AI tokens in this cycle perform actual onchain utility but remain undervalued relative to last cycle's speculative chatbot tokens.
AI & Agents ·
Current artificial intelligence tokens are executing genuine onchain functions this cycle, a substantial shift from the previous generation that largely consisted of chatbots attached to tickers. Research from Sparsity Data identifies Pearl, POD, Surplus, ORBIO, and IMD as examples of tokens where AI performs actual work onchain, contrasting sharply with prior cycles when agents drew market attention through posts rather than tangible utility. The category collapsed dramatically: major tokens from the previous cycle such as Virtual, ai16z, Goat, and AIXBT lost between roughly 90 and over 99 percent of their value from peak to recent levels, conditioning the market to discount AI tokens broadly.
The technical foundation has shifted materially. Models have become simultaneously more capable and far cheaper to operate. Open-weight models released recently score five times higher on capability benchmarks while costing a twentieth of previous-generation models, and autonomy horizons have extended from minutes to hours. This capability and cost improvement removes a primary reason for the previous cycle's discount: tokens are no longer priced on speculative agent potential but on demonstrated onchain execution.
What remains uncertain is whether market sentiment will revalue the category as the distinction between this cycle's functional tokens and last cycle's promotional ones becomes clearer. The discount applied uniformly across all AI tokens regardless of actual utility remains in place, though the structural differences now supporting current projects appear substantively greater.