January 27, 2025 | Read Online

"Que no haya novedad."

— Traditional benediction ("May no new things arise.")

Investing in a new thing: Abundance

The dotcom bubble burst when Cisco's customers stopped ordering twice as much networking equipment as they actually wanted.

The hardware required to build out the internet had been in such short supply that telco companies habitually exaggerated their demand, hoping that placing an order for, say, $100 million worth of equipment would give them a better chance of receiving the $50 million worth they desperately needed.

This practice came to an abrupt halt in 2000 when the market for networking equipment suddenly shifted from a state of under-supply to over-supply.

By the first quarter of 2001, Cisco was forced to acknowledge the new state of affairs, writing off $2.25 billion of inventory that couldn't be sold to customers.

This wasn't just a case of waiting for demand to recover — there was so little use for the piled-up equipment that some of it was sold to recycling companies that salvaged metal components and the rest was destroyed.

Cisco was only one company, but the demise of its order book marked the end of the dotcom bubble in part because the dotcoms themselves were all loss-making.

Only a small number of picks-and-shovel plays like Cisco were making any money from the internet — and suddenly when they no longer were, it was unclear to investors how anyone would make a profit from it.

I relate this ancient history (from memory!) because we may have experienced the Cisco moment of the AI bubble this morning: Shares of NVIDIA fell 17% after a new version of a low-cost LLM, DeepSeek, suddenly got people thinking that Nvidia's customers have been ordering far more GPUs than they really need.

It's unclear to me why this is happening today because the technology behind the DeepSeek version in question (R1) isn't all that new: Technologists have known about the cost-cutting innovation behind R1 since at least last January (even I had at least heard of it).

But this was the weekend people decided to worry about the consequences of a capable LLM being trained using older models of Nvidia GPUs at a cost of $5.6 million — a shocking number given the hundreds of millions that US companies like OpenAI have been spending to build only slightly better ones.

The market believes this will be a problem for Nvidia, which may now see less demand for its newer, more expensive GPUs.

This could also become a problem for the market generally, because we now have to figure out whether anyone other than the picks-and-shovels infrastructure plays can make money from AI.

I think they will, simply based on my memory of the dotcom bust: An oversupply of GPUs in 2025 may play out much like the oversupply of networking equipment in 2001.

Cisco's bad news on orders was ultimately great news for companies like Facebook, Apple, Netflix and Google, which were enabled by the cheap and plentiful broadband that followed.

So, just as the FANG bull market of the 2010s rose from the ashes of the dotcom bust, cheap and plentiful AI may similarly enable all-new use cases that could prove immensely profitable for investors.

The transition from an infrastructure market to a use case market can be rocky, however.

AI has so far been a Field of Dreams kind of thing for investors, who haven't asked too many questions about how the tech industry would recoup the hundreds of billions they've spent on AI infrastructure — we've just trusted that if they build it, the use cases will come.

The advent of DeepSeek suggests that the use cases might get here faster than expected, so investors have to start thinking about what they might be and whether they will be profitable.

This is a much harder game than identifying the picks-and-shovels winners of an infrastructure buildout.

Even if you think AI will become both ubiquitous and beneficial, it's hard to foresee whom the value will accrue for.

"The stock market measures the present value of profits, not the present value of social benefits," John Cochrane wrote this morning. "The profit, and ultimate benefit, of railroads was not so much in the railroad itself, but in the wheat fields of Kansas."

The stock market was down today because fields of dreams are easier for investors to figure out than fields of wheat.

Abundant crypto

Ben Thompson predicted in today's Stratechery note that abundant AI will be bullish for the likes of Apple, Amazon, Meta and Microsoft, which will perhaps have much lower capex bills than we had expected: "In the long run, model commoditization and cheaper inference…is great for Big Tech."

In the short term, however, "the market is working through the shock of R1's existence."

Part of the shock, I think, is a recognition that a substantial portion of the huge valuations we assign to mega-cap tech companies is attributable to the idea that they are the gatekeepers of AI.

But DeepSeek is an open-source model, which likely makes it harder to keep the gates closed.

We've witnessed this in crypto, where John Pfeffer's 2017 thesis has proven prescient: "The open-source nature of public blockchain protocols, combined with intrinsic mechanisms to break down monopoly effects, mean that the vast majority of this economic surplus will accrue to users."

Pfeffer predicted that business moats would not be as sustainable in crypto as they are in traditional companies: "While tens or perhaps hundreds of billions of dollars of value will also likely accrue to the cryptoassets underlying these protocols and therefore to investors in them, this potential value will be fragmented across many different protocols and is generally insufficient in relation to current valuations to offer a long-term investor attractive returns relative to the inherent risks."

That does seem to be how things are playing out, which does not bode well for the crypto investors currently assigning huge valuations to projects that aren't generating much revenue.

AI seems likely to expose these valuations: As the cost of infrastructure plummets, investors will have to find real use cases to invest in, not just picks-and-shovels.

Today's market kerfuffle over DeepSeek suggests that, at least in equities, the search for these use cases is on.

I hope it will be in crypto, too, because that is arguably why we're here: Cutting out the gatekeepers so that more value flows to users and less to owners.

Of course, investing in use cases can be profitable for investors, too, as was the case when the FANG bull market rose from the ashes of the dotcom bubble.

But the history of investing also shows that change can be hard — even in markets, "may no new things arise" qualifies as a blessing.

— Byron Gilliam

The stakes have never been higher. As markets evolve and policy takes shape, Mohamed El-Erian will cut through the noise on macro trends. Anthony Scaramucci will unpack the changing face of finance and crypto. Kristin Smith will lay out the hard truths about regulation and its ripple effects.

📅 March 18-20 | NYC

Don't miss it

Abstract's Creator Flywheel

Luca Netz discusses the upcoming launch of Abstract Chain and its focus on UX. Get his thoughts on Celestia for DA, valuing intellectual property onchain and where L2s and NFTs are headed.

Listen to 0xResearch on Spotify, Apple Podcasts or YouTube.

Eric Balchunas @EricBalchunas

A 2x Melanie ETF before a 1x Melania ETF has been filed. That is unusual. Also first filing to look to track chainlink, Cardano, Polkadot, BNP. All 2x tho.

tw profile: James Seyffart James Seyffart @JSeyff

NEW: @TuttleCapital just filed for 10 different leveraged crypto asset ETFs. Including a bunch of memecoin products and assets that don't have ETPs yet.

9:08 PM • Jan 27, 2025

93 Likes 15 Retweets

23 Replies