AI regulation won't slow the AI race. Here's the evidence
the EU delayed its AI rules. Alphabet doubled its spending.

the EU delayed its AI rules. Alphabet more than doubled its spending.
Alphabet spent $51.4 billion on property and equipment last quarter, up from $24.0 billion a year earlier.
meanwhile Brussels pushed its biggest AI regulation obligations back by more than a year.
the consensus says AI regulation will slow AI down, or that governments could pause it if they really wanted to.
we think the evidence says the opposite: the AI race CAN'T be paused, and regulation is bending around it.
What's actually happening with AI regulation
start with the EU, the place everyone points to as the strict regulator.
→ the AI Act's core high-risk rules were due to apply from 2 August 2026 → a provisional deal in May, which Parliament voted to adopt in June, moves them to 2 December 2027, and to 2 August 2028 for AI built into products like medical devices → the same deal softens the AI literacy duty and delays regulatory sandboxes to August 2027
now the US. the big fight isn't "should we slow down", it's WHO gets to write the rules:
- a draft Great American AI Act would preempt state laws on AI model development for three years
- more than 200 state lawmakers from 42 states signed a letter opposing any preemption
- neither side is proposing a pause
and the money keeps going up.
per Eco-Business's summary of IEA data, Big Tech capex topped $400 billion in 2025 and was expected to rise about 75% in 2026.
Alphabet's Q3 numbers suggest that wasn't optimistic.
Why governments can't pause AI (they're customers now)
this is the part the "just pause it" crowd misses.
governments aren't only referees anymore, they're some of the keenest buyers.
→ the US federal AI use case inventory hit 3,611 use cases across 56 agencies, up from 1,757 a year earlier → HMRC has rolled out 28,000 Copilot licences, after a cross-government trial found about 26 minutes saved per user per day
it's hard to pause something your own tax office now relies on.
and it's not because people don't want rules.
when the Future of Life Institute called for a halt on the pursuit of superintelligence, its US poll found 73% want robust regulation and 64% say superintelligence shouldn't be built until it's provably safe.
a year later? spending up, deadlines pushed back. public support didn't translate into a pause.
it's the half built skyscraper again.
billions already poured in, nobody knows exactly how tall it needs to be, and no board wants to be the one holding a half finished building.
The fair counterargument
to be fair, regulation isn't doing nothing:
→ the EU deal adds a ban on AI systems generating child sexual abuse material or non-consensual intimate content, from 2 December 2026 → it reinstates a registration duty for some systems that claim an exemption
so rules DO shape what gets built and how. they just don't stop the build.
and the brakes that actually bite are physical and local.
in Virginia, Loudoun County started a process that could pause decisions on new data centre and substation applications for up to 12 months.
that's a planning committee and a grid, not a treaty.
What uni won't teach you about AI policy
most policy and law modules teach regulation as a finished text you analyse. in AI:
→ the text keeps moving after it's "passed" (see: the AI Act) → the government is often the customer, not just the regulator → the real constraints can sit in energy, planning and procurement, not in AI law at all
The critical-thinking move: 5 questions for any AI policy headline
- does this rule stop, delay or steer? most "crackdowns" turn out to be steering. say which one it is before you form an opinion.
- what's the date it actually applies? announcement dates make headlines. application dates change behaviour. the EU just moved its own by more than a year.
- is the government the referee or the customer here? if it's both, expect the referee role to soften.
- follow the capex. after a big rule is announced, does company spending fall? if it keeps rising, the rule isn't slowing anything.
- where's the physical bottleneck? grid connections, substations, local planning. that's often where the real "pause" happens.
What to do this month
- read one law firm summary of the EU Digital Omnibus on AI and write down every date that moved.
- open the US federal AI use case inventory, pick one agency and list five things it uses AI for. you'll understand government AI better than most commentators.
- follow one capex number (Alphabet, Microsoft, Amazon or Meta) across the last four quarterly reports and plot it against regulatory news.
- write a 500-word policy memo that runs one AI headline through the 5 questions above. that's the kind of thing a think tank or civil service fast stream actually wants to see.
so TLDR: AI regulation is steering the race, not stopping it. if you want to work in AI policy, learn where the real brakes are.
if you were a minister and could only slow AI down one way, would you go for the law or the grid?
Sources
- EU Lawmakers Reach Provisional Agreement to Delay Key EU AI Act ObligationsSidley Austin (Data Matters) · 22 JUN 2026
- Worries mount about another state AI law preemptionRoute Fifty · 17 JUN 2026
- Alphabet Announces Third Quarter 2026 Results (Form 8-K, Exhibit 99.1)Alphabet Inc. via SEC · 28 OCT 2026
- AI's energy appetite is outpacing deployment of AI-based climate solutions: IEAEco-Business · 11 MAY 2026
- Number of AI use cases in the US federal government more than doubles in a year, official figures showGlobal Government Forum · 22 APR 2026
- Watch out UK taxpayers: 28,000 HMRC staffers just got an AI copilotThe Register · 27 APR 2026
- Prince Harry, Richard Branson, Steve Bannon, and 'AI godfathers' call on AI labs to halt their pursuit of superintelligenceFortune · 22 OCT 2025
- Board Approves Process that Could Pause Legislative Data Center & Substation ApplicationsLoudoun County Government · 15 SEP 2026
Get the next issue by email
theVANTAGEd is written by students to bring critical thinking to AI across real industries. One issue every two weeks.