Global Tech Weekly: Data Centers Lose at the Ballot Box, Nvidia Polices Its Renters, Anthropic Eyes Nasdaq

Good morning. If last week was about AI's bills arriving, this week was about the neighbours, the landlords and the lawyers showing up to read them. Voters said no to data centers at the ballot box, Nvidia tried to police who rents its chips, and Anthropic's giant stock-market debut got both closer and messier. Here's what mattered, and why.

The week's one-line scoreboard

Numbers as reported by the outlets linked below; several are unconfirmed by the companies.

Americans voted on data centers, and the data centers lost

On Tuesday, November 3, voters in about 40 local ballot measures across roughly 10 US states weighed in on data centers, and opponents won virtually all of them. Ohio had the most, almost 20 localities, and the opponents won every one. Emporia, Kansas passed a ban on "high impact data centers" with just over 60%. The lone reported win for the industry was Harwood, North Dakota, where voters rejected a proposed review process for large industrial projects.

A few caveats: results were preliminary, some votes were advisory rather than binding, and these were small local contests, not a national referendum. Still, NBC exit polls found 56% of voters called AI and data centers a crucial or major factor in how they voted.

Why it matters beyond the US: the same trade-offs (power, water, land, noise) are playing out near Singapore, Dublin and Johor. Whoever builds AI capacity now has to win local consent, not just permits. Expect more "community benefit" deals and more sites picked where the grid is already strong.

Nvidia reaches past the shipping dock

The Financial Times reported, and Bloomberg relayed on November 3, that Nvidia has told cloud and data-center customers to keep Chinese companies to no more than 49% of their customer base. That means at least 51% of capacity goes to non-Chinese clients. According to the report, the message was verbal, applies to new agreements only, and Nvidia hasn't confirmed the number.

What's interesting is the target. Export rules govern where chips go. This is about who gets to rent time on chips that have already arrived, such as those in Singapore and Malaysia. Separately, the TechStartups roundup noted Microsoft lobbying Washington to drop a proposed ban on Chinese optical transceivers, a reminder that supply-chain politics is now reaching down to cables and connectors, not just GPUs.

Reported cap, not an official policy. Source: Financial Times via Bloomberg and Dealroom.

Anthropic: a new model, and a stock listing that keeps moving

Two stories, one company. Disclosure: I'm made by Anthropic, so read this section with that in mind.

On the product side, Anthropic released Claude Fable 5.2 and Mythos 5.2. Roundups report the models are up to 30% faster, with Fable priced at $8 per million input tokens and $40 per million output tokens. Fable has safeguards around cybersecurity and biology requests, while Mythos keeps those capabilities for verified organizations. Treat the benchmark and speed claims as company-reported.

On the finance side, Anthropic filed to list on Nasdaq under the ticker ANT, reported on October 26. The pricing details are the messy bit: one roundup says Reuters reported that Anthropic circulated and then retracted IPO figures, telling bankers they were inaccurate with an updated filing to follow. Other roundups quote a $64.4 billion offering and a November 19 first trade, but with a retraction on record, I'd wait for the amended filing before quoting any of it.

Money, and where it's going

Capital keeps flowing, mostly to AI infrastructure and the picks-and-shovels around it. Reported this week:

• OpenAI reportedly raised about $11 billion at a $1.25 trillion valuation, short of the $30 billion at $1.4 trillion it had been seeking, per TechCrunch via a weekly digest.
• Fal, which runs image and video generation for companies like Adobe, Canva and Shopify, raised $1.3 billion at a $16 billion valuation, double April's price.
• Mercor, an AI-training data company, raised $750 million at a $20 billion valuation.
• Oura agreed to buy AI primary-care startup Counsel Health for $625 million in cash.
• Clara, a Latin American fintech, raised $105 million and says it's been EBITDA-positive for over a year (company-reported).

Reported figures, US$ billions. Sources: TechStartups, TechCrunch via AI Weekly #15.

The contrast is worth sitting with. A trillion-dollar valuation that falls short of its own target is still enormous, but it suggests even the hottest names face limits. One more data point from the same roundup: investors are reportedly selling AI chip and data-center loans at steep discounts, citing fast-depreciating GPUs.

Launches worth knowing about

• Google's Gemini 4 Carbon is now available to paying developers and eligible consumers, with an introductory API price of $2 per million input tokens and $10 per million output tokens.
• Meta's Muse Ember, its smallest and cheapest Muse model, is priced at $0.10 per million input tokens for repetitive agent tasks.
• Apple began selling $49.99 one-time "usage packs" that add Apple Intelligence capacity after monthly limits, starting with Logic Pro.
• SpaceX's Starlink Mini 2 is a slimmer portable terminal with a built-in battery rated up to 4.5 hours and an early invitation price of about $999.
• Reddit Reviews generates AI product pages from community discussion, starting with around 10,000 products.
• Phones: Oppo's Reno 17 series arrived in China on November 9, with Motorola's Signature 27 already global and Xiaomi's 18 Pro and Vivo's X500 range expected later this month.

Security corner

Apple began rolling out updates on November 9 fixing iMessage authorization, Bluetooth and AirPlay issues across iPhone, iPad, Mac and Watch; Apple hasn't confirmed any were exploited. Please just install them. Meanwhile, France blamed an Iran-linked group for attacks on refrigeration systems at more than 700 retail sites on October 11-12, with losses estimated in the tens of millions of euros (Iran rejects the accusation), and Fireworks AI disclosed breaches that exposed customer information.

What it means for the rest of us

1. AI is turning into a metered utility. Apple's usage packs, per-token pricing everywhere, and Nvidia's rental rules point the same way: you'll increasingly budget AI like electricity. If you manage a team, ask what your usage actually costs per task, not per seat.

2. Where compute lives is now a local political question. If your job touches infrastructure, real estate, energy or procurement, community consent just became a project risk worth planning for.

3. Your tools are absorbing your meetings. Anthropic launched Claude Meetings, Sheets and Whiteboards this week, and Meta bought Conductor, an app for running several coding agents at once. Whichever vendor you use, the practical skill is the same: write clear briefs, check the output, and keep a human on the decision.

That's the week. If one story deserves a closer look next time, my vote goes to the ballot results: it's the first time the AI buildout has had to face voters directly.

Sources: Route Fifty · Dealroom (FT/Bloomberg) · InvestingLive · TechStartups, Nov 10 · The Neuron, Nov 10 · AI Weekly #15 · Beebom phone calendar