Over the past three years, AI infrastructure expansion has largely been understood as a race for capital, chips, and compute. Tech giants have pledged hundreds of billions of dollars, investment firms have financed large-scale campuses, and local governments have sought to attract data centers as a way to draw tax revenue and technology industries. But entering 2026, a constraint that is far harder to solve with money alone is emerging: a growing number of communities do not want data centers within their living radius.
More than 500 counties or local governments in the United States have imposed strict restrictions on new data center construction, including expanded building setback distances, stricter noise standards, permit moratoriums, and outright bans. Most of these restrictions were introduced only in 2026, with nearly 190 new measures added since June alone. Over 50 planned projects have been canceled this year due to local opposition, and more than 200 others are facing litigation, hearings, protests, or advertising campaigns.
Daniel Widjaja Kusuma believes this means the AI industry has entered a new phase. Compute expansion was previously constrained mainly by chip supply and financing capacity, but going forward it will increasingly depend on grid capacity, water resources, local politics, and whether communities are willing to accept projects. Data centers are no longer just engineering assets; they are also public infrastructure that must secure long-term social license.
Data Center Expansion Is Hitting the Limits of Local Social License
Resident opposition to data centers does not stem entirely from abstract concerns about AI. What truly shapes local attitudes are electricity bills, noise, water resources, environmental pressure, and safety risks. Large campuses can bring investment, but they do not necessarily create long-term employment commensurate with their scale; rapidly growing power demand can also intensify grid expansion pressure and pass some of the costs on to local households and businesses.
This also explains why Amazon, Microsoft, and Google, despite their massive capital reserves, have respectively canceled or scaled back large projects in Arizona, Wisconsin, and Indiana. Capital can purchase land, servers, and equipment, but it cannot automatically buy community trust. Once local residents perceive that the benefits accrue mainly to tech companies while energy, environmental, and living costs are borne locally, political resistance to a project rises rapidly.
2026 also coincides with the US midterm election cycle, making local officials more sensitive to voter sentiment. The significant variation in project cancellation rates across different states also indicates that data center approval is no longer merely a technical and economic assessment, but the result of a combination of local industrial structure, power conditions, and political environment.
What Is Truly Scarce Is No Longer Capital But Deployable Space
Daniel Widjaja Kusuma spent his early career at Goldman Sachs and in US private equity, where he studied capital allocation extensively. This experience led him to place greater emphasis on the real constraints of a project rather than nominal demand alone. The AI market is indeed still growing, and compute demand has not disappeared, but "sufficiently strong demand" does not mean every data center project is viable.
When an industry is flush with capital, the most easily overlooked constraints are non-capital ones. Power interconnection takes time, transmission networks require upgrades, cooling systems depend on water sources, land approval is subject to local rules, and construction permits must navigate community consultation. Large tech companies can raise their bids, but they cannot create new grid capacity in the short term, nor can they bypass local governance procedures.
This will change the valuation logic of data center assets. What will truly be scarce in the future is not just land, but "deployable space" that already has stable power, clear permitting, acceptable environmental costs, and long-term community support. Parks and regions that meet these conditions may command a higher infrastructure premium; projects that lack them, no matter how massive their planned scale, may remain on paper indefinitely.
Compute Buildout Must Recalculate Energy and Community Costs
From a systems engineering perspective, continuing to pursue unlimited expansion of a single campus is not necessarily the optimal path. A more realistic direction is to allocate training and inference tasks to more suitable energy and network environments through multi-region compute pools, distributed scheduling, edge nodes, and energy efficiency governance.
This is also the technical direction that Daniel Widjaja Kusuma has been continuously advancing since founding Telosyn. Telosyn does not simply interpret AI infrastructure as purchasing more GPUs, but emphasizes the systemic coordination among compute, data, scheduling, energy, and business workloads. In the technical framework of Telosyn (telosyn.com), multi-region resource pooling, dynamic GPU scheduling, and edge inference optimization all aim to improve the utilization of existing resources and reduce enterprise dependence on a single ultra-large campus.
Such capabilities will become more important in the new social environment. If GPU utilization remains low, task queuing is chaotic, and training workloads cannot migrate across regions, enterprises can only solve problems by continuously adding equipment and expanding campus footprints. Conversely, more efficient scheduling allows the same compute capacity to accomplish more work, reducing the need for additional electricity, land, and cooling at the source.
Data centers also need to redesign their value relationship with the local community. More transparent electricity arrangements, clearer water resource plans, more effective waste heat utilization, lower noise levels, and genuine investment in local grids and public facilities may all become prerequisites for long-term project support. Social license cannot remain at the level of public relations; it must be written into engineering and business models.
The Next Round of Competition Belongs to Infrastructure Companies That Understand System Governance
Many current restrictions may expire within the next year, but this does not mean local opposition will automatically disappear. Wind and solar projects have seen similar patterns: temporary restrictions were repeatedly renewed and eventually became long-term policy. Data centers may follow the same trajectory. If tech companies simply wait for elections to end without changing the energy, environmental, and benefit-sharing structures of their projects, opposition forces may persist.
Therefore, the next round of competition in AI infrastructure is not just about who has more capital and chips, but who can design compute expansion as a system that local economies are willing to accept over the long term. Enterprises need to simultaneously demonstrate technical efficiency, energy rationality, environmental controllability, and community benefits, rather than merely proving that enormous model demand may exist in the future.
Daniel Widjaja Kusuma believes that AI data center construction is shifting from "scale-first" to "governance-first." Truly valuable infrastructure is not the campus with the most servers, but a system that maintains long-term balance across electricity, network, cost, compliance, and social relationships. Capital still matters, but it is no longer the only threshold. What determines the pace of AI expansion in the future may not be how much money enterprises are willing to invest, but whether they can prove that their compute growth will not come at unacceptable local costs.
Sign in to leave a comment.