Bill 146) proposing a provincial moratorium on AI data centres was introduced in early November 2026.
• Public opposition is growing across Ontario (Oakville and Mississauga have imposed moratoriums; Ottawa residents have organized against projects). Federal voluntary principles also emphasize no cost-shifting to residents and minimizing environmental impacts.The Globe and Mail
Verdict: Ottawa IS NOT a “red carpet” jurisdiction. Timeline risk is real, and political opposition can delay or kill large projects — exactly the risk factor O’Leary flags as likely to eliminate a large share of announced capacity.
Fiber, Latency, and Connectivity
Relatively strong.
Ottawa has solid carrier presence (Equinix, Bell, Rogers, others), local fiber providers, and even Ciena’s advanced networking labs. It works well for government, enterprise, and Canadian-sovereignty workloads. It is not a top-tier low-latency hyperscale interconnection hub on the level of Northern Virginia or certain U.S. or European markets, but it is adequate for many Canadian or government-oriented tenants.
Land, Water, Climate, and Tenants
• Industrial land exists on the outskirts; cool climate is a modest advantage for cooling efficiency. Modern closed-loop designs reduce water concerns.
• Strong government demand: The federal government is in early planning for a ~60 MW data centre in the National Capital Region (estimated $1.5–1.8B). Digital sovereignty and Canadian data localization are policy priorities.
• Hyperscalers are more selective; higher power costs and regulatory scrutiny make pure commercial AI training projects less attractive than in lower-cost power regions. Smaller AI firms, colocation, and government/defence tenants are more realistic near-term users.
Bottom Line
Ottawa is a reasonable location for government, defence, sovereign-cloud, or smaller/mid-scale data centres that value reliability, Canadian jurisdiction, and existing connectivity, and that can tolerate higher power costs and longer permitting timelines.
Ottawa is a weak match for the large-scale, power-price-sensitive AI hyperscale projects O’Leary describes as the winners.
Those tend to favor jurisdictions with abundant cheap power (especially stranded gas or hydro), faster permitting, and lower community/political resistance. Alberta (where O’Leary himself has been active) is a clearer fit on the power and cost criteria than Ottawa.
Success in Ottawa is possible but more constrained and politically contingent than in the highest-ranked locations from the interview.
Data Center Barriers to Entry: Why Power, Permits, and Politics Will Decide the Winners Through 2030
Kevin O’Leary has been unusually blunt about the AI infrastructure boom. In a recent MarketBeat interview, the Shark Tank investor—who is himself developing large-scale data center projects—warned that 60–65% of announced data centers will never get built. The reason is not a sudden collapse in demand (he calls compute demand “insatiable” and roughly 3× current supply into at least late 2027). It is a combination of hard physical and regulatory barriers plus politics.Yahoo Finance
These constraints will shape the industry for the next four years, through the remainder of the current U.S. political cycle, the 2028 presidential election, and the corresponding provincial and municipal decisions in places like Ontario. Sites that clear the barriers early will become scarce and valuable. Most announcements will not.
The Real Barriers to Entry
O’Leary’s list of what is required to succeed is straightforward and unforgiving:
• Land with the right attributes — preferably non-agricultural, within reasonable driving distance of a community for labor, but not so close that noise or visual impact triggers strong opposition (he has cited 3–4 miles as a practical range).
• Water rights — even with modern closed-loop or air-cooled designs that use far less water than older facilities, legal access still has to be secured.
• Dark fiber and low-latency connectivity — tenants (especially hyperscalers) need it; without it the site is incomplete.
• Power — the dominant constraint. Developers must typically bring their own generation (often natural-gas turbines behind the meter) rather than relying on a constrained grid that raises residential rates. Low-cost power (ideally well below 6¢/kWh) is a major differentiator. Pre-development work alone can run $20–30 million per gigawatt for engineering, permitting, and agreements. A large ~1.4 GW project has been described in the $16 billion range.YouTube
• A credible path to permit in 18–24 months, with equipment (turbines, transformers, etc.) arriving on a matching schedule. Delays destroy project IRRs and financing.
These are not soft requirements. Missing any one of them is what O’Leary calls a “fatal flaw.” The complexity is high enough that experienced teams with a track record of delivering large infrastructure projects have a meaningful advantage.
Politics Over the Next Four Years
Politics is the variable that can override an otherwise solid technical package. O’Leary has argued that a large share of the projected failures will come from political reasons rather than pure engineering shortfalls. That dynamic is unlikely to fade between now and 2030.
Several factors are already visible and will remain relevant:
• Electoral incentives. Midterms and local elections create opportunities for candidates to campaign on pausing or blocking data centers. Community opposition is real—many residents do not want large facilities near their homes. Politicians who can claim they “stopped” a project often gain more short-term support than those who emphasize future tax revenue or jobs.
• Ratepayer protection. Jurisdictions are increasingly explicit that data centers must not raise residential electricity bills. This leads to higher rate classes, requirements to fund grid upgrades, and preferences for behind-the-meter generation. Ontario’s 2026 Data Centre Playbook is one example of this approach; similar logic appears in multiple U.S. states.
• Moratoriums and interim controls. Several Ontario municipalities have already imposed temporary freezes while they write policy. In the U.S., state-level and local restrictions continue to appear. Even where no formal ban exists, the risk of one emerging during a permitting process is enough to scare off financing.
• Community benefit tests. Support tends to rise when projects deliver visible local gains—high-paying construction and operations jobs, significant tax revenue, and sometimes surplus power returned to the grid. Projects that cannot demonstrate these benefits face higher political risk. O’Leary has noted that opposition can flip when the economic numbers become concrete.
• Federal and provincial overlays. In Canada, digital sovereignty and data localization add another layer of preference for domestic capacity. In the U.S., the interplay between federal energy policy, state regulators, and local zoning will continue to vary widely by jurisdiction.
Over the next four years these pressures will not disappear. If anything, they will intensify as more large projects move from announcement to actual applications and as electricity demand growth becomes more visible on residential bills. The 2026 U.S. midterms have already shown how energy and infrastructure issues can be politicized; the 2028 cycle will likely do the same.
What This Means for the Next Four Years
The practical implication is scarcity of deliverable capacity rather than scarcity of announced capacity. Sites that already have land control, a realistic power plan, fiber, and a high-probability permitting path will command premium interest from tenants. Everyone else will face the 60–65% attrition rate O’Leary described.
Winners are likely to share a few characteristics:
• They treat the project as a real-estate and infrastructure development problem first, not a pure technology play.
• They diversify across jurisdictions so that a single political decision does not kill the entire portfolio.
• They align early with local economic interests (jobs, taxes, grid contributions) rather than relying solely on national AI narratives.
• They lock in power and permitting milestones on timelines that still make sense for hyperscaler and government demand curves extending into the late 2020s.
Demand itself is not the primary risk. The combination of physical barriers and political risk is. Projects that solve both will be the ones still standing—and leasing—by 2030. Most of the rest will remain announcements.
No comments:
Post a Comment