TheChek News Flow

And rising corporate bankruptcies

Mag 7 Capex

Accounting is boring but important. Particularly important: the difference between a capital expense and an operating expense. A capital expense (buying a big piece of equipment, say) does not count directly against earnings on the income statement, as an operating expense (paying a salary, say) does. Instead, a capital expense appears on the income statement over time, in theory matching the drag on profits to the life of the capital asset. This spread-out expense shows up in a line called “depreciation”.

I see you sleeping at the back. But I drag you through this tiresome point because the most important companies in the world, the Magnificent 7 Big Techs, are running up a huge amount of capital expenditure, mostly on data centres for artificial intelligence. This is cash out the door today, but the expense will only appear in earnings per share over time. The AI arms race has not fully hit profits yet. The question is whether the market has digested the fact that it must do so before long.

Here is capex at the five of the seven that are, to greater or lesser extents, going bananas on capex (at Apple, capex is steady; at Nvidia, the capex money is coming in, not going out):

These are staggering numbers, and they are still growing.

Amazon looks like an outlier, but that is not quite true. One needs to scale the spending to the rest of the company’s financials. For example, one can look at capex as a percentage of revenue:

At Meta and Microsoft, one of every five dollars that comes in the door goes out as capex; at Alphabet, it is one in seven. And the trend is up (wondering about that big mountain of spending in 2022 at Meta? Remember the Metaverse?).

Now we have a sense of the scale of the spending. But what is important for our topic is the scale of the capital spending compared to what is currently being charged against profits. That is, what is capex relative to depreciation expense? As of the past 12 months, here is what that looks like:

Take Meta for example: depreciation expense is 9 per cent of revenue, capex 20 per cent. That means depreciation expense has to go up significantly in the coming years. That does not mean the group’s operating margins have to eventually fall by the difference (11 per cent); data centre capital expenditures will be spread over four to five years. But if the current level of spending keeps up, the drag on margins will be significant.

So, how much margin compression do Wall Street analysts’ estimates price in for these five companies in the next few years, as AI capex comes home to roost? None whatsoever. At all five companies, in fact, operating margins are expected to expand a bit in the next few years. This is possible, of course. Except for Tesla, all of these companies are increasing revenues quickly, and there is plenty of operating leverage in their models. But it won’t be easy. The optimism about Big Tech profits momentum overcomes all.

BigTech Capex

More on Big Tech capex

Several readers wrote in response to yesterday’s piece about capital spending and depreciation expense at the big US tech companies. The gist of their comments was that the market sees through accounting, and looks at cash flow. That GAAP profits of the Big Techs are not representative of how much they are spending doesn’t really matter.

This is right, but only to an extent. Here is free cash flow (operating cash flow less capital expenditures) at the five companies we looked at yesterday:

These companies (except for Tesla) still generate a boatload of free cash, even after shovelling billions at AI data centres. But notice that the trend in cash generation is flat to down. This is easier to see when you look at them in aggregate:

Now, one might look at this and say, “cash flow is cyclical, and the industry has been through down cycles before, like in 2021-2022, no big deal”. And there is some truth to that. But it is worth remembering that during that part of the cash cycle, all these stocks underperformed the market (and except for Microsoft, the underperformance was significant). 

Of course, the important thing is what happens next — how long the wild AI spending continues. It’s hard to say. Meta, Amazon and Alphabet have all suggested that capex will be higher in 2025 than in 2024. We’ll find out more from fourth-quarter earnings reports. 

In terms of earnings per share — as opposed to cash flow — the future is doubly hard to see, because we don’t know how exactly these companies depreciate data centre assets.

Ravi Gomatam of Zion Research Group, which specialises in accounting issues, emphasised to me that modelling future depreciation at big tech companies was highly complex. The companies provide only high-level information about their investments, and those high-level numbers can obscure a lot. Take data centre spending: what part is servers? The non-server infrastructure? The building? If treated individually, each of these would be depreciated at different rates; or they could be depreciated together. On top of that, there is the question of how depreciation rates might be changed, or writedowns taken, when computer equipment is made obsolete by innovation. A lot of assumptions and back testing is required to come up with a reliable projection.

What we can say for sure is that investment, depreciation and cash, and not just AI hype, will matter for tech investors this year.   

The future of QT

Quantitative tightening

Ten-year Treasury yields, at almost 4.7 per cent, are at their highest level since last April. No one likes this. Treasury investors have lost money. Equity investors will see a threat to stocks’ high valuations. Homebuyers foresee more expensive mortgages. And for government officials, the bond market is signalling discontent with fiscal policy.

The causes are mixed. The market seems to be absorbing recent strong economic news, which suggests that inflation might stay a bit high. Investors may be coming to terms with the possibility that full Republican control of the government might lead to bigger deficits. At the very least, the bond market is pricing in rising uncertainty. 

Structural changes in the Treasury market may have a role, too. The Treasury has been issuing a larger share of short-interest debt than in past years, which has made longer-duration Treasuries more scarce. This may have supported prices and driven down yields. Some observers have characterised this as a “secret” monetary policy tool — “quantitative easing by other means”. Treasury secretary Janet Yellen has denied this claim. Incoming Treasury secretary Scott Bessent is said to favour issuing more long-duration debt. It’s possible the market is anticipating Bessent’s approach and selling long bonds.

The rise in yields may subside. But if it persists, adding to the stress in the financial system, what will happen to monetary policy? If inflation does indeed hang around, the Fed will have to keep rates higher for longer. But what of quantitative tightening, or the effort to shrink the bank’s balance sheet after the Covid-19 pandemic response bloated it? Might the Fed halt it to reduce stress on the financial system? Might they even reverse it, and start buying Treasuries — quantitative easing — again?

Unhedged has not written about QT in a while for a simple reason: it’s been working. After QT contributed to 2019’s repo crisis, many feared that the newest QT cycle, started in 2022, may lead to similar panics. But things have been smooth, and there is good reason to think they will remain so.

For starters, QT does not seem to have had a major impact on long yields, so ending it would not give the Treasury market much relief. By buying up Treasuries, the Fed raises the price of Treasuries and depresses yields during quantitative easing (though the size of the impact is debated). At the start of this cycle, many feared that QT would have the opposite effect. But the impact appears to have been more muted: a paper by Wenxin Du, Kristen Forbes and Matthew Luzzetti estimates that it has boosted yields by roughly 8 basis points. 

Next, the Fed would only end QT if faced by a major liquidity breakdown in capital markets, and a jump in Treasury yields does not a crisis make. From Darrell Duffie of Stanford University:

The Fed only buys up securities when interest rates are 0 to stimulate the economy, or when the market for Treasuries gets clogged up. It’s not that rates are high or low, it’s that the market is not functional because there are too many investors selling Treasuries . . . [even] if Treasury yields get very high, the Fed won’t move unless there is market dysfunction. 

QT may be reaching its natural end soon, in any case. There is no “right place” to stop QT — the Fed is feeling its way, or “learning by doing”, in Ben Bernanke’s words. Ideally, QT stops at a point where banks reserves have normalised, but there is still enough liquidity in the system for markets to function smoothly. One approach to finding this level is looking at the sum of bank reserves held at the Fed and balances in the reverse repo window (RRW), or bank reserves plus “the excess reserves that banks can access if they are willing to pay market rates”, according to Joseph Wang of Monetary Macro. Together, the two represent the total amount of money available to the banks on short notice. When the repo crisis hit in 2019, that number was around 8 per cent of GDP. Getting close, but not too close to that number, is a reasonable goal. And we are probably not too far off: the sum of the RRW balance and reserves is around $3.4tn, and 8 per cent of Q3 GDP is $2.3tn. 

Most analysts we spoke with expect QT to end this year — but of natural causes, rather than market stress.

The $20 million deal gives the chip supplier a toehold in the life-sciences market, a strategy it plans to replicate in other industries

Advanced Micro Devices is investing $20 million in Absci, a drug-discovery company based in Washington state, in a move aimed at selling its artificial intelligence chips in the healthcare sector.

The deal is structured as a private investment in a public equity and includes an equity stake in Absci. AMD didn’t disclose the amount of its stake.

AMD, which is based in Santa Clara, Calif., said the investment and partnership between the two companies will help reduce hardware costs and optimize AI solutions for Absci.

The move is AMD’s first attempt to gain footing in life sciences with its AI chips, a space also targeted by its rival Nvidia. In 2023, Nvidia invested $50 million to boost Recursion Pharmaceuticals’ AI-based drug-discovery efforts and provided the underlying hardware for that work.

Mark Papermaster, AMD’s chief technology officer, said its relationship with Absci is the first of many efforts to make its graphics processing units, or GPUs, available to certain industries.

“We’re now expanding our focus into vertical markets and prioritizing healthcare, where we can immediately have an impact on society,” Papermaster said.

For AMD, investments in companies, including Absci, are a part of how it aims to make headway in the GPU market, which is dominated by Nvidia. AMD in December said it was part of a $333 million financing round for cloud company Vultr, for which it aims to become the “preferred” AI hardware provider.

As part of AMD’s investment in Absci, the company will move toward greater use of AMD’s GPUs, said the company’s founder and Chief Executive Sean McClain. Absci currently uses over 470 AI chips, most of which are Nvidia’s GPUs. It will start shifting some of its AI drug discovery workloads to AMD’s GPUs, he said.

Drug discovery efforts using AI require immense amounts of computational power—a constraint that Absci quickly ran up against, according to McClain. That is partly why AMD’s partnership and investment were attractive, he said, because it presented an opportunity to lower Absci’s costs of inference, or the costs of using AI models.

“We’re starting to see this big shift from designing drugs in the wet lab to now designing drugs on AI, and that means compute is extremely important. Our compute spend has skyrocketed,” McClain said.

Absci is also working with AMD to develop hardware and software that will better serve the healthcare sector and AI-based drug discovery work, he added.

The company has about 160 employees and has now raised a total of about $567 million. Absci said it plans to use the new funding to continue building its AI models and to keep working on its internal drug development efforts. Starting in the second half of the year, it said it would have results from a clinical trial for a drug targeting inflammatory bowel disease.

The new agreement puts a stronger focus on obesity and type 2 diabetes

Novo Nordisk NOVO.B 3.65%increase; green up pointing triangle expanded a deal with Valo Health to discover and develop treatments for obesity, type 2 diabetes and cardiovascular disease using human data and artificial intelligence.

The deal extends an agreement signed in 2023 and will see Valo become eligible for increased payments and funding.

Under the original deal, the companies agreed to develop up to 11 drug programs, primarily focused on cardiovascular disease, with Valo eligible to receive up to $2.7 billion in milestone payments, plus research and development funding and potential royalty payments.

The new agreement set out Wednesday expands the scope to put a stronger focus on obesity and type 2 diabetes and includes near-term payments to Valo of up to $190 million.

A further $4.6 billion in potential milestone payments will be made for up to nine new drug programs and Valo will also be eligible for more research and development funding and potential royalty payments.

The companies will continue to use Valo’s drug discovery and development platform that uses patient data and AI to generate new insights and translate them into potential therapeutics.

“We have already begun to realize the potential of combining the capabilities of Valo and Novo Nordisk to advance multiple AI-powered, human-centric programs, and we believe this partnership will help Novo Nordisk fulfill our ambition to expand the number of new drug programs we bring to the clinic,” said Marcus Schindler, executive vice president and chief scientific officer of Novo Nordisk.

The Danish pharmaceutical giant recently reported disappointing results of a closely watched clinical trial testing an experimental anti-obesity treatment that the company hoped would be its next big weight-loss product.