ARPA-H awards Delphi teams up to $117.4M for modular AI biosensor chiplets

ARPA-H said Thursday it awarded Delphi program teams to replace bulky single-purpose biosensors with tiny modular chiplets that report broader biological data for chronic conditions, with an agency commitment of up to $117.4 million over 4.5 years.

Continuous glucose monitors proved people will wear a sensor if the readout is useful. Delphi is the federal bet that the same habit can cover inflammation, hormones, and drug levels with swappable chips, including on-device AI, instead of a new single-purpose gadget for every marker.

On Thursday, 1 October 2026, the Advanced Research Projects Agency for Health said it had named the teams receiving awards for Delphi. ARPA-H is the research agency inside the U.S. Department of Health and Human Services. The record is the ARPA-H news page, “ARPA-H kicks off work to advance next-generation personalized biosensors.” The visible date is October 1, 2026. The page does not print an hour. The line under the headline says the technologies are meant to give people deep, real-time biological readings so they can make decisions about treatment and wellness. The opening sentence says the agency announced the teams that day. Those lines are ARPA-H’s.

What Delphi is supposed to replace. The news page says today’s biosensors are bulky, expensive, and built for one job, such as heart rate or blood glucose only. A biosensor is a device that reads a signal from the body. Delphi would use tiny modular microelectronic pieces, which the page calls chiplets, assembled like building blocks. A chiplet is a small chip that does one job and can be joined to other small chips. The devices those blocks make would report a wider set of biological data for chronic conditions: hormones, inflammatory markers, and medication levels. A hormone is a chemical messenger in the blood. An inflammatory marker is a measurable sign of inflammation. A medication level is how much of a drug is in the body. The page says the aim is more accurate tracking of wellness and health, produced at low cost, for almost any medical use. Those lines are the news page’s. They describe a goal. They do not describe a device a person can buy.

How the pieces are supposed to fit. The news page says teams will apply a chiplet design to biosensing. Separate, specialized, independent pieces would handle power management, signal processing, communications, post-quantum security, and biorecognition, and those pieces would connect to one another. Post-quantum security, on this page, means communication locks meant to hold even if a future quantum computer can break the codes in wide use today. Biorecognition is the part that recognizes a biological target, such as a hormone or an inflammatory marker. The page says every sensor will be plug-and-play, ready to swap onto other wearable platforms. The devices are meant to monitor the body’s inner signals continuously, and to do it safely and securely. Those lines are ARPA-H’s. The page does not say a finished chiplet is already in use.

The money, and the clock. The news page says the agency’s commitment for Delphi is up to $117.4 million over 4.5 years. Up to, in that sentence, is a ceiling, not a check already written for the full amount. Four and a half years is 54 months. That multiplication is arithmetic. The same news page also sets a milestone at 54 months. It does not say in one sentence that the funding window and that milestone are the same clock. Performer awards vary by team. They are contingent on each team meeting research milestones the page calls aggressive and accelerated. Contingent means the money depends on hitting those marks. The award pages say the same thing in a tooltip: funding amounts are ceilings and contingent on meeting aggressive research milestones. Those lines are ARPA-H’s.

The milestone schedule, as the news page prints it. In the first 18 months, teams must show full interoperability and follow post-quantum secure communication standards. Interoperability means a piece from one team can work with a piece from another, instead of each sensor being a closed gadget. At 24 months, teams will show the first prototype device in vivo and name more sensors, to widen the range of biomarkers they can detect. In vivo means in a living body, not only in a dish on a bench. A biomarker is the measurable signal, such as a hormone or an inflammation reading. By 54 months, teams must have finished either a clinical trial, for wearables that go minimally into the body, or a human-factor study, for wearables that stay on the surface. A clinical trial tests a device in people under a study plan. A human-factor study asks whether a person can actually use the device. The page says the path runs from early prototypes through integration, regulatory preparation, clinical trials, and human-factors testing. Regulatory preparation is work toward a regulator. It is not an authorization already in hand. If the work succeeds, the page says, people and their care teams could track a broad range of biomarkers, catch the earliest signs of disease, adjust treatment at home, and reduce hospitalization. If successful is the page’s condition.

Four performer teams, in the order the news page prints them. Massachusetts Institute of Technology will develop a clinical-grade wearable patch that can be extended, for continuous sensing of heart-failure biomarkers. Clinical-grade, here, is the page’s description of the target, not a clearance. The sensor would be powered by heat from the patient and would use advanced, power-efficient on-chip AI. On-chip AI means the analysis runs on the patch, not only on a phone or a server later. New York University will develop a wearable that continuously monitors inflammation, using microscopic probes just under the skin, novel DNA-based sensors, and energy harvested from radio waves, including Bluetooth and Wi-Fi. A DNA-based sensor uses strands of genetic material as the recognition piece. University of Washington will track the perimenopause transition with a non-invasive sensor loaded with engineered proteins that measure biomarkers in sweat. Perimenopause is the transition toward the end of menstrual cycles. By adjusting the sensor’s local pH, those proteins can be refreshed and reused, so the sensor stays sensitive longer. pH is how acidic or basic the liquid is. Non-invasive means the sensor stays on the surface. Novelna Inc. will develop a sensor for inflammation after a myocardial infarction, the medical name for a heart attack, to watch patients afterward and reduce hospital readmission. The wearable would send its signal through the human body, as a conduit, to cut signal loss and save power. Those four descriptions are the news page’s.

Each team’s award page gives a project name, a principal investigator, a start date, and a ceiling. The news page does not print those fields. The award directory lists the same four projects and the same September start dates. All four award pages list a start date of September 24, 2026, a week before the October 1 announcement, and all four sit in the Health Science Futures Office. The MIT project is PYTHIA: Predictive platform for phYsiological and Therapeutic Health Intervention and Adaptive management. The prime awardee is the Massachusetts Institute of Technology, in Cambridge, Massachusetts. The principal investigator is Ahmad Bahai, PhD. The page prints the ceiling as up to $37.9 million. The NYU project is SENTINEL, a skin wearable for multiplexed inflammation monitoring. Multiplexed means more than one marker at a time. The title also says the wearable is based on FETs, a kind of tiny electronic switch used to read a chemical signal. The prime awardee is New York University, in New York. The principal investigator is Elisa Riedo, PhD. The ceiling is up to $30.8 million. The NYU award page says the system will harvest energy from radio sources including Bluetooth and Wi-Fi. The news page adds the word directly. The University of Washington project title is “A remixable chiplet platform for continuous, picomolar, ultra-low-power biomarker monitoring via de novo protein binders.” Picomolar, in that title, means a very small amount of a marker in the sample. De novo means the protein binders are designed for this job. The prime awardee is the University of Washington, in Seattle. The principal investigator is Quansan Yang, PhD. The ceiling is up to $7.5 million. The award page says the proteins can be refreshed so the sensor stays reliable over long periods. The news page says for longer periods. Novelna’s project is PULSE, Proactive Unified Longitudinal Sensing Ecosystem. The prime awardee is Novelna Inc., in Palo Alto, California. The principal investigator is Ashkan Afshin, MD. The award page says the device will use the human body to conduct data, to maximize battery lifetime. The news page says a conduit, to minimize signal loss and optimize power efficiency. The Novelna page prints the amount as “Up to $33.2” and does not print the word million or the letter M. The other three award pages print an M. The three ceilings that print an M are $37.9 million, $30.8 million, and $7.5 million. Added together they are $76.2 million. That addition is arithmetic. It sits under the $117.4 million program ceiling. It is not a remainder the agency assigned in print to Novelna. Those lines stay with the page that printed them.

Who is quoted. Leonard Tender, PhD, the Delphi program manager, said many people are empowered by the continuous glucose monitor for managing blood glucose. He asked the reader to imagine tracking broader, more detailed information about other parts of health, to help prevent life-threatening emergencies, manage chronic conditions, support safe recovery from injury and illness at home, and promote wellness. He said that with Delphi, ARPA-H is doing something fundamentally different: enabling development and driving widespread adoption of biosensors beyond blood glucose, toward sensors that are individualized, affordable, precise, reliable, and reconfigurable. He said those sensors will improve how health, diseases, and recovery are tracked, and will change how disease is understood, anticipated, and ultimately prevented. A second line from Tender says that, like the Greek legend of the oracle at Delphi, the program encourages all Americans to “know thyself.” Those quotations are his, in the release. A quotation is not a sensor already on a wrist.

What the pages do not say. They do not say the Food and Drug Administration has authorized a Delphi device. They do not report how a patient did on one of these sensors. They do not say a device is for sale or already shipping. They do not print one dollar figure that is the same on every team. The program page, “Developing self-monitoring ecosystems to ‘know thyself,’” describes the problem in its own words: today’s wearable or ingestible sensors provide only biophysical data, like heartrate, or metabolic data, like blood glucose, and the devices are bulky, expensive, hard to update, and often designed as a one-off. Ingestible means swallowed. None of the four award descriptions is an ingestible device. The program page’s solution line names hormones, immune markers, and therapeutic drug levels. The news page names hormones, inflammatory markers, and medication levels. Immune markers and therapeutic drug levels are the program page’s wording. The program page also says the chiplet platform is meant to mix and match across wearable and ingestible biosensors. That mix-and-match line is the program page’s. The same page lists an earlier press release, dated March 10, 2026, “ARPA-H launches program to supercharge biosensor device development with next-generation platform.” That earlier item is the launch. October 1 is the day the agency named the teams.

The picture is the official ARPA-H social card for Delphi. On a dark blue field that shifts toward violet, Delphi is set in large white type. The line under the name reads “Developing self-monitoring ecosystems to ‘know thyself.’” A line drawing of a Greek temple fills the right side, in pale cyan. A human silhouette stands in the doorway, at the top of a path of circuit traces. The steps on that path are magenta and purple blocks. The ARPA-H mark, a white H in a hexagon, sits in the corner. The card names the program and the know-thyself line. It does not print a calendar date.

In plain terms, ARPA-H said on Thursday that four teams will build Delphi, a federal push to replace one-job biosensors with small swappable chips. The targets it names are hormones, inflammation, and drug levels, for chronic conditions, plus on-device AI in the MIT heart-failure patch. The agency’s ceiling is up to $117.4 million over 4.5 years, and each team’s award depends on milestones. The award pages date the four awards to September 24, 2026, and print ceilings of up to $37.9 million for MIT, up to $30.8 million for NYU, and up to $7.5 million for the University of Washington. Novelna’s page prints “Up to $33.2” without a unit. The schedule on the news page runs from shared standards at 18 months, to a first prototype in a living body at 24 months, to a clinical trial or a human-factor study by 54 months. The pages do not say a regulator has cleared a device, and they do not say one is for sale.