|
|
Was this forwarded to you? Get The Fourth Estate AI Brief in your inbox →
|
|
|
|
View this issue in your browser
Tuesday Brief
The Compute Consolidation Issue
The interesting thing this week with AI and publishing — two companies are on track to control most of the world's AI computing power by 2028. That's not speculation from a tech blog. It's the projection from SemiAnalysis, the semiconductor research firm that tracks where every GPU ends up. For community publishers trying to figure out which AI tools to bet on, this consolidation matters. Your vendor choices narrow when two players control the infrastructure.
|
|
THE COMPUTE MONOPOLY
The Rundown: A new analysis says the two leading AI labs are on pace to own the majority of usable computing power worldwide within three years.
The details:
- By end of 2026, 40-50% of new compute will flow to OpenAI and Anthropic alone.
- AI labs now generate $50+ million per megawatt of revenue, letting them outbid every competitor for hardware.
- Total AI capital expenditure will exceed $2 trillion annually by 2028.
- Compute pricing could rise from $10-15 million to $40-50 million per megawatt as labs bid up prices.
- Only SpaceX and Meta are positioned as potential third players by hoarding compute.
Why it matters for us: If two companies control the infrastructure, they control the pricing. For a 10,000-circ weekly budgeting $100/month for AI tools, this consolidation means fewer alternatives if your current vendor raises rates or changes terms.
Read at dwarkesh.com →
|
|
|
THOMSON REUTERS' BET
The Rundown: Thomson Reuters announced it's launching a frontier AI model trained on its exclusive legal and financial data.
The details:
- The company will build rather than license AI capabilities, using its proprietary archives as training data.
- Thomson Reuters owns decades of legal filings, financial records, and news content competitors can't access.
- This signals major information companies see their archives as AI training assets, not just content to monetize through subscriptions.
Why it matters for us: Your morgue of 80 years of local coverage, obituaries, and community records might be worth more than ad revenue someday. When a company like Thomson Reuters builds a model on exclusive data, it validates what small publishers own that Big Tech doesn't.
Read at thomsonreuters.com →
|
|
|
THE OP-ED QUESTION
The Rundown: Nieman Lab examines whether opinion sections should accept op-eds written with or by AI tools.
The details:
- The piece investigates current newspaper policies on AI-generated opinion content.
- No industry standard exists yet for disclosure requirements on AI-assisted submissions.
- Opinion editors are developing ad-hoc guidelines before formal policies emerge.
Why it matters for us: That letter to the editor about the school bond issue might already be AI-assisted. Worth developing your submission guidelines now — before you have to make a call on deadline about whether to run something that reads a little too polished.
Read at niemanlab.org →
|
|
|
FRAMING AI COVERAGE
The Rundown: A new piece argues we should cover AI as a medium that changes communication environments, not just a productivity tool.
The details:
- Marshall McLuhan noted new media initially mimic old forms before developing native content — AI is still in this phase.
- Guy Debord's 'spectacle' concept suggests AI will further mediate and commodify direct human experiences.
- The key question isn't 'what can AI do?' but 'how does AI change our media environment?'
- McLuhan argued there's no technological inevitability if we examine what's actually happening.
Why it matters for us: Next time you're writing about AI in local schools or city hall, skip the 'AI will transform everything' framing. Interview decision-makers about WHY they chose specific tools and WHO benefits. Your readers get agency, not inevitability.
Read at theconversation.com →
|
|
|
THE UPDATE
Quick Items Worth a Scan
The Rundown: Smaller developments that crossed our radar this week.
The details:
- Google's Project Genie now generates explorable virtual worlds in real time — early sign of AI-native content forms that don't mimic old media.
- AI labs have shifted from negative gross margins to profitable unit economics, explaining their ability to raise billions in new funding.
- Regulatory concerns may prevent labs from releasing their best models publicly, which could keep advanced capabilities behind enterprise paywalls longer.
Why it matters for us: The profitability shift means AI tools are here to stay. The regulatory uncertainty means the tools available to small publishers may lag what enterprises can access.
|
|
|
|
The compute consolidation story is the one to watch. When two companies control the infrastructure, they set the terms. For now, tools remain accessible and reasonably priced. That may not hold. Worth paying attention to who owns what.
Trevor Slette runs Quadd.ai — AI tools built for publishers.
|
|
|
|
|
|
Run a newspaper. Use Quadd.
AI tools built for small newsrooms. Document intelligence that turns court reports and box scores into copy you can paste into your layout. Audio transcription that finds the quote without you scrubbing for it. AP-style proofreading that tightens copy without flattening voice. Built by someone who knows what a Tuesday night feels like. Free for seven days.
Request a 7-day trial →
|
|
|
