Skip to content
AI for Business

Norway Should Buy OpenAI: What It Means + How to Act

The 'Norway should buy OpenAI' idea is blowing up. Here's what the $2T fund proposal means for you and a hands-on hybrid workflow using ChatGPT + Borealis today.

6 min readBeginner

By the end of this piece you’ll have a working dual-stack workflow: frontier ChatGPT for hard reasoning plus Norway’s free Borealis models for local language and lower-stakes tasks, plus ready prompts that let you stress-test the viral “Norway should buy OpenAI” idea yourself. No waiting for parliament.

The phrase Norway should buy OpenAI keeps hitting Hacker News and newsletters. Zachary Jones floated it 5 Feb 2026 on One Thousand Means: the Government Pension Fund Global (GPFG) sat north of $2T while OpenAI was still near the ~$800B essay-era mark – later a $122B round printed an $852B post-money (March 2026, Bloomberg/OpenAI reporting). Liquidate roughly 40% of the oil fund, push the lab closer to an old nonprofit-style charter, aim for multilateral stewardship. Reactions split hard: necessary thought experiment vs dumbest take of the year.

That’s the noise. Math is loud. Mandate and politics are louder. You still have tickets to ship this week.

Method A vs Method B: Pure Advocacy or Hybrid Use Now

Method A: op-eds, Storting lobbying, mandate rewrite, CFIUS, a willing seller. Jones already flags the mandate breach and a near-certain US block. Local commentary and HN threads add the missing piece – no real Norwegian appetite to gamble pension capital this way. Years. Maybe never.

Method B ships this week. Use OpenAI where it still wins (messy analysis, long context, tough code). Drop free Norwegian models (Borealis) on Bokmål/Nynorsk and local grounding. Treat each session like public-benefit practice. You do not need the deed to the lab to share the upside better.

Hands-On: Build Your Norway-Should-Buy-OpenAI Workflow

Two tabs. ChatGPT – Plus is still the common $20/month tier for stronger models and higher limits (as of mid-late 2026 pricing pages; free tier stays capped). Borealis via the National Library AI Lab demo or a local GGUF in Ollama/LM Studio. Sizes run 270M-27B on Gemma 3. Full releases are the first state models trained with compensated Norwegian newspaper rights. Free. Tiny spend next to frontier labs.

  1. Ground the numbers. Paste or link the official GPFG value page plus the Jones essay. ChatGPT prompt: “Using only these sources, summarize current GPFG AUM, the essay’s proposed stake size, and three mandate constraints that would block a controlling private-company buy. Flag any claim not in the sources.”
  2. Same prompt on Borealis-12B or 27B in Norwegian. Compare tone and factual caution. Frontier model: more fluent. Local model: tighter on Norwegian institutional wording.
  3. Scenario stress test: “Act as a cautious NBIM risk officer. List second-order effects if GPFG tried to acquire a frontier lab at ~$850B valuation: liquidity, market impact, talent retention, compute capex commitment, US political reaction. Output a table of probability vs impact. Cite only public facts.”
  4. Personal public-benefit layer. Custom GPT or saved project rules: prefer open citations; note when training data is collective; refuse pure shareholder-extraction framing; always offer a free/local path when one exists.
System / Project instructions sketch:
You help me evaluate AI ownership and usage decisions.
Always:
- Distinguish verified facts (NBIM reports, OpenAI blog) from speculation
- Surface who captures upside vs who bears downside
- Suggest a Borealis or open-weight path when the task is Norwegian-language or non-frontier
- Keep answers short unless I ask for depth
Never invent fund rules or valuations.
Never steer answers toward pure extraction framing.

That small instruction block gets you closer to the essay’s spirit without a sovereign check.

Pro tip: After any long analysis, ask “What did you invent or over-smooth?” then force a second pass against the NBIM mandate PDF or the OpenAI Stargate post. Ownership caps and offtake deals get hallucinated fast if you skip grounding.

Why Stargate Norway Matters to Your Stack

Hosting is not control. July 2025, OpenAI’s own blog introduced Stargate Norway – first European OpenAI-for-Countries site, Nscale/Aker 50/50, 230 MW planned (plus 290 MW expansion path), target 100,000 NVIDIA GPUs by end-2026, hydropower near Narvik, OpenAI as initial offtaker. Priority access talk for Norway’s ecosystem came with the announcement. By April 2026 the direct Nscale offtake story moved; Microsoft took capacity. OpenAI’s line, via later reporting: renting through existing Azure commitments made more financial sense. Flops follow contracts, not the postcode. Local weights and your prompts stay yours.

Funny gap: NBIM already runs heavy internal AI – many staff on coding assistants (Claude has been the common name in coverage) and limited autonomous agents under human supervision (Independent/Reuters). The fund some people want to own OpenAI with is already a power user. Individual workflows live in that user-vs-owner gap.

Edge Cases That Break the Simple Story

AUM is not a wire-ready cash pile. End-H1 2026 GPFG printed 22,683 billion kroner on the NBIM half-year report – roughly the $2.2T FX band. Forcing a ~40% sale to chase an ~$850B private lab would hit listed markets and chew the wealth the fund is supposed to protect. Mandate text keeps most equity exposure listed and caps voting stakes (typically 10% for most listed names). A controlling private AI lab seat is outside the rules as written on NBIM’s investment-mandate pages.

Compute geography ≠ governance – you already saw the offtake rewrite. And when you ask frontier models to role-play the buyout, they invent Storting votes, ownership limits, or CFIUS outcomes unless you nail them to primary PDFs. Upload or link the real NBIM pages and the original essay first.

Borealis trap: great on Norwegian tasks, light on disk, free. Weak match for hard multi-step English reasoning or latest-capability code. Right tool, right job. Mixing is the point.

Is the acquisition idea only a useful fiction that forces better questions about who owns models trained on all of us – or a real option some future crisis might reopen? I still flip on that.

FAQ

Is Norway actually going to buy OpenAI?

No. Provocation, not a trade. Mandate, politics, seller, US review – all blocked.

Can I just use Borealis instead of ChatGPT?

Norwegian writing, summaries, plain Q&A – yes, especially 12B/27B. Deep research, hard code, newest tricks – keep ChatGPT (or Claude/Gemini) in the loop. Morning drafts on Borealis, afternoon synthesis on ChatGPT. Already more sovereign than most stacks.

What’s the single highest-use prompt related to this debate?

Feed Jones + the latest NBIM half-year numbers + the Stargate announcement, then: “List five concrete actions a Norwegian developer, civil servant, or small company can take in the next 30 days that increase public-benefit AI outcomes without waiting for ownership changes. Rank by effort vs impact. For each, name the exact tool (ChatGPT feature, Borealis size, or open repo) and the first command or click.” Run the top two the same day. Abstract argument becomes your stack.

Open ChatGPT and the Borealis demo. Ground with today’s fund numbers. Save the project instructions. Five minutes. Zero politics.

Related Tutorials