The Regulation Reversal
The Regulation Reversal
When Congress punted on AI policy, the market assumed a singular outcome: industry deregulation through federal preemption. But what’s actually happening is messier and more revealing. The Trump administration isn’t just blocking state AI laws through legislative channels. It’s using executive power, litigation, and selective enforcement to manufacture a regulatory monoculture while simultaneously enabling the exact consolidation it claims to oppose. The deeper signal here isn’t about regulation versus deregulation. It’s about who gets to write the rules when the traditional legislative process breaks down.
Deep Dive
Trump’s Executive End-Run Shows Legislative Breakdown is Now Permanent
Trump’s decision to issue an executive order blocking state AI regulations after Congress twice refused to act represents a fundamental shift in how tech policy gets made in America. This isn’t a victory for either side of the regulation debate. It’s an admission that Congress no longer functions as the primary lawmaker for technology policy, and that executive power, litigation, and corporate lobbying have become the decisive forces instead.
The executive order creates a Litigation Task Force designed to challenge state AI laws on constitutional grounds, claiming they force AI makers to embed “ideological bias” in models. This framing is telling. Trump isn’t arguing that AI regulation is economically harmful or technically infeasible. He’s arguing that state laws constitute unconstitutional speech regulation. This sets up an entirely different battleground, one where the merits of AI safety become irrelevant and the case lives or dies on First Amendment doctrine.
What matters for tech companies and founders is the asymmetry this creates. A fragmented regulatory landscape is genuinely difficult to navigate, but it can also be worked around through selective deployment and feature variations by region. A federal prohibition on state regulation, enforced through litigation, is far more powerful. The administration is effectively saying: major tech companies can operate under whatever framework they choose, answerable only to federal standards that haven’t materialized yet. Congress’s failure to act hasn’t created a vacuum. It’s created an opportunity for executive capture of the entire regulatory space.
This also signals something crucial about the relationship between Trump and the tech industry. The major AI companies wanted federal preemption badly enough to fund lobbying efforts. They didn’t want to negotiate with fifty state legislatures or risk ballot initiatives in California or Vermont. But they also didn’t need to bargain with Congress. The executive order suggests they got what they wanted without the messy process of legislating. That’s a powerful precedent for future policy issues.
The Datacenter Delusion: Goldman Sachs Maps the Four Ways AI Infrastructure Fails
The infrastructure boom that’s driving the entire AI investment thesis has built-in vulnerabilities that nobody in venture or growth equity is seriously modeling yet. Goldman Sachs released a scenario analysis describing four possible outcomes for AI datacenters through 2030, and half of them end badly. This isn’t fringe thinking. This is one of the world’s largest investment banks essentially saying: the math on $1.6 trillion in cumulative datacenter spending doesn’t work if you change any of three variables.
The base case, which Goldman frames as most likely, is a supply-demand squeeze over the next 18 months followed by a normalization after 2027. That’s actually the optimistic scenario. But the other three are instructive. One assumes monetization fails because users won’t pay for AI products they can already access for free. Another assumes corporate cloud spending continues to decline as companies realize their current spend levels are unsustainable. The fourth assumes demand overwhelms capacity, creating a supply crisis that forces builders to play catch-up for years.
What’s remarkable is that three out of four outcomes involve either demand collapse or excess supply. That’s not a warning about one edge case. That’s a warning that the equilibrium everyone is assuming (steady, profitable growth in capacity utilization) is not the most likely path. Some investors have noticed. France’s AXA told Bloomberg it’s exercising “greater caution on the artificial intelligence build-out” in datacenter financing. Norway’s sovereign wealth fund said it’s wary of the sector’s volatility.
For founders and operators in infrastructure, this creates a perverse incentive structure. Right now, building more datacenter capacity is profitable because customers will lease anything available. But if Goldman’s scenarios have even 30 percent probability on the downside, you’re building for a future where occupancy rates crater and pricing power evaporates. The companies building today are essentially racing to build as much capacity as possible before the demand wall hits. That’s not investment. That’s musical chairs with billions in capital.
AI Agents Will Restructure How Brands Survive, and Most Aren’t Ready
British Airways’ CEO Sean Doyle articulated a problem that will reshape customer-facing industries in the next three to five years: if autonomous AI agents become the primary interface between customers and travel providers, being visible to humans becomes secondary to being legible to machines. This isn’t speculative. It’s already starting to happen.
Doyle’s framing is precise. When humans search for flights, they see a website and make decisions based on brand, price, interface design, and reputation. When an AI agent searches for flights on behalf of a user, it’s optimizing for a different set of variables: API availability, data completeness, pricing signals, and integration depth. A beautiful website means nothing to a machine. Neither does brand reputation. What matters is whether the system can read your data, understand your pricing structure, and execute transactions without friction.
This creates a winner-take-most dynamic for AI platforms that serve as gatekeepers. If OpenAI’s ChatGPT becomes the default agent for travel booking, then BA’s entire brand value collapses into a question of whether OpenAI chooses to surface BA as an option. The airline loses control of its own customer acquisition. This is the mirror image of what happened to news publishers when Google’s algorithm became the primary traffic source.
BA is responding with what Doyle frames as a “leapfrog opportunity”: rebuilding its digital infrastructure from scratch, consolidating data silos, and preparing to integrate with multiple agent platforms simultaneously. That’s the right move strategically, but it’s also expensive and requires a multi-year commitment. For smaller airlines and travel providers without BA’s capital reserves, the math is worse. They’ll end up dependent on whoever controls the agent layer.
The deeper implication is that AI advancement isn’t just about model capability anymore. It’s about distribution and control of the interface layer. A company that owns the dominant AI agent in a vertical effectively owns the market structure for that vertical. That’s why Oracle’s partnership with OpenAI for Stargate datacenters matters so much. It’s not just about infrastructure. It’s about who gets to sit between customers and service providers for the next decade.
Signal Shots
ServiceNow Acquires Armis for up to $7 Billion, Doubling Down on IoT Security — ServiceNow is in advanced talks to acquire Armis, which specializes in securing and managing IoT devices. This signals that enterprise security is no longer about perimeter defense or endpoint management. It’s about controlling the explosion of connected devices across corporate networks. ServiceNow is betting that the businesses it serves will pay heavily to avoid the compliance and breach risks that come with unmanaged IoT sprawl. The precedent for valuations in this space is significant, and it suggests that security businesses protecting against fragmented, complex infrastructure command large multiples.
Disney Licenses 200+ Characters to OpenAI for Video and Image Generation — Disney just licensed over 200 characters to OpenAI for use in Sora video generation and ChatGPT image tools. This is the clearest signal yet that content creators have accepted that AI companies will generate licensed content at scale. The deal legitimizes Sora as a commercial product and signals that major media companies are moving from litigation to licensing. This is the transition point where generative AI becomes part of the entertainment infrastructure rather than a threat to it. Expect licensing deals to accelerate across music, publishing, and film.
Home Depot’s Year-Long Security Lapse Shows the Persistence of Legacy Risk — A security researcher discovered that Home Depot exposed access to internal GitHub repositories and cloud systems for over a year after the company ignored his disclosure attempts. This isn’t a sophisticated attack. It’s administrative negligence at massive scale. For founders building security or compliance tools, this is the market opportunity: large enterprises with critical infrastructure struggle to manage basic access controls. The problem isn’t capability. It’s visibility and enforcement.
Reddit Sues Australia to Escape Age Verification Ban, Arguing It’s Not Actually Social Media — Reddit filed suit to exempt itself from Australia’s under-16 social media ban, claiming it’s fundamentally different from Facebook and TikTok because users can browse without logging in. This is a technical argument that misses the regulatory point: governments are targeting the behavioral dynamics of social platforms, not specific technical architectures. Reddit will likely lose this case, but the litigation signals that platform companies are now fighting governments on definitional grounds rather than policy grounds. Expect this to happen repeatedly as regulations proliferate.
Oracle’s Stock Volatility Reveals the Fragility of the Stargate Bet — Oracle’s stock has become a market barometer for AI sentiment, surging and crashing on news about the OpenAI Stargate partnership. When reports surfaced that datacenters might not be completed until 2028 instead of 2027, Oracle’s stock tumbled. This volatility tells you that investors are pricing the entire Oracle thesis around a single customer and a single infrastructure bet. That’s not a healthy position for a software and cloud giant to be in. It means Oracle’s growth story is now entirely contingent on OpenAI’s ability to monetize AI services at the scale required to justify the capex.
China’s Real AI Strategy Isn’t About LLMs, It’s About EVs and Robotics — Tim Wu wrote in the Financial Times that despite US rhetoric about an existential AI race, Chinese state spending is actually concentrated on dominating electric vehicles, robotics, and other autonomous systems. The US has spent $350 billion on AI infrastructure in the past year. China is investing more heavily in applied automation and manufacturing. This is a strategic mismatch: the US is building compute capacity for models that may not generate returns, while China is building manufacturing capacity for products people actually want to buy. The AI race framing obscures the real competition.
Scanning the Wire
TransUnion’s Password Reset Vulnerability Let Me Hijack My Own Credit Report — A security researcher easily exploited TransUnion’s account recovery process to gain unauthorized access to credit reports, a flaw that was left open to potential identity thieves. This is the inverse of Home Depot: instead of exposed infrastructure, it’s a broken authentication flow at a company handling sensitive financial data.
Earth’s Orbit Is Running Out of Space, and the CRASH Clock Is Ticking — The Register reports on orbital congestion as satellite launches accelerate, creating collision risks and debris clouds that could trigger cascading failures. This is the infrastructure limit nobody talks about: you can’t just launch datacenters to space without solving the physics problem of orbital overcrowding.
European Cloud Providers Say EU Should Have Blocked VMware-Broadcom Merger — A trade group argues that regulators made a mistake approving the deal, which consolidated critical infrastructure software. This is a post-facto challenge to a merger that’s already closed, suggesting that the approval process in Europe may have underestimated the competitive risk.
Amazon’s Ask This Book Feature Lets AI Answer Questions About Any Book, and Authors Can’t Opt Out — Amazon rolled out a Kindle feature that answers questions about plot and characters using generative AI, with no author consent or opt-out mechanism. This is the content creator’s nightmare: their work becomes training data for systems they didn’t authorize and can’t control.
CFTC Withdraws 2020 Crypto “Actual Delivery” Guidance, Opening Door to Regulated Crypto Markets — The US Commodity Futures Trading Commission withdrew guidance that required physical delivery of digital assets, a technical change that signals regulatory alignment with crypto infrastructure companies’ desires for easier market access.
SMS Text Message Verification Is Fueling Global Influence Campaigns, Cambridge Study Shows — A University of Cambridge analysis found that cheap SMS verification enables mass creation of fake accounts for coordinated disinformation campaigns. This is a structural vulnerability in how the internet authenticates identity at scale.
VPN Companies Are Lying About Where Their Traffic Actually Exits — An analysis found that major VPN providers misrepresent their server locations to users, a transparency failure that undermines trust in privacy infrastructure.
700Credit Breach Exposed 5.6 Million Records From Auto Dealership Background Checks — The credit verification company suffered a data breach affecting millions of individuals, showing that third-party infrastructure serving the auto industry is a high-value target for attackers.
Outlier
Elon Musk Is Testing Whether Europe’s Content Moderation Laws Actually Have Teeth — Musk is publicly feuding with European regulators after X was fined for content moderation failures, and reports suggest White House officials are backing his challenges to EU enforcement. This is the most dangerous possible dynamic: a US president signaling that American tech companies shouldn’t comply with foreign regulatory frameworks. If this precedent holds, it collapses the entire multi-jurisdictional regulatory system that has kept global tech platforms operating at scale. You’re watching the birth of regulatory balkanization in real time.
See you all in the next one.