Everything About AI
NextFin News - A researcher can now scan the genomes of living and extinct organisms and surface hundreds of thousands of antibiotic candidates in hours instead of years. The instruments doing the work are not microscopes or pipettes but Codex and ChatGPT, deployed inside the transdisciplinary lab of bioengineer César de la Fuente at the University of Pennsylvania.
The math behind the shift is unforgiving. Bacterial antimicrobial resistance was associated with roughly five million deaths in 2021, an annual toll projected to roughly double by 2050. Yet no new class of antibiotics has reached patients in 50 years. The traditional search — sampling plants, animals, microbes, insects, water, and soil, then isolating and testing candidates in an iterative loop — can take years per molecule. De la Fuente's lab treats biology as an information system, training deep-learning models to read the alphabet of nucleotides and amino acids, then scanning genome and protein datasets for patterns that mark a functional antimicrobial. The initial search, once a multi-year exercise, now runs in hours.
The open question is not whether AI can find candidates. It is whether the economics of antibiotic development can keep pace with the speed of discovery.
The Mechanism: Why Code Compresses Chemistry
The core insight is that biology is legible as data.
"The nucleotides that make up DNA, and the amino acids that make up proteins and peptides are sort of like an alphabet," said César de la Fuente, a bioengineer whose lab at the University of Pennsylvania searches for antimicrobial candidates. "Thinking about biology as information enabled us to develop methods that can begin to decipher the organizing principles of life that gave rise to a functional molecule."
That reframing converts a chemistry problem into a search problem. Digital genome and protein databases let scientists query the entire tree of life, but abundance creates its own bottleneck: separating signal from an enormous possibility space. AI is built for that needle-in-a-haystack task — scanning huge datasets, surfacing patterns invisible to human reviewers, and ranking a manageable set of candidates for experimental testing. The productivity delta is the story: where a wet-lab team might characterize a handful of candidates per month, a model can score orders of magnitude more, then hand a prioritized shortlist to the bench.
Codex and ChatGPT sit alongside the lab's own deep-learning models as workflow infrastructure. Lab members use them to brainstorm hypotheses, write and refine code, process datasets, analyze results, and connect ideas across scientific disciplines. The effect is organizational as much as technical.
"We're a highly transdisciplinary group, so we have people from biology, from chemistry, from computer science, from engineering," de la Fuente said. Codex lowers the language barrier between disciplines: a biologist can write a program, and a programmer can tackle a biological problem. "With Codex, someone who has never programmed before, who doesn't understand the language of computer science, can now do it."
The lab also uses ChatGPT as a collaborative sounding board. "Our ChatGPT workspace is receiving input from all these different people that think differently about the problems that we're trying to tackle," de la Fuente said. Members feed it good and bad ideas, review unfamiliar topics, clarify terminology, compare methods across fields, and organize drug-discovery ideas. ChatGPT also lets members work in their native languages, reducing friction in cross-border teams. De la Fuente uses it to shape hypotheses before taking them to colleagues, valuing the ability to access the world's information with a click.
But the workflow carries an explicit guardrail.
"Obviously you have to always double-check for accuracy," de la Fuente said.
Identifying a promising candidate is only the first gate. Scientists must confirm the molecule kills the target microbe, determine the effective dose, and test effects on human cells. Chemists optimize for effectiveness, safety, and stability. Further tests measure toxicity thresholds, how readily microbes develop resistance, how the candidate moves through the body, and whether it can be manufactured reliably. Candidates that clear those hurdles still face regulatory review and clinical trials.
"Ground-truth experiments are essential to validate AI predictions," de la Fuente said. "This will be critical in the life sciences in the years to come if we are to continue scratching the surface of our understanding of biology, which is the most complex thing out there."
For de la Fuente, the work extends a longer scientific lineage.
"We have always relied on tools and machines to understand the world around us," he said. "The telescope illuminated the cosmos, and the microscope revealed the world of the invisible."
Machines now help researchers understand, predict, and engineer biology. "That is what our work is all about."
The Market Signal: Capital Is Chasing the Stack, Not the Molecules
The investment data tells a clear story about where capital believes the value is accruing. The global drug discovery technology market was valued at $32.69 billion in 2025 and is projected to exceed $80.27 billion by 2035, a compound annual growth rate of 9.4 percent. Pharma and biotechnology companies accounted for 55 percent of end-user share in 2025, while academic and research institutes are the fastest-growing segment through 2035. North America held 38 percent of the market in 2025.
Venture activity has followed. Pharma biotools deal value reached $5.7 billion in 2025 across a record 537 transactions, rebounding after three years of contraction. Investors are backing asset-light business models and AI-enabled technologies that support discovery and development — the "scientific stack" that connects computational discovery with physical experimentation. In the broader AI market, Q1 2026 venture funding reached $255.5 billion, eclipsing the full-year 2025 total of $254.4 billion in a single quarter.
Notice what is being funded: platforms, tools, and enabling infrastructure — not pure-play antibiotic programs. The capital is flowing to the pick-and-shovel layer that de la Fuente's workflow exemplifies. Codex lowering the barrier between biology and code is precisely the kind of productivity multiplier that platform investors underwrite. A tool that lets a biologist write algorithms and visualize results without a computer-science degree expands the addressable researcher base, which is a revenue story in a way that a single antibiotic candidate is not.
This is the first-order read, and it is already priced into the platform trade. The asymmetry is visible in the funding mix: pharma biotools deal value reached $5.7 billion in 2025, but that figure sits inside a broader AI venture market where Q1 2026 alone deployed $255.5 billion. Discovery infrastructure is capturing investor attention; antibiotic assets are not. The gap between the two is the clearest market signal in the space. The second-order question is sharper: if discovery becomes cheap and fast, where does scarcity move? The answer points away from molecule identification and toward validation capacity, clinical execution, and — most importantly — the commercial model for antibiotics themselves. When the cost of generating candidates collapses, the marginal value of another candidate falls toward zero; the scarce assets become the wet-lab throughput to test them, the clinical pathways to prove them, and the reimbursement architecture to pay for them.
The valuation implication is uncomfortable for investors who entered AI drug discovery expecting a straight line from better models to approved drugs. AI-designed molecules still have to survive the same attrition that has always governed the pipeline. Industry observers note that even for antibody therapies — a particularly promising area — the number of possible protein structures remains beyond human comprehension, and the model algorithm in drug discovery is not yet where it needs to be. The constraint, in one formulation, is not computation but biology itself: most failures occur because the underlying disease mechanisms are not fully understood. That is why the platform trade and the asset trade are different securities, and why the market's enthusiasm for one does not automatically transfer to the other.
The Counter-Thesis: The Bottleneck Has Moved From Discovery to Economics
The strongest case against the bullish platform narrative is that AI solves the wrong bottleneck. The antibiotic development pipeline faces a dual crisis: a scarcity of drugs in development and a lack of innovation in methods to fight drug-resistant bacteria. Even when new antibiotics materialize, the commercial model is broken. Antibiotics are used sparingly by design — stewardship programs reserve new drugs for resistant cases — which caps revenue precisely when recoupment is needed. The result is a market where the social value of a new class is enormous and the private return is not.
The evidence is in the approval record. Of the large pharmaceutical companies still investing in the space, GSK reported 30 ongoing projects and, in 2025, won approval for the first new class of drugs to combat urinary tract infections in nearly 30 years. One new class in three decades, from one of the few remaining big-pharma participants, is the clearest signal that the constraint is not the ability to find molecules.
This is where the cyclical-versus-structural distinction matters. The acceleration of discovery is structural: once biology is legible as data and models can search sequence space, the reduction from years to hours does not revert. No future regime returns to manual, sample-by-sample prospecting as the primary discovery method. But the funding gap for antibiotic development is also structural — a function of stewardship economics and the misalignment between public-health value and private returns. Neither force is cyclical, which means the tension between them will persist rather than resolve on its own.
The counter-thesis, then, is not that AI fails. It is that AI succeeds at discovery while the system fails to monetize it. Platform companies capture value upstream; pure-play antibiotic developers bear the downstream risk. Investors who conflate the two — assuming that faster discovery automatically translates into viable antibiotic businesses — are making a category error.
The falsifying signal for the platform-bull case is specific and observable: if, over the next three to five years, the number of new antibiotic approvals does not rise materially despite the collapse in discovery time and cost, the value capture thesis shifts decisively toward the validation and commercialization layer — and away from discovery platforms as the primary beneficiaries. Watch the approval count, not the candidate count.
What Comes Next: Three Time Horizons
In the short term, the trade is in the tools. Academic and research institutes are the fastest-growing end-user segment in drug discovery technology, and the democratization of computational biology through Codex expands that base further. Expect continued venture interest in AI-enabled discovery infrastructure, protein-design platforms, and automation that links computation to wet-lab throughput. The pharma biotools rebound — from three years of contraction to a record 537 deals — suggests investors are willing to pay for infrastructure even while they remain selective on assets.
In the medium term, the bottleneck migrates to validation. AI can generate hundreds of thousands of candidates; the constraint becomes how fast labs can synthesize, test, and triage them. Ground-truth experimentation — dose-response, toxicity, resistance development, pharmacokinetics, manufacturability — becomes the scarce resource. Companies that own high-throughput validation capacity, or that can compress the wet-lab loop with automation, capture the next layer of value. This is the second-order trade that the platform rally has not fully priced: the beneficiaries of cheap discovery are not the discoverers, but the validators.
In the long term, the unresolved variable is policy. A structural discovery engine requires a structural commercial model. Options on the table include push funding for R&D, pull incentives that reward approved antibiotics regardless of volume sold, subscription-style payment models, and extended exclusivity. Without one of these mechanisms, the pipeline scarcity documented by global health monitors will persist regardless of how fast AI finds candidates.
Base case: discovery platforms compound as the scientific stack becomes standard lab infrastructure, while antibiotic approvals climb only gradually because the commercial model remains unreformed. Upside case: policy intervention unlocks pull incentives, and the discovery surplus converts into a genuine approvals pipeline — the scenario in which platform and asset holders both win. Downside case: validation capacity and funding fail to scale, and the candidate surplus becomes a catalog of untested molecules, leaving platform valuations supported by workflow adoption rather than approved-drug economics.
The telescope showed humanity the cosmos; the microscope revealed the invisible. AI is now doing the same for the sequence space of life. But the lesson of the antibiotic market is that finding a molecule is the easy part. The hard part is building a system that pays for it to reach a patient.
Explore more exclusive insights at nextfin.ai.
