Semiconductors are now more valuable than software
Welcome to Cautious Optimism, a newsletter on tech, business and power.
Tuesday. The shutdown continues, with political types saying that the “outlook” for it ending is “pretty grim.” Economic coverage of the fallout from the shutdown is neutral, with traders framed as waiting for more on its end. I expect stocks to continue rising because that would be the silliest response to the underlying news landscape.
Why do stocks keep going up? Partially thanks to the OpenAI dealmaking we’ve covered seemingly daily in recent weeks. The FT pegs total OpenAI computing agreements at more than $1 trillion over time, with its recent deals with Nvidia and AMD costing as much as $500 billion and $300 billion for a combined $800 billion price tag. OpenAI’s H1 revenue was $4.3 billion. That’s a lot, but not when you consider the scale of compute purchases that the company in question has agreed to buy. Surely Sam and company have enough compute now. Surely there won’t be more deals announced? — Alex
📈 Trending Up: Jobs growth … AI agents, maybe … app store rents … IBM? … survey consolidation … political bailouts … prison … pharma in India … Twitter …
Signal of the Day: Dell updated its earnings targets this morning, pushing its expected annual growth rate from 3-4% to 7-9%. Naturally, its shares are flying on the news, but Dell’s bump comes from the same AI infra boom that we keep talking about. “Customers are hungry for AI and the compute, storage and networking we provide to deploy intelligence at scale,” Dell said. No shit.
📉 Trending Down: US-Venezuela relations … the joys of flying … chances that Cerebras stays private … conversion therapy bans? … US arms sales to Europe … nuclear safety … Taiwanese independence …
Things That Matter
Semiconductors are now more valuable than software: Schwab’s Senior Investment Strategist Kevin Gordon shared a chart this morning with the tagline: “S&P 500 Semiconductor Industry’s price/sales ratio up to nearly 16x:”
That semiconductor stocks are trading well is known; Nvidia wouldn’t be the most valuable company in the world if that wasn’t true. But when we consider the basket of semiconductor companies trading at 16 times trailing sales, I think it’s fine to wonder if the cohort is not dramatically overpriced. Why? Because you can buy the Bessemer cloud index, a collection of SaaS companies growing at 20% on average for 9.0 times their trailing revenues.
Subscription software (SaaS) trading at a discount to semiconductor companies does make some sense if we zoom into the present moment. Chips are in short supply and margins are solid (Nvidia gross margins were above 72% in its most recent quarter). So, companies like the GPU giant are massively profitable today.
But unlike SaaS, chips are cyclical, prone to dislocations in their supply and market demand; given lag time between rising demand and constituent supply, you can wind up with too little supply (today) or too much supply (tomorrow?). Investors tend to value cyclical industries at a discount to their non-cyclical competitors, everything else held equal. Just ask United why it’s worth 0.54x trailing sales when it kicked off $2.2 billion worth of operating cash flow last quarter against a market cap of $31.4 billion. (The answer is that during periods of economic contraction, airlines are miserable businesses.)
I don’t want to spend all our time crying bubble, but semiconductors stocks are hot because tech companies are tripping over one another to buy compute to meet current and expected future demand. If that party slows, investors could be quickly reminded why you could snag the same semiconductor index for 3x trailing sales a decade ago.
POTUS wants to send in the troops: Continuing Fascism Watch this morning, here’s a rundown of the latest. After a judge blocked the White House from deploying the national guard to Portland, Oregon, POTUS said that he may invoke the Insurrection Act to get around legal blocks. Meanwhile, a different judge declined to block the deployment of military to Illinois, leading to the White House sending troops to Chicago. Military deployment into cities that did not vote for him, with the background chatter of key aides calling political dissidents terrorists, and the hyping of fake threats of disorder as real are not an attractive cocktail.
I’ve had friends recently visit Portland in the state of my birth. Reports thereof indicate that Portland is far from aflame, even if like all major American cities it has its issues. You begin to wonder if the President’s media diet may not afford him a full, fair picture of reality.
Apps in your AI, or AI in your apps? Yesterday, OpenAI announced new partners to bring apps into its AI chatbot. Here’s how the service works, per the company:
When you start a message to ChatGPT with the name of an available app, like “Spotify, make a playlist for my party this Friday,” ChatGPT can automatically surface the app in your chat and use relevant context to help. The first time you use an app, ChatGPT will prompt you to connect so you know what data may be shared with the app.
ChatGPT can also suggest apps when they’re relevant to the conversation. For example, if you’re talking about buying a new home, ChatGPT can surface the Zillow app as a suggestion so you can browse listings that match your budget on an interactive map right inside ChatGPT.
OpenAI wants to make ChatGPT the center of its users’ digital lives. That’s why the company is reportedly working on a browser; digital centrality is incredibly valuable, as Google shareholders have learned. But OpenAI is taking the inverse approach of legacy software companies, baking apps into its AI service instead of AI into its applications. This is mostly because OpenAI doesn’t have a legacy software suite whose cash flows it needs to defend (Microsoft, Alphabet), and because it’s currently not doing that much inside of an AI chat interface.
But with developer tools launched into preview this week so that more apps can be built into ChatGPT, expect the software companies of the world trip over themselves to join the fun. There are, the company announced yesterday, now some 800 million weekly active users of ChatGPT. That’s heading to one billion rather quickly — Q1 2026? — meaning that if you do something online, you are going to want to plug it into OpenAI consumer/prosumer audience.
At a minimum, apps in ChatGPT will earn usage share from existing methods of accessing app-ified services. That’s probably a net negative for traditional search services. The bigger question is how much capability you can create by bringing apps into an AI chatbot interface.
Here I get a little bit skeptical. Typing in all my requests feels fast when I need something that traditional software cannot provide. Commanding actions and requests via natural language prompt will feel slow as shit when we need to do things like “add one more row” or “wait go back” that were very quick via button clicking in the prior era of software.
So how good can apps prove in ChatGPT? It’s a multitrillion dollar question, given the value of software out there today. Value that OpenAI wants to help generate inside its own domicile.
After doodling that out, I saw that Stratechery made a similar point.
Are these prediction markets useful?
The owner of the New York Stock Exchange (NYSE) plans to put a staggering $2 billion into Polymarket, the well-known prediction marketplace that recently earned the green light to enter the US market. Lots of folks want a bite at the same apple. Kalshi competes with Polymarket, as does Robinhood, which added prediction betting to its product lineup earlier this year.
The investment makes certain strategic sense. Some of the hedging that individuals once did through specific shareholdings, or short-bets against select equities, can be executed via prediction markets. And, prediction markets offer up a wider array of events to wager on or against. We can therefore understand why an exchange might want to get in on the fun; if you help people trade already, why not help them trade more?
There’s reason to be bullish on demand, too. Robinhood is seeing impressive growth in its prediction market business, with volume rising to over two billion contracts in Q3 alone, or about half of the company’s total volume to-date: