Delphi Digital: Is DeSci experiencing a second wave of opportunities?

Null

Author: Muhammad Yusuf, Delphi Digital Researcher

Translation: Yuliya, PANews

Editor’s Note: Although BIO experienced nearly a month of rapid growth, do not be fooled by short-term wealth creation myths and token celebrations. As AI and open-source models lower the barriers to drug discovery to near zero, true medical breakthroughs still cannot bypass the long and rigorous clinical cycles. Faced with the crypto community’s lack of patience for speculative hype and the traditional research sector’s “funding winter,” DeSci indeed has a good opportunity to break through. But what it needs is not a speculative game that switches tracks in a few months, but rather the establishment of a funding infrastructure tightly bound to clinical milestones and governed by professionals to ensure sustainable development.

A AI Agent called peptAI designed a new type of ADHD (Attention Deficit Hyperactivity Disorder) peptide candidate drug from scratch in just 24 hours, passing through eight validation processes, and output a molecule ready for direct wet lab testing. The lab work cost only a few thousand dollars. As the supporting platform for this project, BIO Protocol’s token surged by 105%. Within just a few hours, half of Crypto Twitter’s bios included the word “DeSci,” just like six months ago when they all added “AI.”

Today, open-source protein folding models can match AlphaFold3 at zero licensing cost, public bioactivity databases cover 2.5 million compounds, and wet lab validation costs have dropped below $2,000. AI is significantly compressing the costs and time of drug development. Over the past week, I’ve been trying to understand what truly sets this apart this time.

Passing a Phase I clinical trial doesn’t prove much

Current data suggests that drugs discovered by AI have an 80-90% success rate in Phase I trials, compared to about 47% for traditional benchmarks. But no one clarifies that Phase I tests whether the drug will cause death, not whether it can cure a disease. Passing this stage only means your compound is safe enough to continue research, but it still needs to go through subsequent layers of screening until finally approved by the FDA.

Fewer than 40 AI-discovered compounds have reported Phase II data, and none have completed Phase III. Insilico Medicine’s Rentosertib is currently the most advanced AI-discovered compound, with positive Phase IIa results for idiopathic pulmonary fibrosis announced in mid-2025 (published in Nature Medicine), and Phase III patient recruitment began in China in Q4 2025. If all goes well (completion of recruitment in 2027, data readout in 2028, FDA review in 2029), the best candidate among the 173 drugs in the pipeline will still require at least three years. Several compounds in this pipeline were shelved in 2025 for failing to meet endpoints in atopic dermatitis, schizophrenia, and cancer. Independent analysts estimate a 60% chance that the first AI-designed drug will be FDA-approved by 2027, but so far, no AI-designed drug has completed this process.

Is crypto Twitter really suitable for true DeSci?

Now, keep these timelines in mind and look at the chart of BIO Protocol. Its token fell from $0.89 to $0.018, then rebounded 105% on news of peptAI, with a trading volume of $720 million and a market cap of about $68 million. The entire DeSci funding logic is based on a hypothesis: token holders will patiently wait for clinical projects that take seven to ten years, but in reality, even before Phase I data is unblinded, Crypto Twitter has already shifted to the next narrative.

Pump Science even leaked its private keys on GitHub, spawning a bunch of fraudulent tokens, one of which was called Cocaine, and the enforcement of IP-NFTs has never been tested in court.

Open Source vs. DeSci

If we avoid reflexive self-deception and long-term speculative hype, open science might bring a glimmer of hope for DeSci.

In October 2025, the OpenFold alliance released OpenFold3 under the Apache 2.0 license. It is fully trainable and commercially usable, built on over 300,000 experimentally determined structures (unlike AlphaFold3, which is restricted by Google for academic use only). Boltz-2 from MIT and Recursion jointly predicted protein structures and binding affinities at a speed 1,000 times faster than physics-based methods.

Baker Lab released RFdiffusion3 in December. ChEMBL has 2.5 million bioactive compounds with complete ADMET profiles, accessible for free to anyone with a laptop. The infrastructure that pharmaceutical companies spent millions of dollars to build internally is now available on GitHub under permissive licenses, and five pharmaceutical companies are currently federated training on their proprietary drug-protein libraries through the Federated OpenFold3 Initiative.

No one talks about these on Crypto Twitter because there are no tokens to trade, and I highly doubt the core contributors of these codebases are excited about issuing tokens.

Funding shortages in science

In 2025, over 7,800 NIH and NSF grants were terminated or suspended, with more than $5 billion frozen. NIH (the U.S. National Institutes of Health, the largest public biomedical research funder globally, with an annual budget of about $47 billion) maintained its budget, as Congress continued appropriations, but the government still froze the funding pipeline. New competitive grants dropped sharply from 11,659 in FY2024 to 6,095 in FY2025, a 48% decrease. The success rate for researchers applying for funding fell from 21% to 13%, with Fred Hutch losing $508 million and Harvard losing $945 million.

This funding gap is precisely why DeSci’s pitch can gain opportunities if operated properly. In July 2025, Gero, funded by VitaDAO, signed a research and licensing agreement with Chugai Pharmaceutical (a Roche subsidiary with a market cap of about $100 billion), with milestone payments up to $250 million. This was the first DAO-funded project to produce something truly valued in the nine-figure range by a real pharmaceutical company. The process was completed smoothly without governance disputes or flight, making it one of the most significant events in the field.

A long four years

This year, 15 to 20 AI-discovered drugs entered Phase III trials, but data readout for Rentosertib won’t happen until 2028 at the earliest, meaning it will be years before we have a definitive answer on whether any of this can translate into effective human medicines. Regardless of tokens, open-source tech stacks will continue to cut costs, and the funding vacuum will keep pushing researchers toward anyone willing to write a check. Today, open-source protein folding models match AlphaFold3 at zero licensing cost, wet lab validation costs are below $2,000, and NIH just announced its lowest funding success rate in twenty years.

Even if AI fulfills all its supporters’ promises and halves the drug development timeline, it still takes four to five years from discovery to approval—assuming Phase III success rates actually improve. In an industry where investment confidence shifts with quarterly earnings calls, four years is a long wait, and token holders even find six months of holding unbearable, like a life sentence.

The costs of drug discovery and innovation are falling every quarter, regardless of token popularity. NIH’s funding shortages are also concerning. Perhaps between these two, a viable model exists where tokens fund specific trials and milestones, while governance is handled by professionals. Gero’s collaboration with Chugai is the first proof that DAO-funded projects can produce something pharmaceutical companies are willing to pay nine figures for. Besides eliminating hype, I also wonder if anyone is building resilient financing infrastructure for truly decentralized science.