Trends›AI regulation doesn't bind
AI regulation doesn't bind
Through 2030, AI/tech regulation (safety/liability, antitrust against the hyperscalers, training-data copyright) does not materially slow the frontier labs or hyperscaler capex.
66 ↓1 confidence, computed 23 Oct 2026
- Legal
What rests on thismarkets it underwrites or threatens
Threatens
Evidencethe schedule the figure draws
The evidence, newest first
- May 2026
for · observable The Digital Omnibus (agreed in provisional trilogue 6 May 2026) defers EU AI Act high-risk obligations from 2 Aug 2026 to 2 Dec 2027 (and 2 Aug 2028 sectoral), conceding the regulatory infrastructure was not ready. EU AI Act Digital Omnibus defers high-risk obligations (Council/Consilium) - Feb 2026
for · primary MSFT/GOOGL/AMZN/META guided to ~$725B of 2026 capex, up 77% from 2025's ~$410B — no regulatory or antitrust event has dented the build-out. Tech AI spending ~$725B in 2026, +77% YoY (CNBC) - Jan 2026
against · claim Universal/Concord/ABKCO filed a $3.1B copyright suit against Anthropic (28 Jan 2026) and both Anthropic and OpenAI are reportedly earmarking investor capital for copyright liabilities — a recurring, escalating cost stream. Music publishers file fresh $3.1B suit vs Anthropic (PYMNTS) - Jan 2026
against · observable Judge Stein affirmed an order compelling OpenAI to produce a 20M-log ChatGPT sample in the NYT MDL, a discovery defeat analysts say will 'effectively dictate settlement terms' — litigation exposure rising, not resolving. OpenAI must turn over 20M ChatGPT logs in NYT case (Bloomberg Law) - Dec 2025
for · primary Trump EO establishes a DOJ AI Litigation Task Force (active 10 Jan 2026) to challenge state AI laws, conditions BEAD funding on avoiding 'onerous' AI laws, and seeks a federal statute preempting state AI regulation. Executive Order: national AI policy framework, preempting state AI laws (White House) - Dec 2025
against · claim On 9 Dec 2025 the European Commission opened a formal antitrust investigation into Google's use of online content to train its AI models, while FTC/DOJ continue probing Microsoft-OpenAI and Nvidia — an active enforcement front, no remedy yet. EU opens antitrust probe into Google AI-training content (Euronews) - Sep 2025
neutral · observable Anthropic settled Bartz for ~$1.5B but raised $13B at a $183B valuation the same week (settlement ~0.8% of value); the underlying Alsup ruling held LLM training on legally-acquired books is fair use. Anthropic pays authors $1.5B to settle copyright (NPR) - Sep 2025
for · primary Judge Mehta's remedy (finalised Dec 2025) rejected DOJ's structural demands — no Chrome or Android divestiture — imposing only behavioural data-sharing, citing AI as a natural competitive check. Google keeps Chrome — no structural breakup in antitrust remedy (NPR) - Jul 2025
against · observable GPAI obligations entered force 2 Aug 2025; OpenAI, Anthropic and Mistral signed the voluntary GPAI Code while Meta refused as legally overbroad — the regime is real enough to force a frontier-lab posture and split the field. Meta refuses to sign EU's GPAI code of practice (TechCrunch)
What would move itthe measures watched
- watchCombined hyperscaler capex (MSFT+GOOGL+AMZN+META), annual $ and YoY. The cleanest test of whether ANY regulation binds. Rising/accelerating = holds; a guidance cut attributed to a legal/antitrust constraint = breaking. $725B for 2026, +77% YoY — no regulatory drag visible.
- watchEU AI Office enforcement vs GPAI providers post-2 Aug 2026 (fines, recalls). Fining powers (up to 3% global turnover) switch on 2 Aug 2026. Zero/token = paper tiger; a real fine or recall against a frontier lab = teeth.
- watchEU AI Act high-risk applicability dates (further Omnibus slippage). Each deferral is a 'for' signal. High-risk dates already pushed from Aug 2026 to Dec 2027/Aug 2028. Further delay = Brussels blinking; hard re-imposition = bear.
- watchCopyright outcome type: training INJUNCTION vs $ damages. Damages are cost-of-doing-business (Anthropic $1.5B ≈ 0.8% of valuation). A training-data injunction (NYT v OpenAI) or forced model retraining would bind; money alone doesn't.
- watchStructural antitrust remedy on a hyperscaler (divestiture vs behavioural). Google search case ended with NO breakup — behavioural only. A forced divestiture (Chrome on appeal, or Microsoft-OpenAI/Nvidia probes) would be the binding event.