Data centers are now a midterm issue. Wall Street has already underwritten $3 trillion of them.

A bipartisan warning landed on AI's biggest companies this week. House GOP leadership has begun telling them, in private meetings and in public comments, that the public image of their data centers has become a midterm political liability rather than a Democratic-only issue. The shift is the surprise, and it puts the buildout on a collision course with the roughly $3 trillion that the Wall Street Journal has now put a number on.

The political signal arrived in the same news cycle as three independent readouts from the financial side. Anthropic told its bank syndicate it could file for an initial public offering as early as the end of August, in a deal that could match or exceed SpaceX's record listing, Bloomberg reported. Broadcom, the custom AI chip supplier behind several hyperscaler builds, is reportedly raising up to $100 billion in debt financing to keep orders filled. And the Wall Street Journal has quantified what most capex headlines understate: real AI spending by big tech is roughly $3 trillion higher than their balance sheets show, once joint ventures, supplier financing, and long-dated purchase commitments are included.

The combination matters because the data center buildout is now operating on two clocks at once. Politically, it is a bipartisan target. Financially, it has already been sold in pieces, to public debt markets, to private credit funds, and to chip suppliers who have committed capacity years in advance. AI labs can no longer set the pace of the buildout on their own; voters and rate-setters have just as much leverage over what gets built and when.

The political shift is the surprise. Until this month, opposition to data centers had been treated as a local zoning fight: water rights, transmission lines, backup-generator noise. The new GOP posture, described in industry coverage, moves the issue up the party. Leaders worry that the visual footprint of a 3,000-GPU campus, the small headcount relative to the property tax bill, and the broadside that AI is being "shoved down people's throats by big tech" while companies "are eager to tell people how many jobs it's going to eliminate" have become a usable campaign image. That language is no longer confined to environmental groups.

On the financial side, the convergence is doing the same kind of work. Anthropic's annualized revenue hit $6.5 billion in July, per CNBC, which is why bankers are taking the August IPO window seriously. A listing at the scale being discussed would convert some of the buildout's private credit exposure into public equity and give the market a chance to price the same off-balance-sheet exposure the WSJ analysis is sizing. Broadcom's reported $100 billion debt raise is the cleaner read on what is actually committed: a financing that size is a bet by lenders that hyperscaler demand is durable enough to underwrite multi-year fab expansion, and once that paper is sold, the lender has as much reason to want the buildout to continue as the operator does.

SiliconANGLE analyst Dave Vellante put a name on the result this week when he wrote, in a column, that the AI buildout has become "too big to fail... quietly, at least." That phrasing is a shortcut, and the column is opinion. Read against the WSJ figure and the Broadcom raise, it points to a more concrete fact: the question of whether the buildout continues is no longer a question the labs alone can answer. It is shared with the credit committee, the rate-setter, and now, the voter.

The first binding test is timing. If Anthropic files in the August window the syndicate has been discussing, the public market will price the same exposure the off-balance-sheet analysis is trying to measure. If the filing slips, and a "could" is not a "will," the next read is whether Broadcom can place the full $100 billion at terms that assume the buildout continues at current pace. Either outcome feeds back into the political fight, because the buildings the GOP is now asking AI firms to defend are the same buildings those credit lines were sized to build.