Joe Wiesenthal on tariffs, the Moody's downgrade, rising global bond yields, and why the AGI narrative feels like a business story now

Summary

Joe Wiesenthal, host of Bloomberg's Odd Lots, argues that the AI safety narrative is effectively dead as a market force — replaced entirely by competitive product pressure and business logic.

Tariffs and the Moody's downgrade

The 10% baseline tariff that survived Trump's April rollback is not nothing. Wiesenthal draws the Brexit comparison deliberately: the UK is still a wealthy country, but it hasn't grown the way it might have. America could be on a similar trajectory — the damage slow and ambient rather than dramatic. He dismisses the Moody's downgrade as largely irrelevant since the US borrows in dollars it can print, but flags two genuine risks: a stagflationary policy mix of tariffs plus tax cuts is bad for Treasuries, and the perception that US policy can be reversed unilaterally at any moment is, at the margin, eroding international willingness to invest in dollar assets.

Rising global bond yields

Wiesenthal previews an upcoming Odd Lots episode with Steven Englander of Standard Chartered. The thesis is structural: governments worldwide are being forced to spend more — Germany and others can no longer assume US security guarantees — while tariffs are making supply chains less efficient and more inflationary. Central banks will need to keep rates elevated to hit inflation targets. Englander thinks the US may get short-term rate cuts, but his base case is that the window closes fast and rate hikes could be back on the table by 2026.

Middle East deals and Jensen Huang

Wiesenthal frames the Trump Middle East tech deals through a Cold War lens: what if stable geopolitical relations are now defined by who you agree to sell chips to, rather than shared governance values? He's not strongly opinionated on the foreign policy question, but from a pure business standpoint, Jensen Huang traveling the world and signing hundred-billion-dollar GPU deals while boxing out Huawei in the Gulf reads as straightforwardly effective.

AGI as a business story

The safety framing that defined AI discourse two years ago feels like yesterday. The market is dictating pace — companies race ahead not because of ideology but because competition leaves no alternative. DeepSeek displaying its reasoning forced OpenAI to respond in kind. That's product competition, not mission.

Wiesenthal's more pointed observation is that OpenAI's recent moves — hiring a CEO for product and revenue, reportedly exploring a social network, buying a browser — look like investments in building a profitable consumer internet company, not like a lab racing toward transformative AI. He thinks OpenAI could plausibly pivot to openly calling itself a consumer internet company, leaving other labs to carry the AGI narrative without that commercial cushion.

His steelman for the AGI narrative is narrower than most: the sci-fi framing may have been necessary to generate the nonprofit investment and unprofitable research that produced genuinely useful consumer and enterprise technology. The nuclear analogy cuts both ways — the bomb was the bad ending; cheap, abundant nuclear power was the good ending that never arrived. He hopes AI follows the better path.

Google's product problem

Wiesenthal argues Google's core issue isn't research quality — it's access. ChatGPT ships a feature and users find it at chat.com. Google releases something extraordinary and buries it inside a fragmented product map of Gemini apps, research previews, and NotebookLM. The structural read is that Google, unlike Zuckerberg at Meta, doesn't have the ownership concentration to simply declare that Google.com is now an AI product and force the transition.

Meta's AI rationale

On what Meta actually gets from its AI spend, the answer is more legible than it looks. The algorithm serving ads on Instagram is already a large transformer model — a training run comparable in scale to GPT-3 or GPT-4, just optimized for behavioral prediction rather than language. The Reels ranking system is a second major vertical at similar scale. Zuckerberg built double the data center capacity after being caught flat-footed by TikTok's algorithmic shift, ended up with surplus compute, and used it to train Llama. Internally, LLMs now handle payment fraud detection, content moderation for minors, customer support, and dozens of other non-consumer-facing tasks — workloads Meta would otherwise be paying OpenAI or Google to run. The chatbot UI paradigm isn't a natural fit for Meta's social products today, but as image generation matures toward token-based architectures, the use case shifts from generating AI slop to natural-language photo editing, which fits the Instagram surface much more cleanly.