This report looks at four extensions where that shift shows up very differently: .ai, .co, .gg and .us. It also tests how well NetMark’s pricing model handles them.

Country codes are taking a bigger share of the market

The count share more than doubled in five years, and the jump came after 2023. The dollar share grew faster still, from 5.5% to 16%. It would be 23.4% if not for a single sale: ai.com went for $70 million in February 2026. Even the biggest .com deal of the period was driven by artificial intelligence.

.ai: from a country code to a category

.ai is Anguilla’s country code, but buyers now treat it as the extension for artificial intelligence. The market changed after ChatGPT launched in late 2022.

In 2025 and 2026 alone, the database records 352 reported .ai sales worth $23.2 million. The top sales are one-word names bought by AI startups: bot.ai ($1.2M), wisdom.ai ($750K), cloud.ai ($600K), genesis.ai and lotus.ai ($400K each). Before 2023, a big .ai sale was $50,000.

The lesson is that a ccTLD can be repriced by a single technology wave. The price follows what the letters mean to buyers, not the country behind them.

.co: the steady alternative to .com

.co (Colombia) has grown more slowly, with no AI-style jump. The median reported .co sale was $10,000 in 2026, up from $7,200 in 2023. That is roughly three times the .com median of $3,500.

The top of the .co market keeps climbing. symphony.co sold for $180,000 in July 2026, nova.co for $160,000 in November 2025, and cv.co for $130,000 in August 2026. Short dictionary words carry the most value. Real three-letter English words on .co sold for a median of about $15,000 since 2023, with car.co at $90,000 at the top.

.co also shows how fragile boom-era prices can be. Its 2021 median of $11,728 fell to $5,000 a year later. Buyers who price from 2021 sales will overpay.

.gg and .us: two very different stories

.gg (Guernsey) has been adopted by gamers, because “gg” means “good game.” Reported sales are rare but rising in value. The record in the database is x.gg at $115,000 in July 2026, followed by hype.gg ($40,000) and nano.gg ($30,000). With only six reported .gg sales in 2026, a median of $13,500 is a signal, not a benchmark.

.us is the opposite case. It is a true national extension with no second meaning, and its prices have stayed flat. The median reported .us sale has held between $4,600 and $7,600 since 2023, and the best recent sale was power.us at $17,000. Reported .us sales fell from 13 in 2019 to one so far in 2026.

The contrast makes the main point. Country codes rise in price when buyers give the two letters a meaning. A country code that just means its country behaves like a regional extension.

How well the pricing model reads country codes

NetMark Intelligence is centered correctly on .ai and .co, but it misses rare high-end .gg sales and underprices .us. We tested its retail estimate against recent reported sales (2024–2026), pricing each name as of the day before it sold.

The .ai misses run in both directions, which fits a market still finding its level. The .gg misses are the standout names: x.gg sold for $115,000 against an estimate of $43,370. On .us, the model sits about a third below actual sales, so it is conservative where the market is quiet.

What this means for buyers and sellers

• Price the meaning, not the extension. .ai and .gg gained value because the letters mean something to a buyer group. .us has no such meaning, and its prices show it.
• Use recent sales only. Sales from the 2021 boom and from before ChatGPT mislead in both directions. Comps older than three years should be treated as background.
• Compare like with like. A three-letter .co word and a six-letter .co word are different markets. Length and real-word status matter more on country codes than on .com.
• Expect wide ranges. Country-code markets have fewer sales, so one strong buyer can set a record. Treat single high sales as ceilings, not as the going rate.
• Watch the next wave. .ai shows that a ccTLD can be repriced within two years. The next extension to gain a meaning could move just as fast.

About the data

Yearly medians and market shares use the 28,853 sales reported by DNJournal in the NetMark comps database, January 2019 to October 2026, so every year is measured the same way. Named sales and the three-letter .co comparison also draw on the database’s other sales feeds. Country codes are two-letter extensions plus second-level forms such as .co.uk. Medians are used instead of averages, so one record sale cannot skew a year. 2026 covers January to October. Small samples (.gg, .us) are stated where they appear. Model results come from NetMark Intelligence v6.1, a shadow model that is not used for live pricing.

To see how wide the gap really is, we took 200 domains at random from the NetMark comparable-sales database, appraised each one with a simple formula, and then checked the results against two data-driven methods built on hundreds of thousands of recorded sales. The results say a lot about how domain prices are actually set.

How we tested

We drew 200 sales at random from NetMark’s comparable-sales database: every name sold for at least $100 between January 2023 and October 2026, across all extensions and sales venues. Each name was then valued three ways.

1. A formula appraisal (the Math model). It scores the name on visible traits such as extension, length, hyphens, numbers and dictionary words, then maps the score to a price range. This is how most instant online appraisal tools work.
2. Comparable sales. NetMark’s comps engine looks for recent sales of similar names in the same extension and category. The name’s own sale was hidden from it, so it had to price the name blind.
3. The NetMark Intelligence model. A statistical model trained on about 479,000 recorded sales. It was asked for a price as of the day before each sale, using only data available then, and for the right market: end-user retail or investor auction.

We compared each estimate with the real sale price. Because domain prices span from $100 to millions, we measured the error as a ratio, not in dollars: an estimate of $2,000 for a $1,000 sale is “2x high”.

What the numbers showed

The formula was the least reliable way to price a name: it came within 2x of the real price only 32% of the time. The median sale in the sample was $3,217.

Three findings stand out.

• The formula prices names in a few fixed boxes. It gave 103 of the 200 names the same $1,000–$3,000 range. Those 103 names actually sold for anything from $105 to $600,000. The real price fell inside the formula’s range only 27% of the time.
• Extensions outside .com break the formula. For the 75 non-.com sales it was never within 2x. It valued stocks.si at $24; it sold for $25,000. The two data-driven methods handled these names about as well as .com.
• Comparable sales are only as good as the comparables. The comps engine found no trustworthy match for 88 of the 125 .com names and declined to guess. When it did match, half its estimates were within 2x.

The chart shows the most important pattern. Every method overvalued the cheapest sales and undervalued the most expensive ones. The data models were close in the middle of the market, where most sales happen: NetMark Intelligence came within 2x on 97% of the $2,500–$5,000 sales.

Why value and price drift apart

A domain does not have one value. It has a likely price in a given market, on a given day, to a given buyer. Four forces explain most of the gap we measured.

• The market decides the price. The same kind of name sells for very different amounts depending on where it sells. In our sample, expired-domain auctions closed at a median of $305, while end-user sales on marketplaces like Sedo and Afternic had a median of $3,550. The formula gave the expired names a typical estimate of $1,732, five times what they fetched.
• The top of the market is driven by single buyers. socrates.com sold for $600,000 and liquid.org for $300,000. No model priced either above $80,000. A six-figure sale usually means one buyer needed that exact name, and no pattern in past sales can predict that buyer.
• Surface traits miss meaning. A formula sees that win.ai and gats.io are short names on popular extensions. It cannot see that one is a common word with obvious buyers. Models built on real sales pick up demand for words, categories and extensions that a points system never captures.
• Every estimate leans toward the average. Models learn from the bulk of the market, which sits between $1,000 and $5,000. Pulled toward that middle, they overvalue the rare cheap sale and undervalue the rare expensive one. That is the pattern in the chart.

What it means for buyers and sellers

Treat any single appraisal as a starting point, not a price tag. The practical lessons from these 200 sales:

1. Ignore formula appraisals for pricing decisions. They are fine for sorting a list, but in this test they were within 2x of the real price less than a third of the time.
2. Price by market, not by name alone. Decide first whether you are selling to an end user or to investors. The same name can be worth ten times more on a marketplace listing than at an expired auction.
3. Check more than one data source. Recent comparable sales and a sales-trained model together cover more names than either alone. When they disagree badly, the name deserves a closer look.
4. Expect the extremes to surprise you. Below $1,000, models are often too optimistic. Above $20,000, they are almost always too low. Great names are worth testing with a higher asking price, because the buyer who needs them is not in anyone’s model.

The bottom line: a domain’s value is an estimate of where the market is likely to settle. Its sale price is where one buyer and one seller actually met. Good data narrows the gap between the two, but it never closes it. That gap is where careful pricing pays off.