How to Hedge the AI Boom and the Unrest in France | Robert Armstrong

Amazon.com

Across the shows: 31 speakers in other episodes, 39 bullish and 7 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBullish

  Amazon is among the most successful businesses ever, funding AI capex from strong cash flow.

  Armstrong names Amazon among the hyperscalers that are among the most profitable and successful businesses in history. Their scalable, high-barrier businesses with low incremental costs generate the cash flow they are now using to finance AI infrastructure. He sees this as a much stronger starting point than the leveraged borrowers of earlier bubbles.

  “you have Microsoft, ... Alphabet, Amazon, Meta, these are truly, depending on how you measure it, the most successful businesses the world has ever seen.”at 18:06

Meta Platforms

Across the shows: 61 speakers in other episodes, 49 bullish and 33 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBullish

  Meta is among the most successful businesses ever, funding AI capex from strong cash flow.

  Armstrong groups Meta with the most profitable businesses in history. Scalable operations, very low incremental costs and extremely high barriers to entry make these companies very profitable. He says they have committed their cash flow to AI infrastructure, which makes the current boom financially stronger than the 1990s or the 2008 credit bubble.

  “you have Microsoft, ... Alphabet, Amazon, Meta, these are truly, depending on how you measure it, the most successful businesses the world has ever seen.”at 18:06

Microsoft

Across the shows: 33 speakers in other episodes, 33 bullish and 5 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBullish

  Microsoft is among the most successful businesses ever, funding the AI build-out from strong cash flow.

  Armstrong argues Microsoft is one of the hyperscalers that are the most successful businesses the world has ever seen. Their highly scalable models, very low incremental costs and extremely high barriers to entry make them money-printing machines. These firms have gone all in, using their cash flow to finance AI data centers. That financial strength is why he sees the AI boom as starting from a better place than past bubbles.

  “you have Microsoft, ... Alphabet, Amazon, Meta, these are truly, depending on how you measure it, the most successful businesses the world has ever seen.”at 18:06

Alphabet

Across the shows: 46 speakers in other episodes, 34 bullish and 26 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBullish

  Alphabet is among the most successful businesses ever, funding AI capex from strong cash flow.

  Armstrong lists Alphabet among the four to six most profitable businesses in history, with scalable models, low incremental costs and very high barriers to entry. He says these hyperscalers have gone all in with their cash flow to finance AI infrastructure. That financial base is why he sees the AI boom as financially sounder than past euphorias.

  “you have Microsoft, ... Alphabet, Amazon, Meta, these are truly, depending on how you measure it, the most successful businesses the world has ever seen.”at 18:06

NVIDIA

Across the shows: 51 speakers in other episodes, 41 bullish and 15 bearish calls, past 30 days

• Robert Armstrong · Financial TimesBullish

  Nvidia is the AI bottleneck, collecting most AI spend and recycling profits into the ecosystem.

  Armstrong argues GPUs are the bottleneck everyone needs, so a large share of AI spending, he guesses 50% to 80%, passes through Nvidia's balance sheet. Nvidia is investing some of that money back, guaranteeing data center leases and investing directly in OpenAI and Anthropic. He uses the gold rush analogy: the picks-and-shovels seller is now also buying mines. He says this circular finance is not terribly dangerous, but it is worth noting that profits at this stage are being recycled into later-stage bets.

  “Well, in this case, the picks and shovels company, Nvidia, is actually buying mines too.”at 21:00

AI & jobs

• Robert Armstrong · Financial Times

  AI job losses are not yet visible; past tech transitions did not raise unemployment, though AI may differ.

  Armstrong says nobody has been fired by AI yet, judging by headline statistics, even though there are anecdotes. Programmer employment is holding up. Youth hiring is harder, but he does not think that is clearly an AI effect, and it is not massive. History shows new jobs appear and transitions vary, and past technological revolutions did not raise medium-term unemployment. He admits AI as a general-purpose technology may differ and could permanently damage employment, so he calls it a wait-and-see.

  “nobody's been fired yet, if you believe the headline statistics.”at 30:26

• Robert Armstrong · Financial Times

  Commercial driving jobs will eventually all be taken by machines, with no clear destination for workers.

  Armstrong says driverless 18-wheelers already run on the road and do a good job, being safer and able to drive 24 hours. Local routes are still too hard for the technology, but it will eventually handle them too. Driving is a large, reasonably paid, medium-skilled source of US employment. He thinks all commercial driving jobs will sooner or later be done by machines, and he doesn't know where those workers will go, perhaps into service jobs.

  “all the commercial driving jobs sooner or later should be taken by machines because the machines will do them better and more safely.”at 34:40

• Robert Armstrong · Financial Times

  AI will eliminate many driving and service jobs; even a benign transition will create lasting political upset.

  Armstrong expects many jobs to disappear, with driving and service jobs more important than factory jobs, and data centers in the mix. He says even if displaced workers find new jobs quickly, which he calls very optimistic at about six months, changing jobs is hard and the transition is very uncomfortable, with uncertainty about self-worth and finances. He argues that even in an economically benign scenario, a long or short period of displacement could make many voters upset and dominate the political narrative. He applies this to his own job as a journalist: a computer may already do part of it and will likely do it as well later, so he worries about a forced job change.

  “A lot of jobs are going to disappear. We can talk about driving jobs. We can talk about factory jobs. ... Driving jobs are probably more important. Service jobs are probably more important.”at 38:51

AI infrastructure & data centers

• Robert Armstrong · Financial Times

  The AI capex cycle is the biggest in US history, about 3.6% of GDP over eight years.

  Armstrong says the AI investment boom is bigger than previous capex cycles, as the economist's estimate shows. He has checked the methodology of a Brookings paper by an NYU professor and finds it solid. The estimate multiplies the cost per megawatt of AI computing capacity by the megawatts that must be built, then compares that to US GDP. The result is about 3.6% of GDP over eight years, roughly a percentage point bigger than earlier cycles.

  “He's the guy who has come up with this number 3.6% of GDP over an eight year period. ... It seems like a pretty solid piece of work.”at 12:39

• Robert Armstrong · Financial Times

  Data center permitting is getting harder, worsened by a vetoocracy and the AI industry's PR.

  Armstrong, citing Fukuyama's idea of a vetoocracy, argues that many people have a veto over building infrastructure in the US, which weakens the social contract. He says his team has written that political risk around data centers keeps rising and building them gets harder. He attributes part of this to the AI industry's bad PR. Labs talked up existential risk to recruit engineers, and that has fed local opposition.

  “my team has written a little bit about the political risk around data centers, that it becomes harder and harder to build these things.”at 26:40

AI bubble & tech valuations

• Robert Armstrong · Financial Times

  The AI boom is financed by the most profitable businesses ever, a stronger starting point than past bubbles.

  Armstrong contrasts today's AI spending boom with the 1990s bubble and the 2008 credit boom. The key difference is that the spending is funded by the four to six most profitable businesses in history. These firms have highly scalable businesses, very low incremental costs to grow and high barriers to entry, so they can fund capex from cash flow. He compares this to a rich homeowner levering up with a strong income and lots of equity, versus the weaker borrowers of 2006-2008. He says the starting point is financially much better than in previous euphoric episodes.

  “The most important one is that standing behind this great adventure in how much money you can spend is the four to six most profitable businesses in the history of the world.”at 17:51

• Robert Armstrong · Financial Times

  AI boom is not near collapse; committed cash flows and profitable backers give it legs.

  Asked whether the cycle can keep going, Armstrong gives a hesitant yes. He says the euphoria has legs because there are committed cash flows, data center and chip contracts, and profitable companies involved. He stresses that contracts can break if people decide the compute is not useful, and that mood is ultimately what drives outcomes. He says he is not one of the imminent-collapse people.

  “my hesitant answer is yes. And my tendency to think is that this euphoria period has legs.”at 22:50

Credit markets

• Robert Armstrong · Financial Times

  Widening CDS on data center firms shows the odds of the AI build-out going wrong have risen.

  Armstrong says the recent spike in CDS spreads on data center firms shows that the marginal chance that not everything goes as expected has risen. He notes that CDS prices speculation as well as hedging, so you cannot read exact default probabilities from them. But the move shows the volatility of opinion: a perceived 1% chance of the situation going badly may have moved to around 2.5%. In a complex, opaque industry, he argues that this change is meaningful and must be reflected in prices and in animal spirits.

  “Those have spiked. I think what that's showing you is that the marginal chance that not everything is going to go as expected has gone up.”at 16:33

Power demand & the grid

• Robert Armstrong · Financial Times

  Energy may become the next AI bottleneck after GPUs, though he has not studied it.

  Armstrong concedes he misspoke in saying GPUs were the only bottleneck. They are the bottleneck now, but the next one might be energy. He reads a lot that says so but has not done the work, so he treats it as another area of speculation. He does not give a firm view on the timing or severity.

  “The next bottleneck might be energy.”at 24:42

AI regulation & safety

• Robert Armstrong · Financial Times

  AI labs refusing to accept liability for harms will backfire if a real accident happens.

  Armstrong says AI leaders have never committed that if the technology does harm, the buck stops with them. He thinks they avoid this because it would make them liable and hit their equity valuations. He argues they may think society will handle the consequences, but if there is a real accident they will be finished, with new CEOs and liability. He says it would be better to stake their reputations and wealth on keeping people safe.

  “They may think they can get away with that. I don't think they can. If there's a real accident, they're done.”at 28:40

Robotics & autonomous vehicles

• Robert Armstrong · Financial Times

  Driverless trucks already do long-haul well and will eventually take local routes.

  Armstrong says there are already 18-wheelers on the road with no driver. As far as he can tell they do a great job, are safer, drive 24 hours and never speed. Local routes are still too hard for the technology, but for long haul the trucks already perform well. He expects them to come for local routes eventually.

  “for the long haul, they already do a great job and they'll come for the local routes eventually.”at 34:14

Manufacturing & industrial policy

• Robert Armstrong · Financial Times

  China shock cost manufacturing jobs but was not catastrophic; technology drove most of the loss.

  Armstrong cites FRED data: US manufacturing employment fell from about 16-17 million around 2000 to about 11.5 million by 2010, a loss of roughly 5 million jobs. Manufacturing output did not fall, and the job loss was about 4 million of total employment of around 140-150 million. He concludes the China shock was significant but not catastrophic for the manufacturing base, and that most of the job loss was probably technology and would have happened anyway. He sets this against the political narrative of a left-behind rust belt.

  “So 5,000,000 jobs went away. As a corollary to that, manufacturing output did not go down in America. ... it doesn't look like the China shock was a catastrophic event for the manufacturing base in The United States. ... Most of it was probably technology. Most of it was going to happen anyway.”at 37:15

US elections & politics

• Robert Armstrong · Financial Times

  US anger is more than economics; negativity sells, and the current president exploits it.

  Armstrong notes that the unemployment rate is low and real earnings are not falling fast, so economics explains only part of the anxiety. He argues that negative stories sell in journalism, and that a politically savvy president has grasped this more fully than any predecessor. Combined with an atomized society, where people are less connected through church, work, and community, he thinks this drives the anger in the country.

  “I feel like we have politics now and in particular, a president who has grasped more completely than any president before the political analog of that, which is that negativity really sells, especially to an atomized country.”at 47:31