Stubborn Bond Markets in the Face of Mounting Fiscal, Monetary, Geopolitical and AI Challenges

August 21, 2026

GARI STRATEGIC BRIEFING AUGUST 21ST, 2026

By Michal Kořan

If you were amazed by the pace of things in the past few weeks, I hazard a guess that the past four days must make your head spin even faster. Even more intriguing is how they all fit nicely together in to one eloquent mosaic.

1. Scott Bessent is learning the hard way that fooling the bond markets is a tad bit trickier than fooling the oil and the stock markets;

2. Broadcom joins NVIDIA in turning from a cash cow into a debtor;

3. The protection sellers do not like this turn – the CDS spreads over the key parts of the tech sector keep climbing, Oracle now reportedly at 219 bp, but it is the pace of the climb that should make people think;

4. As the fight for capital between the AI ecosystem and the US government picks up, OpenAI reportedly rushes to an IPO to get out of the door before it is too late, which increases the competition between the government and the AI ecosystem for money;

5. How to conclude the Iran “expedition” under the conditions of drying up funds and ammo?

6. (optional point) AI gullibility vs. human mistrust in the river of Babylon.

I. On Tuesday, Aug 19th, US Treasury Secretary Scott Bessent made a fresh attempt to rein in long-term borrowing costs from multi-year highs, sending Treasury yields and the dollar down. Just two weeks after releasing its planned schedule for buybacks this quarter, the Treasury Department on Wednesday said it’s “increasing, by at least double, the size of liquidity support buyback operations” for securities dated from the 10-year to the 30-year sector.

So essentially, the Treasury is infusing new liquidity into the system, helping the Fed avoid making any decision on rates (looking forward to following the Jackson Hole meeting next week), by buying its own debt to prevent said debt from becoming ever more expensive – and untenable. However, what the markets seem to be telling us is – we are not buying your story anymore, this is not about headlines and/or short-term liquidity, this is about the fundamentals and the fundamentals look creepy.

What I find interesting – and encouraging – is that it seems to be much more difficult to “fool” the bond markets as opposed to the stock and oil markets. I was definitely fooled – there was a lot of frustration over the course of the Iran conflict when it came to oil and stocks as nothing seemed to have made sense. The bond markets are tougher as they are (thank God) still linked to the fundamentals in a more meaningful way.

For example, in relation to algo-trading that has made the markets a bit of a mess lately – according to one study of CME order-book data, median order lifetimes after the initial Iran shock fell from 148 to 66 milliseconds in WTI oil, from 106 to 18 ms in S&P futures, and from 46 to just 7 ms in Nasdaq futures. Seven milliseconds! :-D

In these markets, it is easier to operate on headlines (themselves boosted by bots); thus huge parts of the markets rode on keywords, momentum, positioning and volatility that trigger algorithms which trigger other algorithms, rather than on sound analysis. But then there are US Treasuries. They are algorithmically traded too, but the bond market seems to be tied down by a much heavier set of “real-world” constraints: expected Fed rates, inflation, oil, SOFR and swaps, repo financing, TIPS, FX hedging, foreign yields, Treasury issuance. And ultimately, there is the rather inconvenient question for the US government: who is going to hold those trillions in marketable US government debt and at what price. Just to be sure to add the obvious – this is a condition that Trump´s administration largely inherited from the past 90 years of fiscal dominance in US politics and policy, but, at the same time, gravely helped by the Iranian adventure.

Amusingly, Mr. Bessent is trying to throw the right words around to see what sticks. During yesterday´s interview, Treasury Secretary Bessent went out of his way to promise fiscal consolidation. After the DOGE fiasco, and knowing that there is little that the US can realistically do in this area (hmmm, raise corporate taxes? :D ) Likely, the Secretary hopes that words such as “deficit reduction”, “fiscal discipline” and “consolidation” can feed directly into the models pricing the long end of the Treasury curve. The problem is that eventually somebody still has to buy the bonds. Which leads to the second point.

2. Broadcom joins NVIDIA in turning from a cash cow into a debtor. Bloomberg reports that Broadcom Inc. is in talks with lenders to raise more than $60 billion in debt for an AI chip financing deal that will benefit Anthropic PBC and other companies. This is a similar (albeit about ten times smaller) deal that NVIDIA made a few weeks ago. In simple terms – major hedge funds and banks (such as Blackstone, BlackRock, KKR, Apollo, Morgan Stanley) lend the needed capital to the tech companies, which then lend it to their prospective customers to buy their stuff. Since the Japanese and other carry trades that kept fueling the US stock and bond markets are a bit less certain these weeks, this looks more like a “bank of last resort” kind of action rather than a normal financing procedure. Where does the money ultimately come from this time? Yes – pension funds, insurance funds, and mobilizing treasury bonds.

This concept runs on a “bookkeeping”, but rather fundamental change – GPU and chips are no longer treated as computer components, but something akin to toll roads and factories: they are now long-term productive assets. However, this does not change that, a) in anything between two to five years, these parts will inevitably become obsolete, b) a loan is a loan, and debt is a debt, c) it runs on trust that the AI machine will start making money, preferably as soon as in 2027. Which brings China into the fray for the first time in this text – China is puffing out much cheaper and competitive models and solutions. This is not to say that China has won but even having the discussion about Chinese competitiveness in the entire race should make investors very worried.

3. And it does, which leads to the third point.

Credit default swaps (essentially insurance against default) climbed from essentially zero for the biggest tech companies to sometimes alarming highs (Oracle) in just a few months or became a thing at all for others (such as unsinkable NVIDIA). They are still far from the likes of Lehman Brothers in 2008 but the speed at which they are rising is certainly something to watch for. In any case, the money needed is now dug out from just about every and any corner, which leads to the fourth point.

4. As reported by Bloomberg this morning, both OpenAI and Anthropic (two companies whose success the entire current AI commodification cycle hinges on) seek to accelerate their IPOs. (At record levels, obviously, why be beaten by Elon Musk?) The acceleration seems to suggest, among others, that they both need to get out of the door before something major goes wrong, such as the realization that their valuation might, in the end, amount to very little. After all, reportedly, around 40 % of OpenAI revenues come from its mother, Microsoft.

Again, this is yet another sign of heightened competition for money between the government and the AI ecosystem. The government is in a very unenviable position here, though. It needs the money that AI is gulping down by the buckets, but it needs the AI to prosper – and fast. Let us list just a few points here (more to come on that topic in the next few days): First) AI, or so it is hoped, will bring enough fresh breath of productivity, thus helping out with the sovereign debt and bringing prosperity and a future for the entire American society.

Second, should the US succeed in “owning” the entire AI supply chain (see Pax Silica) from rare earths (see Greenland), through other commodities (see Ukraine), to data-centers, cloud and the AI itself, it would be a welcome addition, if not replacement for, the essentially free money flowing into the US economy as supplied for decades by a) the dollar serving as a reserve currency (see Mr. Vance calling reserve currency a “curse” yesterday) and/or b) the petrodollars which over the past decade or so have been threatened by multi-currency oil trade (see Venezuela and, maybe Iran /wink, wink/).

Third, the AI will also serve as a tool to beat China and others in the security and defense sectors. As you can see for yourself, there are a lot of eggs in one basket and to me, the basket seems to be more than dubious, but about that later. More importantly – the US Treasury faces an unresolvable dilemma – it competes for money with the same entity it desperately needs to succeed. Good luck with that. What would certainly help would be to get rid of the little problem called “the expedition into Iran”. For the time being it is clear that the US cannot engage in a major armed escalation – it would further unhinge bond yields, inflation, potentially damage the stock markets (who knows, though). Oh, and there is the ammo deficit. So the latest piece of strategy is to isolate Iran to death, according to a plan unveiled by Scott Bessent on Monday. The hurdle here is that for that plan to work, China would have to get on board, which leads to the fifth point.

5. The US is enforcing a naval blockade of Iran's ports and has already subjected Iran to decades of tough sanctions, so it's unclear what meaningful economic actions Washington has left. Perhaps the only leverage would be if China plays along – Treasury Secretary Scott Bessent said verbatim: "you are either with us, or against us" clearly speaking to Beijing. As Bloomberg wrote – rather than a show of strength, Bessent´s words reflect a worrying lack of US options to wind down the unpopular conflict. Beijing is the largest buyer of Iranian oil (around 90 %) and the US does not seem to have much leverage to change that.

It will be interesting to see how the US targets Chinese banks and such under the conditions, among others, that China still controls critical parts of the supply chains that make the manufacturing of much-needed weapons possible. The U.S. defense-industrial base remains materially dependent on China for a number of critical upstream materials, processed materials, magnets and lower-tier components. Another source of leverage that China has is the AI card – once again, should Beijing prioritize conflict over cooperation in the AI sector, this would result in throwing unpleasant buckets of sand into the US AI-financing machine. Which leads to the last point.

6) The last point is very optional and admittedly much more philosophical.

The current AI commodification cycle (LLMs, image and video generation) runs on trust that AI will actually be helpful and productive. I already see many key mismatches for this to pan out well, more on that in a new brief sometime in the near future, so at least a few points here.

Let us start with trust itself. We are living in a society where trust is a rare commodity. Words have increasingly different meanings for different countries, for different societies, for different communities within those societies, for different individuals within those communities. And we are speaking about the key terms, such as freedom, democracy, war, peace, prosperity, and equality. We are living in times where you can mold and fold each of these words into a context that makes the same word the exact opposite in the hands of two different people and unintelligible still for others. How do we expect any language model to make sense of this mess and to provide any meaningful guidance based on this?

Further on, as the volatile and often senseless oil and stock trading during the past six months showed, AI is very gullible when it comes to headlines and carefully manicured public trigger slogans. Not only is AI gullible, but more importantly it propagates its gullibility into society and the economy at large. It does so, for example, precisely through the algo-trading in the past months. Otherwise, I can hardly see oil prices below 100 USD, with so many oil reserves depleted and having anywhere between 5 and 10 million barrels offline per day. True, China surprised by its own demand destruction and quick shift to EVs and there was likely more oil slipping through the Strait of Hormuz, but it is not only oil, it is gasoline and diesel and all the petro-products we are talking about. After all, the high margins of refiners speak a more truthful story. What I am trying to say is that relying on a gullible AI (and making good money by it) is already having real-world consequences which will be the more painful the longer the mismatch between oil prices, demand, and reality goes on.

So what we have is a gullible AI on the one hand, increasingly skeptical humans on the other, and that is not all. There is another mismatch, still. Simplifying to the extreme, at least according to some metabolic efficiency theories, one way to think about intelligence is the ability to achieve a given goal while expending as little energy and resources as possible. In that sense, the level of intelligence can be thought of as something like the value of the achieved outcome relative to the energy spent achieving it. Such a goal is always stated individually, so for the AI to be truly productive and helpful, its key organizing principle would have to be “energy efficiency” for its user. Only – it is not. The key organizing principle of the current AI commodification cycle is set by its designers, and it is: to spend as little energy as possible at the aggregate level (computing capacity, electricity, etc.) to keep users engaged. Efficiency and productivity at the individual level are merely collateral outcomes here.

Third key point against the case of current AI as a future productivity vehicle is linked to the first one: economic and social productivity can only rise (or be maintained at all) if there is trust – in the form of institutions, laws, and general trust in society. Why is it so? Because with a lack of trust, instead of productivity everyone needs to spend more energy (capital, time) on hedging, redundancies, resilience. All these eat away at efficiency, thus at productivity. I think there is no one these days who would argue that society as such bathes in trust in the past decade or so. In fact, across almost all societal domains, trust is thin and running out. The more complex the system (society), the more trust is needed. That is because the more complex the society, the longer and more complicated the journey leads from one's action to a realized counteraction (in the form of punishment, or reward, let us call it responsibility). Responsibility becomes more abstract and with it the need for trust increases. AI and AI-fuelled digital social networks infuse unprecedented and exponentially growing amounts of complexity into our social and economic systems, thus proactively making sure that trust is evaporating on an hourly basis.

So, we, humans, find ourselves in a vicious circle – we need AI to increase future productivity, but productivity runs on trust, which the current AI cycle actively depletes. And I am not even writing about the lack of resources for the AI infrastructure, droughts versus cooling, etc. etc. So it seems that the only thing markets and some governments seem to trust is that AI will make things somehow better, but no one has seen it (at least so far). It feels more like belief rather than trust, for we live in societies where trust in institutions, politicians, markets, corporations, the future, and ourselves is thin and thinner still. We are trying to find a firm footing and prosper in the biblical Babylonian world, where the most fundamental words take on shifting meanings at an accelerating pace, and AI, which itself contributes to this state, is called upon and prayed to as the coveted remedy at the same time. Feels very much like history repeating itself.

About GARI Enterprise

GARI Enterprise turns global complexity into forward intelligence. Built on nearly a decade of R&D, GARI operates a proprietary digital simulation of the globalised world, connecting more than 70 billion data points across markets, trade, supply chains, logistics, geopolitics, climate, energy and the wider economy. Through GARI Market Lens, we provide forward commodity price projections, volatility signals and early warnings to support procurement, trading and strategic planning, alongside an interactive War Room for testing future scenarios and “what-if” events against real-world data. GARI also works directly with companies and institutions on tailored forward-looking analysis, scenario intelligence and strategic decision support — helping clients understand not only what is happening now, but what is likely to happen next and how to prepare for it.

Test it yourself — get one week of free access to GARI Market Lens and see what’s coming next.