🔮 AI's expensive gap; superintelligent bureaucracy; Copernican trauma; supershoes & BYD's sexy mishap ++ #481

An insider’s guide to AI and exponential technologies

Ideas of the week

The $500 billion question. The AI gold rush has created a $500 billion annual revenue gap between infrastructure investment and earnings, according to Sequoia Capital. The shortage of GPUs has turned into a glut, and there is a substantial shortfall in future earnings to return that capex. Paradoxically, AI companies are underperforming earnings growth expectations on average, yet AI stock valuations continue to rise (as research analyst

Joachim Klement

argues). Since 2023, companies with significant AI exposure that are on the S&P 500 have underperformed growth expectations by 26 percentage points, while non-AI peers have missed by 32 percentage points. AI stocks have seen their growth expectations and valuations rise more than their non-AI counterparts. Economic sentiment is currently tied to AI. As Klement notes, “US tech dominance is narrowing to fewer and fewer stocks, namely those involved in building generative AI and the infrastructure needed to run it.” LLMs are proving to be game changers. Claude 3.5, for example, demonstrates reliable, intelligent utility. As I’ve previously argued, we’re likely to see substantial overcapacity in computing build-out, as we have in previous infrastructure waves (like railways and digital telecoms). Such over-investment is to be expected, but in the medium term demand will vastly exceed those current investments. For example, in 2005, after the telecoms bubble burst, 85% of US broadband capacity was unused. Internet traffic levels grew some 300 times to 2022. It doesn’t, of course, mean that some share prices won’t get hammered.

See also:

• A good Goldman Sachs discussion on whether GenAI investments can yield a return.

• See also: A critique of Breakthrough Energy Ventures – a climate-focused venture capital fund founded by Bill Gates and other billionaires – argues that many of its investments are misguided due to the founders’ outdated biases, particularly in nuclear, carbon capture and fringe renewables.

Leviathans. In 2020, author and professor of politics David Runciman discussed his provocative thesis with me, which remains highly relevant today. He posits that modern AI systems and long-standing institutions such as states and corporations have for centuries functioned as ‘superintelligences’, reducing complex realities to narrow decision inputs. Since the 17th century, we’ve developed strategies to align these institutional superintelligences with human interests through legislation and constitutionalism – essentially, through politics. The challenge of aligning markets with climate goals, as journalist Martin Wolf’s analysis shows, underscores the critical need for these policy tools in long-term alignment. Markets drive innovation and create efficiencies, expanding policy options and resources. However, the myopia of the markets in pricing long-term risks requires a strengthening of the government’s capacity to deal effectively with these far-reaching challenges. This lesson is crucial as we approach AI adaptation.

Homo sapiens’ identity crisis. Philosopher of technology and social theory Benjamin Bratton’s “The Five Stages of AI Grief” challenges conventional AI narratives, framing our reactions as a “Copernican trauma” that decenters human exceptionalism and troubles our self-image. Bratton outlines and critiques four perspectives: denial dismisses the importance of AI; anger sees it as a threat; bargaining assumes we can control it; depression foresees inevitable catastrophe. We deny the potential of AI, rage at its implications, bargain for control and despair at its inevitability. These reactions, Bratton argues, hinder our understanding of AI’s true nature. Instead, he proposes a “non-grief” approach that sees AI as part of the broader evolution of intelligence. With this perspective, AI is emergent rather than controllable, and encourages thinking beyond human-centred concerns. It recognises AI as an ongoing transformation of intelligence itself.

• See also: Neuroscientist and EV member Anil Seth asks, “what are the implications of AI systems that appear conscious, even if they aren’t?”

Regulatory arbitrage. As Brussels wields its Digital Markets Act against Meta’s ‘pay or consent’ model and France prepares antitrust charges against Nvidia, the regulatory spotlight is on Big Tech. However, while regulators fixate on these familiar skirmishes, a new front is quietly opening up in the AI arena. Amazon and Microsoft, with a finesse that would make Sun Tzu nod in agreement, are executing a strategy of “reverse acquisitions”. By hiring most of the team from would-be OpenAI competitors Adept and Inflection and licensing their technology, they achieve de facto acquisitions while skilfully avoiding merger scrutiny. This manoeuvre – predicted by Reid Hoffman – exploits the gap between Silicon Valley’s breakneck evolution and the glacial pace of regulatory adaptation. While the authorities struggle to rein in yesterday’s monopolies, a new form of market concentration emerges unchecked. Can our regulatory frameworks evolve at Silicon speed, or will the future of AI be shaped by those most adept at regulatory arbitrage?

---

Data

Globally, 40% of all EV batteries now use LFP (lithium iron phosphate). This type of battery does not contain any nickel or cobalt, resulting in dramatically lower mineral intensity compared to other lithium-ion types.

By 2024, more than 50% of the EU’s electricity will be generated from renewable sources.

There are now 65,000 public charging points for electric vehicles in the UK, an increase of 46% year on year.

Coursera’s global enrollment has quadrupled to 140 million since the start of the Covid-19 pandemic.

US home prices are now 47% higher than they were in early 2020.

US imports of Chinese golf carts surged from $148 million in 2020 to $916 million in 2023, a sixfold increase.

---

Short morsels to appear smart at dinner parties

📦 A new Meta AI model can generate 3D models from text.

👟 Carbon fibre supershoes are rewriting the running record books.

🥴 A BYD ad in the Euros accidentally sent Germans to a sex toy website.

🇨🇱 Why is Chile so long? 

📍 SandboxAQ unveiled a navigation system using AI and quantum for when GPS is jammed.

🪖 British war ships are getting laser guns. 

💾 The Japanese government is finally getting rid of floppy disks. 

---

Preview

This is a preview of my Saturday column this week.