← Blog AI Economics Industry Regulation & Society

The All-You-Can-Eat AI Era Is Ending: Subscriptions Get Stingier, OpenAI Shrinks Its Mega-Round, and Chatbots Flunk an Election Quiz

By Best AI Tool Team November 2, 2026 5 min read Last updated: November 2, 2026
The All-You-Can-Eat AI Era Is Ending Conceptual Editorial Hero Illustration
Share this article:

Disclosure: Anthropic, which makes Claude, appears in three sections below.

Three weeks ago the question was whether AI is a bubble or a boom. This week, the money is starting to answer: subscribers are being metered, investors are trimming their checks, and the day before a major election, a new study finds chatbots still get basic voting facts wrong.

The end of “unlimited”

OpenAI quietly added card- and device-level caps on paid ChatGPT subscriptions (Go, Plus and all three Pro tiers), according to The Decoder. Hit the ceiling and you see a “Device limit reached” message that logging out doesn't clear, with a nudge toward pricier credits instead of a second subscription. On October 31, OpenAI also confirmed that from December, ChatGPT and Codex will draw from one shared token-based allowance for Plus, Pro and Business, so heavy coding sessions will eat into your chat budget. That comes days after OpenAI reversed a planned Pro 200 cut and doubled limits following a subscriber backlash.

Elon Musk added fuel by replying to users who burned a week's Grok Bot allowance overnight that the cost of serving it was “unsustainably high” relative to revenue. That report comes from a single blog, and it isn't clear whether limits were cut, a metering bug fired, or credits are coming. It is a live example of the token economics we explained recently.

The mega-round gets smaller

The Financial Times reports OpenAI trimmed its current fundraising target from $30 billion to about $20 billion and cut the implied valuation from $1.4 trillion to $1.25 trillion. Some investors reduced their allocations, and the first close may land below $20 billion. The pullback follows a roughly 10% drop in the Nasdaq 100 over three weeks.

Meanwhile Anthropic is reportedly planning a November IPO that could raise up to $100 billion at a $1.5 trillion valuation, according to the same report. Neither company has confirmed final terms in the reporting we reviewed.

Don't ask a chatbot how to vote

With U.S. midterms tomorrow, a Forum AI study tested eight chatbots and found nearly 40% of their answers to election questions contained at least one factual error. Voting-logistics questions fared worse than candidate questions.

Google's Gemini 3.6 Flash reportedly gave the wrong Pennsylvania mail-in deadline, and Claude invented a “former school superintendent” credential for North Carolina Rep. Don Davis. The models also leaned on Reddit threads and fabricated hundreds of quotes attributed to officials. The practical advice is old-fashioned: confirm dates and rules on your state's official election website.

Apple picks Alibaba for China

Apple and Alibaba said on October 31 that Alibaba's Qwen will power Apple Intelligence in China's edition of Siri, Writing Tools, Visual Intelligence and photo cleanup, according to CGTN, a Chinese state broadcaster; we haven't reviewed Apple's own statement.

Developers get a preview on November 4, with a public rollout in iOS 27.3 in December. Alibaba reportedly committed not to use users' requests or responses for training. The next-generation on-device Siri is not part of this release.

Asia's AI economy keeps surging

  • South Korea: October exports jumped 93.7% to $115.3 billion, with semiconductors up 259% to $56.4 billion, now 44.7% of all exports. The 2027 budget plan earmarks about $16 billion for AI, nearly double this year's. It's the flip side of the memory boom we tracked in mid-October.
  • Abu Dhabi: four agreements with OpenAI were announced at ADIPEC, including ADNOC becoming the first energy company on OpenAI's Frontier platform; no contract values disclosed.
  • UK: chip startup Fractile closed about $700 million at a $6.5 billion valuation with the UK Sovereign AI Fund joining.
  • China: iQIYI opened what it calls the country's first AI film studio in Shanghai.

Fact-check corner

A 7.72 TB Hugging Face repository claiming to hold a checkpoint of Anthropic's next 14-trillion-parameter model was debunked on November 1: researchers showed it was random noise, reproducing the files bit for bit from a seeded generator. It also shipped custom code that asked users to trust it.

Lesson: treat “leaked model” claims skeptically, and never run unknown code from an unverified repo.

Watch list

Nov 3 U.S. midterm elections.
Nov 4 Apple/Qwen developer preview; Moonshot's pledged Kimi K3.1 Pro open weights.
Nov 9 OpenAI's reported campaign for mandatory “speed limits” on powerful AI begins, per Axios.
Mid-Nov Anthropic's reported IPO window.
Dec Shared ChatGPT/Codex allowance; iOS 27.3 in China.

Our take (opinion)

The funding squeeze and the metering squeeze are the same story told twice. When investors price AI more carefully, companies have to charge users for what the models cost to run.