Everything About AI
NextFin News - Larry Fink, chief executive officer of BlackRock Inc., warned that delays in the build-out of artificial intelligence driven by public opposition will make the technology the "domain of large firms," limiting access for everyone else. Speaking on a panel Tuesday at the Canada Investment Summit in Toronto, Fink argued that the faster capacity is built, the more AI can be democratized — a pointed intervention in a debate that intensified over the weekend after Anthropic CEO Dario Amodei called for a slowdown in AI development.
The tension is stark: the very voices calling for caution are, in Fink's framing, accelerating the concentration they fear. If AI development slows, the fixed costs of compliance, safety testing, and capital-intensive infrastructure do not fall — they become a higher floor that only the largest players can clear.
The Event and the Warning
Fink made the remarks during a panel discussion at the Canada Investment Summit, a two-day gathering in Toronto hosted by Prime Minister Mark Carney to attract foreign investment into the country. The summit drew hundreds of executives and global asset managers; in his welcome-reception remarks, Carney said the world's largest investors, managing more than $120 trillion of assets, would come to "peer into our shop window." Carney has set a goal of stimulating $1 trillion in investment in Canada over the next five years, pitching opportunities in energy, infrastructure, critical minerals, and advanced manufacturing.
"The faster we can build out more capacity, the more we can democratize and make it available for everyone," Fink said. Delays in the build-out of AI because of public opposition, he warned, will make the technology the "domain of large firms."
The timing was deliberate. Fink spoke a day after Amodei, whose Anthropic is one of the leading frontier AI labs, warned in a weekend essay that AI could soon "outrun our ability to understand and control it" if left unchecked. OpenAI CEO Sam Altman and xAI founder Elon Musk said they agreed with the call for greater caution. President Donald Trump pushed back on Monday, writing on Truth Social that "the only control or 'guardrails' that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!"
Fink is not a neutral observer in this debate. BlackRock, the world's largest asset manager with a record $14 trillion in assets under management at the end of 2025, has positioned itself as a financier of the AI build-out. Its AI Infrastructure Partnership, formed with Global Infrastructure Partners, Microsoft, Nvidia, and Abu Dhabi's MGX, completed a $40 billion acquisition of Aligned Data Centers on July 21. On August 10, during a CNBC panel, Fink said the United States needs more than 70 gigawatts of new power capacity to support AI operations, at a cost of $50 billion to $60 billion per gigawatt — roughly $500 billion in immediate funding needs, with trillions more required in the years ahead.
That same day, Nvidia announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to establish financing platforms intended to mobilize more than $500 billion of third-party capital for the AI build-out. "The AI buildout will require unprecedented investment and a skilled workforce to turn that investment into the infrastructure that will help power future growth," Fink said in the Nvidia statement. "This partnership deepens our relationship with NVIDIA, including through the AI Infrastructure Partnership, and brings together NVIDIA's leadership in accelerated computing with BlackRock's ability to connect long-term capital to essential infrastructure."
The question is not whether AI is risky. It is whether slowing it down changes who wins.
The Opposition That Fink Is Warning About
Fink's warning lands on a political landscape that is already moving against rapid build-out. Community opposition to data centers has hardened through 2026. Recent polling found that 41% of Americans would oppose a data center being built within three miles of where they live, up from 28% in January 2026, while support dropped from 37% to 24% over the same period. A March Gallup poll found 70% of people opposed data centers in their areas.
The opposition has translated into concrete delays. A report by the tracking service Data Center Watch found that nearly half of U.S. data centers planned for 2026 have been delayed or canceled, pointing to a 7-gigawatt shortfall in available power as grid bottlenecks, transformer shortages, and community opposition stall projects. The same report determined that in the first quarter of 2026 alone, at least 75 projects valued at $130 billion were disrupted by local opposition. During the first six weeks of 2026, more than 300 state data center bills were filed across the United States, with proposals for statewide moratoria introduced in 14 states by legislators from both parties.
Some jurisdictions have already acted. New York established the first statewide moratorium on hyperscale data centers, framed as a bid to keep residents' electricity bills down and protect environmental resources. The city of Denver announced a moratorium on new data centers to reassess regulations, and Ypsilanti, Michigan, passed an emergency pause on approvals while it evaluates zoning and infrastructure impacts. In Utah, demonstrators gathered at the state capitol in May to oppose the proposed Stratos data center, a project speculated to cover about 40,000 acres and use 9 gigawatts of power.
This is the pushback Fink is warning about — and it is not abstract. When a state pauses approvals, the firms that can wait it out are the ones with the deepest balance sheets. The ones that cannot are the smaller developers, the challenger labs, and the communities that hoped cheaper access to AI would follow faster deployment.
The Economics of a Slowdown Favor Scale
The mechanism behind Fink's warning is straightforward but often overlooked. AI development carries enormous fixed costs: data centers, semiconductor capacity, electricity, and increasingly, safety testing and compliance teams. A slowdown or new regulation adds more fixed costs without reducing the base. In any industry where the cost structure is dominated by upfront capital and compliance, market share consolidates toward firms that can spread those costs over the largest revenue base.
Fink made essentially this point in his annual chairman's letter to shareholders in March, writing that companies with the "data, infrastructure, and capital to deploy AI at scale are positioned to benefit disproportionately." The logic cuts against the stated intentions of the slowdown camp. If AI development is paused or licensed, the firms best equipped to absorb the pause — those with deep balance sheets, in-house regulatory affairs functions, and existing capacity — lose the least. Startups and smaller labs, which depend on rapid iteration and cheap access to compute, lose the most.
The wealth-inequality thread Fink has pressed throughout 2026 sharpens the point. In his March letter, he wrote:
The massive wealth created over the past several generations flowed mostly to people who already owned financial assets. Now AI threatens to repeat that pattern at an even larger scale.
He also wrote, in the letter's discussion of AI:
AI may accelerate this trend further. The companies with the data, infrastructure, and capital to deploy AI at scale are positioned to benefit disproportionately. That is not unusual, and none of this is inherently problematic. Market leadership has always shifted with technological change. The broader question is who participates in the gains. When market capitalization rises but ownership remains narrow, prosperity can feel increasingly distant to those on the outside.
The power data underscore the scale of the capital barrier. The U.S. Energy Information Administration said on September 9 that U.S. power consumption will rise to record highs in 2026 and 2027, driven by AI-hungry data centers and electrification. The EIA projected power demand will rise from a record 4,195 billion kilowatt-hours in 2025 to 4,270 billion kWh in 2026 and 4,349 billion kWh in 2027. Goldman Sachs Research has forecast U.S. data center power demand climbing from 31 gigawatts in 2025 to 41 GW in 2026 and 66 GW in 2027.
Here is the crux: a slowdown that delays new capacity while demand keeps compounding means the firms with existing capacity — hyperscalers and the infrastructure consortia backing them — gain pricing power. Scarcity is a moat.
BlackRock Is Not a Disinterested Referee
There is a second-order implication that Fink's framing does not surface. The asset manager warning that delays concentrate AI is itself positioned to finance the infrastructure that only large players can afford. The "democratization" frame and the business interest point in the same direction — more build-out, faster — but the beneficiaries are not symmetric.
Who benefits from speed: infrastructure owners, chip makers, utilities, and the large cloud providers that can monetize scarce capacity. Who is exposed: smaller AI labs, startups dependent on cheap open access to frontier models, and the "democratized" users Fink invokes. Since the launch of ChatGPT in November 2022, much of Wall Street's AI-driven market gains have been led by companies at the center of the boom, pointing to a more narrow set of winners — a pattern Fink's own letter highlights.
There is a tension here that deserves scrutiny. Fink diagnoses concentration as a social risk, yet the policy solution his firm advances — build faster, mobilize private capital — channels more capital through the same large institutions that already sit at the center of the market. BlackRock's assets under management reached a record $14 trillion in 2025 as clients entrusted the firm with nearly $700 billion of new assets. A financial system that routes the AI build-out through a handful of asset managers and infrastructure consortia may deliver capacity faster, but it does not obviously broaden ownership of the gains.
BlackRock's shares have rallied alongside the AI capital cycle, and the market prices the firm as a bet on its continuation. The exposure is the point: the messenger's incentives run in the same direction as his message. That does not make the message wrong. It does mean the democratization argument should be weighed against the balance sheet behind it.
The Counter-Thesis: Could a Pause Help Challengers?
The strongest case against Fink's framing deserves a direct answer. A slowdown could, in principle, give smaller players time to catch up, reduce the risk of a winner-take-all lock-in, and allow open-source and lighter-weight models to mature. If frontier capability races ahead unchecked, the argument runs, the gap between the top lab and everyone else widens irreversibly; a pause narrows the diffusion lag. There is also a strategic argument: safety incidents from rushed deployment could trigger a far harsher regulatory backlash than a negotiated slowdown. From this angle, Amodei's voluntary caution is the least-bad path to avoiding a draconian regime that would truly entrench incumbents.
The counter-thesis holds only under conditions that are unlikely to be met. A pause helps challengers only if it is universal, binding, and time-limited. Voluntary slowdowns by responsible labs are not binding on well-capitalized private actors or state-backed competitors abroad. In that asymmetric game, the firms that slow down cede ground to those that do not — and the compliance burden that eventually follows still falls hardest on the small. A voluntary pause is, in effect, a cartel of the conscientious, and cartels without enforcement leak.
There is a second flaw in the counter-thesis: it assumes the alternative to voluntary caution is a well-designed regulatory regime. The political reality is messier. The 300 state bills filed in the first six weeks of 2026 were not a coordinated framework; they were a scatter of moratoria, tax-incentive suspensions, and zoning pauses driven by local grievances over water, noise, and electricity bills. That is not the careful, scaled compliance that would protect small entrants. It is the kind of fragmented friction that large firms absorb and small ones do not.
The falsifying signal is specific. Fink's concentration thesis weakens if the United States adopts a binding, universally enforced licensing regime for frontier AI models within the next 12 months, with compliance costs scaled to firm size. Watch for two things: first, a federal AI licensing bill with meaningful small- and medium-enterprise carve-outs; second, open-source model capability at fewer than 10 billion parameters closing to within 10 percent of frontier benchmarks without proportional capital. If neither materializes by the third quarter of 2027, the large-firm domain call stands.
Structural, Not Cyclical
This is a structural shift, not a cyclical fluctuation. The cost structure of AI — capital intensity, proprietary data advantages, and regulatory fixed costs — is a regime change, not a temporary imbalance. It will not revert on its own. Cyclical claims require evidence of mean reversion; there is none here. Once capability and infrastructure concentrate in a handful of firms, the advantage compounds: more data improves models, better models attract more users, more users fund more capacity. That is a flywheel, not a wave.
The time horizons point in the same direction but at different speeds. In the short term, over the next six to twelve months, expect volatility around the regulation debate; infrastructure and semiconductor names tend to benefit from any resolution that favors "build faster." Over the medium term, one to three years, consolidation among AI labs is the base case — smaller players acquired or priced out, while infrastructure owners capture scarcity rents. Over the long term, three years and beyond, the open question is whether policy can re-engineer the cost structure itself through public compute capacity or compliance scaled to firm size. Absent that, AI remains a large-firm domain regardless of the political cycle.
Three scenarios frame the path. The base case is no binding federal pause: build-out continues, concentration increases, and infrastructure owners and large labs win. The upside case for diffusion is federal investment in public compute plus scaled compliance that lowers the entry floor, allowing smaller firms to regain footing. The downside case is a major safety incident that triggers abrupt, blanket regulation; smaller firms exit, and concentration accelerates beyond even Fink's warning.
The signals to watch are concrete: EIA data center demand prints, federal AI legislation, open-source benchmark convergence, and deal flow through the AI Infrastructure Partnership and rival consortia. The EIA's next Short-Term Energy Outlook and the Goldman Sachs data center demand revisions will tell whether the power bottleneck is tightening or easing. State-level moratorium votes — particularly in the 14 states where they have been proposed — will show whether the local opposition hardens into durable barriers or softens as communities negotiate benefit agreements.
Speed and caution are supposed to be opposites. In AI, they have the same beneficiary: the firms that can afford to run, and to wait.
Explore more exclusive insights at nextfin.ai.
