AI Power Public Unease

Executive summary

Amazon's blockbuster compute deal with Anthropic — a $100 billion-plus, decade-long commitment securing up to 5 gigawatts of AI infrastructure — has landed in a public that is divided, watchful, and overwhelmingly safety-first. A new pulse survey of 144 Americans finds no consensus on whether the partnership is good news, a warning sign, or just business as usual.

Nearly 40% of respondents welcomed the expanded partnership, citing the AI capability gains that more compute enables. But 31% flagged the deal as a troubling concentration of power inside big tech, and 26% shrugged it off as routine competition. No single reaction commanded a majority — a clear signal that optimism about AI progress and anxiety about who controls it are running in parallel, not in sequence.

The most decisive number in the survey has nothing to do with Amazon or Anthropic directly: 63% of respondents named safety and responsible development as the single most important factor in AI development — outpacing innovation by more than 2.6 to one. That hierarchy matters because the Amazon–Anthropic deal is explicitly about scaling raw capability. Public legitimacy for that kind of investment depends on visible, credible safety commitments — and right now, trust in Amazon to deliver them is mixed at best.

Context

This survey was fielded in April 2026 with 144 U.S. respondents after Amazon announced it was deepening its partnership with AI startup Anthropic — the company behind the Claude model family — by dramatically expanding Anthropic's access to Amazon Web Services computing infrastructure. The deal is not incremental. Amazon has committed more than $100 billion over the next ten years, secured up to 5 gigawatts of compute capacity for Anthropic, and is investing an additional $5 billion immediately on top of $8 billion already deployed. More than 100,000 customers now run Claude through Amazon Bedrock.

The announcement arrived in the middle of an infrastructure arms race unlike anything the tech industry has seen. Amazon CEO Andy Jassy has outlined roughly $200 billion in capital expenditure for 2026, almost entirely AI-driven. The company added 3.9 gigawatts of new power capacity in 2025 alone and expects to double its total power footprint by the end of 2027. AWS AI revenue is already running at over $15 billion annually.

Amazon is not alone. The top five U.S. hyperscalers — Amazon, Google, Microsoft, Meta, and Apple — are collectively on track to spend $720 billion in capital expenditure in 2026. Epoch AI data shows that more than 60% of all global AI compute is already owned by U.S.-based hyperscalers, with Google holding roughly a quarter of that share. EU competition chief Teresa Ribera has publicly flagged concern about the entire AI stack — models, training data, cloud infrastructure, energy — becoming concentrated in a handful of corporations.

Into this landscape, the survey asked four questions: how respondents feel about the Amazon–Anthropic shift toward compute-intensive AI, what concerns they hold about big tech controlling AI resources, how much they trust Amazon specifically to manage AI partnerships responsibly, and which single factor matters most to them in AI development overall. Free-response answers were analyzed for thematic dimensions around tech outlook, control authority, and motivation trust. The results capture a public trying to reconcile genuine excitement about AI's potential with structural unease about who is building it — and on whose terms.

Findings

A Three-Way Split With No Winner

When respondents heard about Amazon's expanded compute deal with Anthropic, they divided into three roughly equal camps — and that division is itself the story. At 39.6%, the largest share expressed a positive reaction, accepting the premise that more computing power drives better AI. But close behind, 30.6% said the deal worried them because it concentrates too much AI power in big tech's hands. Another 26.4% called it neutral — just competitive capitalism doing what it does.

The absence of a majority view is significant. It means no single narrative about this deal — not the capability-optimist frame, not the concentration-alarmist frame — has broken through to dominate public understanding. For Amazon, Anthropic, and the broader AI industry, that ambiguity is both an opportunity and a risk: the story of what this deal means is still being written.

Safety First, By a Landslide

The clearest signal in the entire survey has nothing to do with Amazon specifically. Asked which factor matters most in AI development, 63.3% of respondents chose safety and responsible development — more than double the 24.5% who chose innovation and capabilities. Competition between companies, often cited as a driver of progress, registered just 7.2%.

That 2.6-to-1 margin is not a soft preference. It reflects a durable public hierarchy in which capability gains are welcome, but only if they arrive inside a framework of accountability. The Amazon–Anthropic deal is almost entirely framed around compute scale and capability expansion. From the public's vantage point, that framing is missing the most important ingredient.

External polling reinforces the finding. A Fathom national survey from April 2026 found that 83% of Americans support slowing frontier AI development until risks are better understood, and 71% trust independent experts over tech companies for AI governance. The respondents in this survey are not outliers — they reflect a broad national posture.

Big Tech's Profit Motive Is the Default Assumption

When asked directly what concerns them about big tech controlling AI computing resources, respondents' free-text answers clustered around a consistent theme: that companies like Amazon are primarily profit-driven, not mission-driven. Automated dimension scoring placed the average response at –0.38 on a scale where –1 represents pure profit-motive attribution and +1 represents confidence in societal alignment. That lean toward profit skepticism was statistically robust across 75 respondents who engaged with the dimension.

On a related dimension — whether government or big tech should hold authority over AI computing resources — respondents averaged –0.33, indicating a modest but clear preference for government retaining oversight rather than ceding control to corporations. The distribution was polarized, not clustered: opinions spanned the full range, but the center of gravity tilted toward public accountability.

Trust in Amazon specifically followed a predictable pattern. Respondents who expressed higher trust in Amazon were measurably more likely to view the partnership expansion positively — confirming that trust functions as the gateway variable. The problem for Amazon is that high trust is the minority position. Respondents who distrust big tech's motives and are pessimistic about compute concentration are not a fringe; they represent roughly a third of the sample.

Anthropic's Safety Brand Is Being Tested

Anthropoc has built its entire market identity around being the safety-conscious AI company — the one that puts guardrails ahead of growth. That positioning should theoretically resonate with the 63.3% who ranked safety as their top AI priority. But two pieces of evidence complicate the picture.

First, the profit-motive skepticism in this survey extends to Anthropic by implication. Respondents didn't distinguish between Amazon and Anthropic as a safer alternative — they expressed concern about the big-tech AI ecosystem as a whole. Second, external reporting reveals that Anthropic recently revised its Responsible Scaling Policy, removing so-called "if-then" pause commitments that had been central to its self-governance credibility. Critics called it "the breaking of a meaningful promise." For the 63% of respondents who prioritize safety above everything else in AI development, that policy retreat — if it registers — could undercut the very justification Anthropic offers for its rapid infrastructure scaling.

The deal's narrative depends heavily on Anthropic's safety brand holding. Right now, that brand is under internal and external pressure simultaneously.

Conclusion

The Amazon–Anthropic deal is one of the largest AI infrastructure commitments ever made. The public knows it — and is watching carefully.

The survey's three-way split on the deal reflects a market moment in which the old tech-optimism default no longer holds automatically. Nearly 40% of Americans are willing to say more compute means better AI. But a combined 57% are either skeptical of big-tech power concentration or simply unmoved by the capability argument. That's not opposition — it's a legitimacy gap that neither Amazon nor Anthropic has yet closed.

The most actionable signal is the 63% safety supermajority. Any company seeking public trust for massive AI infrastructure investment needs to lead with governance, not gigawatts. Visible, independent, and enforceable safety commitments aren't just good ethics — they're the price of entry for the segment that matters most.

Watch Anthropic's Responsible Scaling Policy closely. If further weakening of its self-governance commitments becomes a news story, the 63% who prioritize safety will be the first to notice — and they overlap directly with the respondents who are already skeptical of big tech's motives. The capability story Amazon and Anthropic are telling is compelling. The safety story still needs to catch up.