On Tuesday, Goldman Sachs CEO David Solomon did something unusual for a bank that has spent two years championing the artificial intelligence boom: he talked about what could go wrong. Speaking on the firm's quarterly earnings call, Solomon said that localized political volatility around data centers is a risk worth watching ahead of the November midterm elections, and that rising financing needs combined with rate pressure could squeeze returns on some AI projects. He still expects significant infrastructure to be built and says Goldman is well positioned to participate — but the hedge itself is the story. When Wall Street's most prominent AI bull starts pricing in political backlash, financing stress, and community resistance, it signals a shift in the broader AI conversation. And that shift matters far beyond data centers. It matters to any business putting an AI voice on the phone.
Key points
- Goldman Sachs CEO David Solomon flagged election-driven political volatility and rising financing costs as risks to the AI data center buildout during the bank's October 13 earnings call.
- A Columbia Business School study estimates the AI buildout could consume roughly 3.6% of U.S. GDP annually through 2032, with complex financing structures posing potential systemic risk.
- Community opposition to data centers — over power costs and environmental impact — mirrors growing consumer wariness toward AI phone calls.
- As AI spending faces sharper scrutiny, businesses should favor AI tools with transparent pricing, measurable outcomes, and no inflated claims.
- Consent-first design, AI disclosure, and immediate opt-out handling are becoming durable advantages as AI regulation tightens.
According to Reuters reporting from October 13, Solomon pointed to two converging pressures on AI infrastructure investment. The first is political: election outcomes will shape how states and localities handle zoning, power allocation, and incentives for data center projects, and that uncertainty is already being priced into decisions. The second is financial: the sheer scale of capital required for AI projects, layered on top of a higher-rate environment, could make some projects unviable. Goldman has skin in this game. In August, Nvidia announced a partnership with six major financial institutions — Goldman Sachs among them — to launch compute platforms targeting more than $500 billion in third-party capital for AI infrastructure. So Solomon's caution is not the view of a distant observer. It is a partner in the buildout telling investors that the road ahead includes friction, not just growth. Separately, a paper by Columbia Business School professor Stijn Van Nieuwerburgh, prepared for a Brookings Institution conference last month, estimates the AI buildout could consume roughly 3.6% of U.S. GDP annually through 2032 — more than $10 trillion — and warns that increasingly intricate financing structures could pose systemic risk. The study frames the buildout as proportionally larger than electricity, railroads, interstate highways, or the internet. Those figures are worth verifying against the original paper, but the direction is clear: AI has become an economy-wide capital project, and economy-scale capital projects attract scrutiny.
One of the most telling threads in the Reuters reporting is community opposition. Residents across the United States are pushing back on data center projects over electricity demand, potential utility bill increases, and environmental impact. In other words, ordinary people are asking a reasonable question: who bears the cost of this AI buildout, and who consented to it? If that question sounds familiar, it should. It is the same question consumers are asking about AI phone calls. The public's relationship with AI is being shaped right now — by utility bills in one context, by unexpected robocalls in another. Trust in AI is not built at the scale of trillion-dollar projections. It is built or destroyed in small, repeated interactions: whether a call discloses that it is AI-assisted, whether a person can reach a human, whether an opt-out is honored immediately, whether the caller had permission to dial in the first place. This is why consent-first design is not a legal checkbox for AI calling. It is the difference between AI outreach that compounds goodwill and AI outreach that compounds resentment. At My AI Call Center, we run structured campaigns only against approved, permissioned, or reviewed contact lists. We check list sources and consent records before a single call goes out, and we tell clients plainly if a list will not support the campaign — before they spend anything. That discipline exists precisely because the backlash Solomon is describing is real, and it lands hardest on the businesses that skipped the discipline.
There is a second lesson in Solomon's comments that has nothing to do with politics and everything to do with arithmetic. When financing costs rise, projects only survive if their returns are real and measurable. The same is true for AI tools at the level of an individual business. The past few years have produced plenty of AI spending that is hard to justify on a spreadsheet — seat licenses nobody uses, platforms billed monthly whether they produce outcomes or not, dashboards that measure activity rather than results. In a market where Goldman Sachs itself is stress-testing AI returns, business buyers should be stress-testing their AI vendors the same way. Our answer to that is simple: we report what actually happened. Every campaign ends with a named outcome report — disposition codes, per-call notes, routed follow-up requests, opt-out and DNC logs. Calling starts at 9 cents per connected minute, tiered by volume, with the rate agreed before launch and locked for the campaign. Setup and management fees are quoted upfront. No per-seat charges, no platform bill, no minimums you did not choose. This is not modesty for its own sake. It is a bet that in a post-hype AI market, the winning tools will be the ones whose numbers hold up to inspection. Trillion-dollar projections are exciting. A campaign that pays for itself and shows you the receipts is useful.
Solomon's political-volatility warning also has a regulatory dimension that outbound calling operators should not ignore. Growing political attention to AI infrastructure — siting disputes, power debates, community hearings — is part of a broader trajectory toward more AI regulation, not less. For AI calling specifically, the framework is already established. AI-generated voices are treated as artificial voices under the TCPA, which means prior express consent is required before dialing. State-specific quiet hours, calling-day restrictions, and registration rules apply and vary by jurisdiction. Disclosure obligations are expanding, not contracting. Businesses that treat compliance as a design principle — rather than a patch applied after launch — will find themselves with a durable advantage as enforcement tightens. Our approach reflects this: AI disclosure on every call, keyword opt-outs honored immediately and carried into DNC records, escalation paths to a live human, recording only with disclosure and consent, and data that is never shared, sold, or used to train shared models. Campaign requirements still vary by location, industry, contact type, and consent status, and clients remain responsible for getting appropriate legal guidance before launch. But the direction of travel is clear, and the businesses that build consent-first habits now will not be scrambling when the rules harden.
The Van Nieuwerburgh systemic-risk framing carries one more practical lesson. If intricate financing structures are the risk vector in the AI buildout, then small and mid-sized businesses should be especially wary of AI commitments that mimic that complexity — long platform contracts, heavy upfront capital, infrastructure-heavy deployments that cannot be unwound. The safer pattern is the opposite: managed services with one clear goal per campaign, quoted in full before launch, cancellable, and measured against a single outcome. That is how we structure every campaign — from speed-to-lead follow-up calls that reach new leads within minutes inside approved windows, to renewal and retention calls placed 30 to 60 days before the renewal date, to database reactivation blitzes run as structured multi-touch campaigns over two to four weeks. Each campaign starts with a question, not a platform: what do you need the call to accomplish? Everything — script, calling windows, escalation path, routing of outcomes back into your CRM — is scoped around that one answer, and nothing launches until you approve it. In an environment where even Goldman Sachs is hedging its AI exposure, the businesses that will come out ahead are the ones keeping their AI commitments small, clear, and accountable.
Solomon's comments do not signal the end of the AI buildout — he was careful to say Goldman remains well positioned and expects significant infrastructure to be built. What they signal is the end of the AI buildout's honeymoon. Political scrutiny, financing pressure, community resistance, and skeptical buyers are now part of the landscape, and every business deploying AI — whether a $500 billion compute platform or a five-hundred-call reminder campaign — will be judged by the same standard: did you get consent, did you disclose, did you deliver, and can you prove it. That standard is not a burden. For the businesses that meet it, it is the moat.