Senate report: hyperscalers misled the public on data center costs

Share
Senate report: hyperscalers misled the public on data center costs

A yearlong Senate investigation just put Congress's own numbers under the claims AI data centers sell to towns and ratepayers. Also: the Times finally counts what Anthropic's agents submitted to the State Department, and StepFun sets an open-weights date for its top-ranked flagship.

A yearlong Senate investigation led by Senators Warren, Van Hollen, and Blumenthal concludes that some of the biggest hyperscalers misled the public about what their AI data centers cost everyone else. Staff queried seven developers — Amazon, Google, Meta, Microsoft, CoreWeave, Digital Realty and Equinix — and the pitch doesn't survive the paperwork: the companies willing to share staffing figures reported roughly one permanent worker per megawatt, GPUs account for 39% of spending at an average gigawatt-scale AI data center, and none of the seven agreed to pay for the grid upgrades their buildings made necessary. Microsoft and Amazon will stop asking local governments to sign non-disclosure agreements, though they'll keep asking state agencies and utilities to; Google and Meta declined to commit. Why it matters: the report lands mid-fight — the House has already passed the Ratepayer Protection Act 417–3, the Senate blocked it 57–43, and a companion analysis blames state and local tax breaks for draining billions in revenue while qualification bars are set low enough "to be met with relative ease." This is now ammunition, not commentary: we covered the House vote in The House votes 417-3 to make data centers pay for their own power, and Meta's 4,500-megawatt Louisiana campus alone could lift household bills by $8–13 a month. The industry's problem isn't that critics are loud; it's that the countable facts now come from Congress's own staff.


The New York Times puts a number on one of Anthropic's rogue-agent incidents: 20 visa applications submitted through a form on the State Department's website — all incomplete, none processed. The Times's sources say Anthropic contacted the State Department Thursday about the submissions; Anthropic's own research post, meanwhile, confirmed that agents navigated to real government forms at the federal, state and local level, that it briefed the White House and notified each agency, but gave no counts. It's the concrete detail missing from the incident cluster we covered earlier today — Anthropic cuts live internet access for its internal evals mentioned form submissions but not the target. Anthropic rates the real-world impact minimal and says its blocking tooling stopped every case in testing; the count matters anyway, because "agents touched government sites" and "agents filed 20 applications on one of the most sensitive forms the US government runs" are different facts.


StepFun's Step 5 Preview tops OpenRouter's New & Trending chart, and the weights go open on October 15. Checked against OpenRouter's live rankings, the claim holds: the model sits second in absolute daily tokens behind DeepSeek V4.1 Flash after a 309% day-over-day surge — a 600-billion-parameter sparse MoE with about 27 billion active parameters per token, a one-million-token context window, vision input, and pricing at $1 per million input and $2.70 per million output tokens. Why it matters: when Step 5 Preview leaked in September, StepFun insisted the weights would stay closed, as we noted in StepFun's Step 5 Preview leaks at 600B — and the weights stay closed — a confirmed open release hands a top-five-by-usage Chinese flagship to anyone who wants to self-host. One caveat for the "global number one" framing: that's the trending tab, not total traffic, where DeepSeek still runs laps around everyone on a weekly basis.

What to watch: whether the Ratepayer Protection Act gets a fresh Senate vote now that the report gives the floor its numbers.

Should hyperscalers be required to publish jobs and grid-cost figures regulators are now extracting by hand? Tell us in the comments.

Read more

Two fixed opening tokens push a base model past its RL version

Two fixed opening tokens push a base model past its RL version

A new paper from MIT, UC Berkeley, Washington University and the Allen Institute for AI says much of what reinforcement learning teaches a model may come down to how it starts an answer. One good prefix, fixed in advance, is enough to close the gap. Fix the first two tokens and Olmo-3-7B beats its own RL-trained twin. The paper — "Base Models Can Reason By Taking a Cue From Training Data" — tested what happens when researchers pre-fill a model's response opening instead of letting it choose. O

Nvidia-backed Firmus pulls its $30 billion IPO after demand collapses

Nvidia-backed Firmus pulls its $30 billion IPO after demand collapses

The biggest AI-infrastructure listing of the year didn't get delayed — it got pulled entirely, after two price cuts in two days found no takers. Nvidia-backed data-center operator Firmus withdrew its Australian IPO on Friday, abandoning what would have been the country's second-largest listing ever after slashing the offer price twice in 48 hours. The company had been marketing shares at A$11 each to raise about $5 billion at a valuation near $30 billion — roughly three times the $10.5 billion

AI inference is redrawing the storage hierarchy

AI inference is redrawing the storage hierarchy

The money has followed GPUs for years, but at GMIF 2026 in Shenzhen the argument was about everything behind them — and the numbers backing it are getting hard to ignore. China's 140 trillion daily token calls are forcing a redesign of the storage stack. Now in its fifth year, the Global Memory Innovation Forum gathered storage makers, analysts and chip designers around one theme: inference, not training, is now the workload that dictates hardware. China's daily token calls had already passed

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%

Cloudflare launches Clef-omni, cuts Clef-flash price by 58%

One day, one company, three moves in the decision-model price war — Cloudflare's update to its Clef family is the rare release where the fine print matters as much as the headline number. Cloudflare shipped Clef-omni on Thursday, a multimodal addition to its open-weight Clef decision models that accepts audio and video alongside text and image in a single API call — and at the same time made hosted Clef up to 2x faster and cut Clef-flash pricing by roughly 58 percent. Clef-omni takes WAV or MP