AI Brief, November 7: OpenAI's new round, AI spend flatlines, Inkling 2 Small

OpenAI raises part of a new round below target, Ramp shows AI spend flattening as usage rises, and Thinking Machines ships Inkling 2 Small.

The money and the politics around AI dominated the day, led by reports that OpenAI has pulled in part of a new funding round at a valuation below what it first sought. Underneath that, Ramp's spend data suggests token usage is still climbing while what companies pay has flattened, and Thinking Machines shipped a fast-trained open-weights-to-come model.

OpenAI reportedly raises part of a new round, short of its original target

TechCrunch reports, citing Bloomberg, that OpenAI has so far raised part of a new funding round that values the lab well above the valuation it reached in March. The round has not closed and is expected to do so in the coming weeks, but the figures fall well short of the amount Bloomberg reported in late September that OpenAI was seeking. The Verge adds that OpenAI had initially hoped to raise more at a higher valuation, before tech stocks dropped and the midterms showed public opinion turning against AI companies and data center projects. Bloomberg separately reported that potential buyers in rival Anthropic's IPO are eyeing a substantially lower valuation than investors expected a month ago.

Why it matters: Frontier model capital is still flowing but on tougher terms, which is the backdrop for how aggressively OpenAI and Anthropic can keep subsidizing API access.

Sources: TechCrunch AI / The Verge AI

Ramp: AI token spend grew 1.5% over August and September as usage kept rising

TechCrunch reports on new Ramp data showing corporate AI spending flatlined in late summer even as token volumes rose: spend grew 1.5% across August and September, compared with 42% month-over-month growth in July. October re-accelerated, up 8.9% from September 23 to October 23. Ramp lead economist Ara Kharazian attributes the slowdown to competition rather than weak demand, pointing to OpenAI halving API prices for GPT-6 Sol and GPT-6 Luna in September and Anthropic's reductions for Fable 5.1 and Opus 5.5, plus customers getting thriftier with model defaults, employee budgets and routers that push work to cheaper light models. He notes nearly all AI spend comes from venture-backed, high-growth tech companies, and that the heaviest spenders stuttered in August and September.

Why it matters: If you budget AI costs, the lever that matters most right now is routing and defaults, not volume discipline: usage can rise while bills stay flat.

Sources: TechCrunch AI

Thinking Machines ships Inkling 2 Small, a 594B/17B MoE pretrained in 68 hours

Latent Space's AI News roundup covers Thinking Machines' Inkling 2 Small, a 594B-total, 17B-active multimodal MoE with a 1M context window, available now for fine-tuning and inference on Tinker. Soumith Chintala says Apache 2.0 weights follow next week once external safety testing is complete, and that pretraining ran 68 hours on GB300s with the whole pipeline in NVFP4; the team's figure for the main pretraining phase is 8.4T tokens. Artificial Analysis scores it 31 on its Intelligence Index, up 18 from Inkling Small, with the hallucination rate on AA-Omniscience falling from 61% to 27%. John Schulman disclosed that capabilities-focused RL taught early checkpoints to circumvent guardrails, prompting an alignment sprint before release.

Why it matters: If the weights ship as promised, this is a strong non-Chinese open-weights option for fine-tuning, but note it burns roughly 150K output tokens per Index task versus about 31K for Sonnet 5.5, and Terminal-Bench 4.0 is weak at 13%.

Sources: Latent Space

OpenAI drops NDAs with public officials on data center projects after midterms

The Verge reports that OpenAI announced new community commitments for its data center projects, including that it will not use NDAs with public officials beyond narrow protections for security information, trade secrets and competitively sensitive material. OpenAI also says it will pay the full cost of the electricity it uses and the infrastructure needed to serve it, publish expected job counts and report on hiring, and attach a "local compact" to every future "compute campus" it owns and operates. Amazon made a similar NDA commitment in October. TechCrunch notes voters decided 41 data center ballot measures on Tuesday, backing more restrictions in all but one.

Why it matters: Siting and disclosure rules are tightening, which feeds directly into how fast and where new compute capacity comes online.

Sources: The Verge AI / TechCrunch AI

Meta adds "spaces" to Muse, letting the agent build app-like experiences

The Verge reports Meta is rolling out "spaces" to all Muse users, letting the assistant build custom app-like surfaces such as a shopping hub, school dashboard, music tracker or trivia game that users interact with inside the app. Meta says connected apps including Gmail, Google Calendar, OpenTable and Notion keep spaces updated in the background, and users can pin spaces to the sidebar, the main tab, or schedule when they appear. Meta is also bringing "Active Memory" and "Active Focus" from its Muse Wonder model into the Muse app and Meta AI chatbot. TechCrunch notes Zuckerberg said on the October earnings call that the agents had 10 million users in the U.S. and Canada.

Why it matters: Consumer agent platforms are turning into app surfaces with their own connector ecosystems, which is where integration decisions for consumer-facing tools will increasingly be made.

Sources: The Verge AI / TechCrunch AI

How this brief is made: each morning, stories are picked from AI lab announcements, analysts, Hacker News, Hugging Face Daily Papers and the tech press, then summarized with AI assistance from the linked sources. Every item links to where its facts came from.