Author: Muhammad Yusuf,Research @Delphi_Digital

Compiled by: Deep Tide TechFlow

Deep Tide Overview: The AI agent peptAI designed an ADHD candidate drug in 24 hours, the BIO Protocol token surged 105%, and crypto Twitter was instantly flooded with "DeSci." But looking at the data calmly: drugs discovered by AI do indeed have a Phase I clinical success rate of 80-90%, but Phase I only tests "will it kill you," the real determinant of whether it can cure is Phase III, and currently, none of the 173 AI drug pipelines worldwide have completed it. The most optimistic prediction is that the first FDA approval will not be seen until 2028, and the attention cycle of the crypto market cannot sustain even 6 months—this game's endgame is either DeSci finding a financing mechanism compatible with the 4-year clinical cycle or another round of narrative bubble bursting.

AI Drug Discovery

An AI agent called peptAI designed a new type of ADHD peptide candidate drug from scratch in 24 hours, completed eight validation processes, and outputted a molecule that can be sent directly for wet laboratory testing. The laboratory work only cost a few thousand dollars. The BIO Protocol token surged 105%, and within hours, half of crypto Twitter's bios had added "DeSci," just as they had all added "AI" six months earlier.

Open-source protein folding models can now reach the level of AlphaFold3 at zero licensing cost, and publicly available bioactivity databases cover 2.5 million compounds, with wet lab validation costs under $2000. AI is compressing the cost and time of drug discovery.

I spent a week trying to figure out what was truly different this time.

Passing Phase I Clinical is Not a Big Deal

The commonly cited figure is that the Phase I clinical success rate for AI-discovered drugs reaches 80-90%, while traditional drugs hover around 47%. But what no one adds is that Phase I only tests whether the drug will kill you, not whether it can cure you. Passing Phase I merely means your compound is safe enough to continue research, but you still need to go through the entire remaining process all the way to FDA approval.

Fewer than 40 AI-discovered compounds have reported Phase II data, and none have completed Phase III. Rentosertib from Insilico Medicine is currently the most advanced AI-discovered compound, having published positive Phase IIa results for treating idiopathic pulmonary fibrosis in Nature Medicine in mid-2025 and starting Phase III enrollment in China in Q4 of 2025. If everything goes perfectly (enrollment completed in 2027, data published in 2028, FDA review in 2029), what you see is at least a three-year timeline—this is still the best candidate among 173 drug pipelines, with several compounds having been shelved in 2025 for failing to meet endpoints in atopic dermatitis, schizophrenia, and cancer. Independent analysts estimate the probability of the first AI-designed drug receiving FDA approval by 2027 is 60%, and not a single AI-designed drug has completed this process yet.

Can Crypto Twitter Sustain Real DeSci?

Now keep these timelines in mind and take a look at the K-line chart of BIO Protocol. The token dropped from $0.89 to $0.018, then surged 105% after the peptAI news, with $720 million in trading volume swirling within a $68 million market cap. The entire financing logic of DeSci presumes that token holders will patiently wait 7 to 10 years for clinical projects, while crypto Twitter will shift to the next narrative before Phase I data unblinding.

Pump Science leaked its private key on GitHub, birthing a host of scam tokens, including one actually called Cocaine. The enforceability of IP-NFTs has never been tested in court.

Open Science VS DeSci

If we do not fall into the reflexive self-deception and chronic speculation, DeSci holds a glimmer of hope within open science.

In October 2025, the OpenFold Alliance released OpenFold3 under the Apache 2.0 license. Fully trainable, commercially usable, based on 300,000+ experimentally determined structures (unlike AlphaFold3, which Google has locked under academic-specific restrictions). Boltz-2, co-launched by MIT and Recursion, predicts protein structures and binding affinities at a speed 1000 times faster than physical methods. The Baker Lab released RFdiffusion3 in December. ChEMBL contains 2.5 million bioactive compounds with complete ADMET profiles, available for free to anyone with a laptop. What pharmaceutical companies used to spend millions on building infrastructure is now available on GitHub, licensed under a permissive license, with five pharmaceutical companies currently federating their proprietary drug-protein libraries through the federal OpenFold3 program. No one on crypto Twitter talks about these because there are no tokens to trade, and I genuinely question whether the core contributors of these codebases would be excited about tokens.

Research Funding Crisis

In 2025, over 7,800 NIH and NSF grants have been terminated or suspended, with over $5 billion in funding frozen. The NIH (National Institutes of Health, the largest public funder of biomedical research on Earth, with an annual budget of about $47 billion) budget remains flat because Congress continues to appropriate funds, but the government has still frozen pipelines. New competitive grants dropped from 11,659 in FY 2024 to 6,095 in FY 2025, a 48% decline. The success rate for researchers applying for funding has fallen from 21% to 13%, with Fred Hutch losing $508 million and Harvard losing $945 million.

The funding vacuum is the reason DeSci has a chance at promotion—if done right. In July 2025, VitaDAO-funded Gero signed a research and licensing agreement with a Chinese and foreign pharmaceutical company (a Roche subsidiary, market cap about $100 billion, not a meme) with milestone payments up to $250 million. This is the first time a DAO-funded project has produced something that a real pharmaceutical company is willing to value at nine figures. This process was completed without governance disputes or runaways, and remains the most significant event that has occurred in this field.

Four Years of Eternity

This year, 15 to 20 AI-discovered drugs will enter Phase III, with data for Rentosertib not expected to be published until 2028, meaning a real judgment of whether all this can translate into effective drugs in humans is still years away. Whether or not there are tokens, the open-source tech stack will continue to compress costs, and the funding vacuum will keep pushing researchers towards anyone willing to write checks. Open-source protein folding models can now reach the level of AlphaFold3 at zero licensing costs, with wet lab validation costs under $2000, while NIH has just recorded the lowest grant success rate in twenty years. Even if AI delivers on everything its supporters have promised, cutting the drug development timeline in half, you will still see a 4 to 5-year cycle from discovery to approval, assuming that the Phase III success rate improves in that optimistic scenario. Four years is an eternity, and in an industry where portfolio beliefs shift with quarterly earnings call rotations, token holders treat a 6-month holding period as a life sentence.

Whether or not anyone buys tokens, the costs of discovery and innovation are getting cheaper each quarter. The funding vacuum created by the NIH is also a problem. Between these two reasons, there may be a viable version in which tokens fund specific clinical trials, set concrete milestones, and governance rights are handed over to knowledgeable individuals. The Gero/Chinese and foreign pharmaceutical deal is the first proof that DAO-funded projects can produce something that real pharmaceutical companies are willing to pay nine figures for. Putting hype aside, I am curious whether anyone will build a defensible funding infrastructure for real DeSci.