# Dark Fibre & Dark Silicon — Unsolicited market thoughts from an ex-hedge fund tech & hardware analyst, who's very out of practice — retained record

> Retained reddit record from reddit:ValueInvesting.

Canonical: https://highsignal.app/data/records/faa09b7b45965a27

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ValueInvesting)  published 2026-11-08 19:06 ingested 2026-11-09 02:38

[original source ↗](https://www.reddit.com/r/ValueInvesting/comments/1xr69cz/dark_fibre_dark_silicon_unsolicited_market/)

## retained text



```
I wanted to take some time this morning to share a few ideas that I feel are increasingly relevant in today's market. I'm a former HF research analyst — I covered technology & hardware. I'm a little less attuned than I used to be, but I thought I would share anyway. Please forgive my lengthy piece — it’s a bit of an essay. I am always happy to answer questions or discuss things further. This is not meant to be confrontational to those who disagree — it's merely meant to offer a different angle.

**Commentary & Context**

I have seen a lot on here comparing the current AI dynamics to the Dotcom bubble. At the same time, I have seen many on here dismiss the idea that they share any similarities — most often from the position that the Dotcom bubble was driven by profitless, high-growth business models.

I come from a different angle. I believe the two share exceptional similarities, but not in the way most think, even in the case of former Fed Chair Jerome Powell.

Between 1993 and 2002, $500bn was spent on a fibre optic buildout by US companies — roughly half was spent by new alternative carriers, where eight brand new national networks were laid from scratch. By the fall of 2001, over 100mm miles of fibre had been laid during the previous two years. The main drivers of the bull market run were not the unprofitable internet startups but the network equipment providers that enabled the fibre buildout (Four Horsemen) — they contributed \~30% of the NASDAQ’s gains that decade. By 2002, less than 5% of the total network capacity that had been laid in the previous decade was being utilized. Adoption rates among the general populace were far below what investors and the telecom giants had anticipated, not to mention the incredible duplication among the hyperscalers. All of this to say, for those who have not made the connection yet — that the current AI buildout and the telecom buildout share incredible similarities. Maybe a better name would be the Telecom bubble rather than the Dotcom bubble. This is to say that there is an extreme likelihood that we have overbuilt data centre capacity and that, ultimately, it will be underutilized in the proceeding decade.

Chip vintages also play an important role in this. If you watch the Satya Nadella interview, you would have heard the concern that hyperscalers possess too much of the same vintage of data centre capacity that isn’t necessarily forward-compatible with the next generation of Nvidia chips. This is the key problem with CoreWeave. Their business is incredibly capital-intensive, and it’s not feasible for them to continuously purchase the next generation of NVDA chips without using some erroneous accounting practices to improve on-paper performance.

None of this is to say that AI won’t be revolutionary or not have an incredibly profound impact — because after all — did the internet not also have a profound impact on us?

**Further Concepts — bubble dynamics**

Capital Cycle Theory & mean reversion. As a new industry grows and attracts windfall profits, it garners new entrants. These new entrants increase competition, driving down prices. The decrease in pricing power dampens margins — and over the long run — decreases profitability to the point where returns on capital fall below the cost of that capital. What typically follows is a slow burnout of market competitors and consolidation of the industry’s assets until the asset base can eventually generate returns that meet the cost of capital again. History offers us many examples of this — John D. Rockefeller’s consolidation of the oil industry in the early 20th century, which created the first modern monopoly, is a good one.

Further to the idea of a company’s ability to generate returns on its asset base is Tobin’s q — the ratio of a company’s enterprise value to the replacement cost of its physical assets. When q is well above 1, the market is valuing each dollar of installed capital at several dollars, which gives management a powerful incentive to keep adding capacity. Over the past five years, Meta’s average q has been roughly 8x and Alphabet’s roughly 11.7x. A high q can persist for years, even after a bubble has popped, because most of the premium reflects the existing franchise rather than the new capital. What matters is the marginal response: whether the market is still paying for the next dollar of capex. Through 2025, it was — Meta added roughly $140bn of capex and $1.35tn of market cap. In 2026, that has reversed: both companies have spent more than in any prior year, and market caps have fallen (Meta’s q has slipped to \~6.2x, Alphabet’s to \~9x). That divergence — a high average q with a negative marginal response — is typically an early warning sign that the market expects returns on new capital to fall below the cost of that capital.

**Interest Rates**

Between June 1999 and the market peak in May 2000, the Fed raised interest rates 175bp to cool a frothing economy. Today, we have entered another tightening cycle, where the Fed is looking to calm inflation —largely driven by the current administration's tariff policies and what seems to be a strong job market based on ADP’s recent numbers. The relevance is that, like in the Dotcom bubble, the asset base’s growth was supported by a robust credit market. As rates rise, they threaten the industry’s ability to support this continued expansion — one of the reasons we have seen a huge uptick in off-balance-sheet arrangements over the past five years. This protects their core operating businesses from the potential fallout of overbuilding. Global Crossing’s $22bn in fibre was liquidated for $500mm in January 2002.

**More on off-balance-sheet arrangements & credit**

During the Dotcom bubble, carriers sold each other capacity swaps (IRUs) on fibre. Global Crossing sold capacity to Qwest; Qwest sold an equal amount back. Both booked the sale as revenue and capitalized the purchase as an asset. No cash changed hands, both income statements grew, and the fibre glut was disguised as demand. Nortel, Lucent and Cisco lent their customers the money to buy their equipment — tens of billions by 2000. The suppliers’ revenue was real only as long as the buyer performed. WorldCom booked $11 billion of fibre operating costs as capex. Not off-balance-sheet so much as a misuse of accounting standards, but it served the same purpose: keep reported earnings growing while cash went to fuel the asset build-out. CoreWeave, Oracle and others borrow against GPUs and long-term hyperscaler contracts to build capacity that Meta and Alphabet then lease. The hyperscalers get compute without the asset's capital intensity; the credit risk sits with the bagman — which is where vendor financing sat in 2000. Before 2019, under the old lease standard (FAS 13), operating leases didn’t show up in a company’s liabilities — only a footnote disclosure of future minimum payments. Most capacity arrangements were structured to qualify under FAS 13: a 20-year IRU on a fibre pair was typically booked by the buyer either as an operating lease or as a prepaid service contract and then amortized through the income statement over its life. So a dotcom, ISP or enterprise that had committed hundreds of millions of dollars of future bandwidth payments showed no corresponding liability. This kind of aggressive behaviour in maintaining EPS performance leads companies to do dishonest things — management’s compensation is largely driven by share price performance: the more their stock goes up, the more they are paid. This phenomenon directly coincides with the idea of Tobin’s q.

**Back to the roots — what is value?**

Well, value is an interesting concept that can apply to any company at any time. I’m an avid listener of a podcast called Behind the Balance Sheet — I suggest you listen to it. The podcast is focused on the idea of value investing. Some guests have strict definitions of what value is; others are a little looser with it. Over time, I have come to my own definition through the framework mentioned above: value exists in industries that have experienced the Capital Cycle, where they are now fully consolidated and mature, where the return on capital now exceeds the cost of capital, and where the only capex remaining is maintenance — thus balancing the concept of Tobin’s q. An example of this: the remnants of the Dotcom bubble — Telecoms… and what’s happening now in the telecom space? The threat of new entrants and disruption in a mature market that is about to become youthful again, twenty-five years later, with the innovation of SpaceX’s Starlink.

My suggestion to all: prepare for the coming rout the data centre glut is bringing, and search for value where you can protect your beta exposure. The problem with today’s market structure is that measuring a business’s cost of capital is almost impossible without incorporating a skewed beta measure. Therefore, a company’s cost of capital likely reflects the current performance of AI leaders, given the index composition. Because the major indexes are market-cap weighted, the beta calculation will naturally reflect the company’s movements relative to big tech. I’m not sure if that’s still the case, but I recall a period when pharma had a negative beta because it lagged AI —which supposedly prices the risk of those investments as very risk-averse — which is certainly not the case, but I digress — that’s a conversation for a different topic. Seek cash and healthy balance sheets that are capable of weathering the storm.
```



document fetched 2026-11-09 00:38 · canonical https://reddit.com/r/ValueInvesting/comments/1xr69cz/dark_fibre_dark_silicon_unsolicited_market
