Right, and still poor.

Three questions, and only one of them is fun

Almost every investment argument you will read collapses three separate questions into one. They are worth pulling apart, because they have wildly different difficulty levels and wildly different failure modes.

1. Is the trend real? Will this technology, this shift, this demand actually arrive?
2. Does this company capture it? Of everyone standing in the path of that trend, which one converts it into durable profit rather than watching it pass through?
3. Is the price right? Given a correct answer to both of the above, does the current quote leave you any return?

The first question is the one everybody argues about, and it is by far the easiest. Trends are visible. You can read about them. Enough smart people are shouting about any given trend that the consensus is usually directionally correct, and being contrarian about the trend itself is usually just being wrong early.

The second is genuinely hard and gets a fraction of the attention.

The third barely gets discussed at all, which is remarkable, because it is the only one of the three that determines your return. You can be completely right about questions one and two and still lose money for a quarter of a century. We have a clean example of exactly that, and it finished running its course only recently.

The control experiment

On March 27, 2000, Cisco Systems closed at $80.06 and became the most valuable publicly traded company in the world, passing Microsoft. Its routers and switches were the physical backbone of the internet. If you had to name one company that was correctly identified as essential to the coming era, it was this one.

Everyone who bought Cisco that day was right. Right about the internet, which transformed the world roughly as advertised. Right about Cisco, which did not fail, did not commit fraud, and did not get disrupted into irrelevance. Revenue has nearly quintupled since 1999. Profits roughly quadrupled. Earnings per share grew about eightfold. Margins stayed healthy the entire time. The thesis worked.

Cisco closed above that March 2000 price for the first time on December 10, 2025, at $80.25.

Twenty-five years and eight months. That is not a market being irrational for a while. That is a full professional career, spent being correct.

The counterexample matters too, and I want to state it before someone else does. Other companies from that same era, correctly identified as internet winners, fell just as catastrophically and then went on to produce extraordinary returns for anyone who held or bought the wreckage. Amazon is the obvious one. Same era, same correctness about the trend, opposite outcome for a long-term holder.

So the lesson is not "the internet was a bubble." It plainly was not. The lesson is that two companies can be equally right and produce opposite investment outcomes, and the variable that separates them is not insight. It is the relationship between the price paid and the growth that followed. One grew into an absurd multiple fast enough. The other did not, despite doing nearly everything right.

Why question three is so hard to hold onto

Knowing this does not make it easy, because the failure mode is emotional rather than analytical.

When you have done the work on questions one and two, and you are right, the conviction you have earned feels like it should also settle question three. It does not. Being right about the technology generates a feeling of certainty that then gets spent on the price, which is the one place certainty is worth nothing.

It is also socially expensive to hold. Declining to buy a company you have just spent two thousand words praising reads as fence-sitting. Saying "this is excellent and I am not buying it" satisfies nobody: the bulls think you are timid, the bears think you are captured. A rating that says the business is superb and the price is not is the least popular thing you can publish, which is roughly a guarantee that it is underprovided.

And the arithmetic is genuinely uncomfortable. A high multiple is not a prediction that a company will do well. It is a prediction that it will do well and that the market will still think so later. You are underwriting two things, and the second one is not in the company's control.

Where this leaves the AI question

I have written that the AI build-out is real and that the physical layer collects the rent. I believe that. It is an answer to question one, and partly to question two. It is not an answer to question three, and I want to be careful not to let the first two smuggle in the third.

One comparison is worth sitting with, precisely because it is uncomfortable for anyone bullish today. At its March 2000 peak, Cisco traded at roughly 31 times sales. That number was later used as the canonical example of dot-com excess.

The most recent report on this site, on Palantir, put its enterprise value at about 54 times this year's guided revenue. That is roughly 1.7 times the multiple Cisco carried at the top of the most famous bubble in living memory.

I want to be fair about the limits of that comparison, because a dishonest analogy is worse than none. Cisco at its peak was not growing 93% a year and did not run a 62% adjusted operating margin. Palantir's business quality on these numbers is genuinely better than Cisco's was, its balance sheet is stronger, and it is profitable on a GAAP basis. A higher multiple on a better business is not automatically a mistake.

But the mechanism is the same, and it is the mechanism I care about. At 54 times revenue, you are not forecasting that Palantir does well. You are forecasting that it does well and that the market still pays a very large multiple after it has. Cisco's investors got the first half and lost a quarter of a century on the second.

What this does to the coverage board

It is worth saying out loud that this belief has a visible cost in the published record. Three of the last four ratings on this site are HOLDs, on businesses I described in each report as excellent. Costco, Texas Pacific Land, and Palantir are all, in my view, better companies than most of what trades. I declined all three on price.

A reader is entitled to look at that and wonder whether I am simply too cautious, or whether a HOLD is a way of avoiding commitment. It is a fair challenge and I would rather raise it than have it raised for me. My answer is that a rating that never says no is not a rating, and that the scoring protocol is designed to catch exactly this: if the HOLDs are systematically wrong, the record will say so on the schedule, and I will not be able to quietly restate them.

Which is the point of publishing the dates in advance.

What would prove me wrong

A note with no failure condition is enthusiasm with a longer time horizon, so here are the conditions for this one.

• The HOLDs are wrong on the record. If the calls I declined on price systematically beat the benchmark by more than five percentage points over their scored twelve months, then the discipline is costing more than it saves and I should recalibrate. First results land in July 2027.
• Multiple compression stops being a thing. If today's highest-multiple businesses grow into their valuations without a de-rating, over a full cycle including a recession, then the Cisco mechanism is a period artifact rather than a rule.
• Quality proves to dominate price over long horizons. The strongest counterargument is that for genuinely exceptional businesses held for twenty years, entry multiple barely matters. If the evidence supports that at the multiples on offer today, my framework is too conservative for the best companies and I would be paying a real cost for the caution.

That last one I hold most loosely. It has the best argument behind it, and I am not certain I am right.

Sources

Cisco price history and the December 10, 2025 milestone are as reported at the time; the peak close of $80.06 on March 27, 2000 and the $80.25 close on December 10, 2025 are split-adjusted. Valuation multiples at the 2000 peak, and the revenue, profit, and earnings-per-share growth figures cited across the period, are as reported in coverage of that milestone. The Palantir multiple is from our own initiation report, computed from company guidance and the August 14, 2026 close.

• CNBC on Cisco reclaiming its dot-com peak
• Yahoo Finance on the length of the trough
• Summary of the peak multiples and the fundamental growth over the period