The consensus read on Trump’s Super Intelligence Task Force is that it is mostly theatre: a rebranding exercise dressed up in executive-order language. The structural questions underneath it are rather more uncomfortable than that.
What the Super Intelligence Task Force Actually Does
On 4 October, President Trump announced the formation of the Super Intelligence Force (SIF) via Truth Social, naming Jay Clayton, Andrew Ferguson, Emil Michael and Scott Kupor as leaders, reporting to Trump and chief of staff Susie Wiles. The framing was grandiose: ‘bigger than the Industrial Revolution, and the Internet.’ Taken at face value, that is easy to dismiss.
Taken as policy architecture, it is harder to ignore. According to Politico, the task force’s general approach is intended to push forward the doctrine of applying existing legal liability laws to AI companies, as outlined in an executive order Trump signed in June. That is not nothing. Applying existing liability frameworks to AI outputs, rather than crafting a purpose-built regulator, could expose AI companies to tort and product-liability claims without requiring new legislation. The implications for companies carrying roughly $2 trillion in combined valuation, a figure cited in commentary by Holman W. Jenkins Jr. in The Wall Street Journal, would be material.
Politico also reports that the White House had drafted an executive order over the summer to create a FINRA-like regulator for AI, but moved away from it after Meta chief executive Mark Zuckerberg opposed the plan. The retreat is instructive: the administration is manoeuvring between competing industry pressures, not simply deregulating. The Super Intelligence Task Force is where that manoeuvre lands for now.
Meanwhile, NBC News reports that venture capitalist David Sacks had previously served as AI czar before stepping down when his term as a special government employee expired. The SIF is, in part, filling the institutional gap his departure left.
The task force’s remit extends beyond headline-grabbing liability questions. According to Yahoo Finance, the group will examine how federal agencies receive warnings about AI-related incidents, including data breaches, cyberattacks and attempts to bypass safeguards built into AI models, and recommend ways to improve the government’s response using existing authorities. That is unglamorous plumbing, but it is also where AI governance will actually be tested.
The Global Crossing Problem Nobody Wants to Price In
MishTalk‘s Mike Shedlock draws a comparison the bulls would prefer not to examine: Global Crossing. Founded in 1997, the company set out to build a 100,000-mile undersea fibre-optic network connecting 200 cities across 27 countries. At its peak in 1999, its market valuation reached $47 billion, despite never generating an annual profit. The company spent $15 billion laying cables into a market that simultaneously overbuilt. When dot-com customers collapsed, so did the revenue base. Global Crossing’s January 2002 bankruptcy filing listed $12.4 billion in debt against $22.4 billion in assets, at the time the fourth-largest in US history.
The accounting dimension is the part worth sitting with. Global Crossing would, per Shedlock’s account, book $100 million in swap revenue upfront as immediate profit, while capitalising the $100 million it spent and spreading that cost across decades. Little to no physical cash or actual services changed hands. The pattern was self-reinforcing until it wasn’t.
The AI parallel Shedlock draws is not that the technology is worthless. It is that circular investment structures, where AMD pays OpenAI to be a customer, where Nvidia and OpenAI have comparable arrangements, resemble the vendor-financing loops that amplified the telecom bust. Jenkins, writing in the Journal, gets to a similar destination via a different route: ‘The thing to worry about is whether their spending will pay off, and whether government will let it.’ In his new book, Kevin Roose documents that as far back as 2017, OpenAI leaders Sam Altman and Greg Brockman considered future government nationalisation so likely that they debated obtaining advance payments from China, Russia and the United States for tools those governments might expropriate.
The consensus may be overweighting the technology’s transformative potential and underweighting the simple question of when, and how much, profit actually arrives. Models are getting cheaper. China, per Shedlock’s October 2025 analysis, holds structural advantages in cheap electricity, open-source models and lower capital requirements. The result is price pressure on the returns that justify current valuations. The Super Intelligence Task Force could, at the margin, shape which companies survive that pressure, particularly if liability exposure becomes asymmetric. That is the detail the rebranding noise is drowning out.
