NSF Is Putting $75 Million Into Instruments That Do Not Exist Yet — And It Renamed the Program Before Publishing It

There is a specific kind of funding announcement that rewards the people who read it early: the one where the agency commits a dollar figure before it writes the solicitation.

That is what happened on October 8, 2026. Inside a large NSF release tied to the White House's roughly $6 billion science package, the National Science Foundation said it will "soon release" a funding opportunity for Super Intelligence-Native Instruments (SINI), attached to the NSF Programmable Cloud Laboratories Network (NSF PCLNet), with an anticipated investment of up to $75 million over five years.

No solicitation number. No deadline. No award ceiling. No proposal page limit. Just a commitment, a program name, and a design philosophy: the instruments NSF wants to fund are digital-first — built from the architecture up for automated, data-dense, model-driven science rather than retrofitted for it.

If you build scientific instruments, operate a core facility, or run an automation group, this is the most useful thirty days you will get all year. The solicitation does not exist, which means the requirements are not yet fixed, which means your conversations with program officers still land before the text hardens.

The rename is not a footnote

Start with the thing that makes this announcement strange to search for.

NSF's own October 8 release calls the program Super Intelligence-Native Instruments (SINI). Trade coverage published within hours of the same announcement calls it Artificial Intelligence-Native Instruments (AINI). Both are reporting the same $75 million, the same five-year horizon, and the same PCLNet attachment.

The split traces to a terminology directive. As reported the same week, NSF officials have been directing that references to "artificial intelligence" in agency grant language be rewritten as "Super Intelligence," following a White House executive order — a rewrite that, by at least one account, reaches further than the order itself requires, extending to abstracts and historical documents the order did not mandate.

Treat that as an operational fact rather than a political one, because the consequence is concrete and immediate:

If you track NSF opportunities by keyword, "AI" is no longer a reliable filter. A saved search on "artificial intelligence," an RSS alert keyed to "AI-enabled," a Grants.gov keyword subscription built in 2025 — all of them now have a blind spot the size of NSF's AI portfolio. The $75 million instruments program is the first case where the renamed term is the only term on the agency's own page.

Practical fix, today: add super intelligence, SI-native, SI-enabled, and SINI to every alert you maintain, and keep the AI variants rather than replacing them. Coverage, agency pages, and third-party aggregators are going to disagree with each other for months, and the renamed version is the one that will appear on the actual NOFO.

What PCLNet already is, and why SINI sits on top of it

SINI is not a standalone idea. It is the instrument layer of a build-out that is already two phases deep, and understanding the lineage tells you what kind of proposal will be competitive.

Phase one — the test bed (August 2025). NSF published NSF Test Bed: Toward a Network of Programmable Cloud Laboratories (NSF 25-541) on August 5, 2025: up to $100 million, up to six awards, each at roughly $5 million per year for four years. Eligibility was broad — institutions of higher education, nonprofits, and for-profit organizations. Initial priority areas were biotechnology and materials science. The program is run out of the Directorate for Technology, Innovation and Partnerships (NSF TIP). The pitch was remote-accessible, AI-enabled laboratories that execute user-programmed workflows, with automation across all three phases of an experiment: design and prediction before, real-time sensor monitoring and condition control during, analysis and next-step planning after.

Phase two — the network (July 2026). NSF scaled the test bed concept into a national network, PCLNet, at roughly $400 million across about twenty nodes. Reporting has rendered the headline figure variously as $380 million and $400 million, which is normal for an announcement that bundles several appropriations lines; treat the number as approximate and the architecture as the real signal. We covered that build-out in NSF's $380 million programmable cloud laboratories push.

Phase three — the instruments (announced October 8, 2026). SINI. Up to $75 million over five years for the hardware and architecture that will populate those nodes.

That sequence matters for one reason: the nodes came first. SINI awardees are expected to "enhance NSF PCLNet" — meaning the instruments are not being funded as free-standing capital purchases. They are being funded as contributions to a network that already has, or will have, operators, users, and data pipelines. A proposal that treats the instrument as the deliverable is answering a question NSF did not ask. A proposal that treats the instrument as a node-compatible, remotely programmable, API-addressable component is answering the one it did.

What "digital-first" probably means in review

NSF has not published criteria. But the language it chose — instruments "specifically designed for the modern era of data," designed so that digital access is intrinsic to the architecture rather than bolted on — maps onto a reasonably predictable set of requirements. Based on how PCLNet's earlier phases were scoped, expect reviewers to look for some version of:

• Programmatic control as the primary interface. Not a GUI with a scripting plugin. An instrument whose full capability surface is reachable by API, because the intended user may be a model rather than a graduate student.
• Native data output, not exported data. Structured, metadata-complete, machine-readable at the point of capture. If the instrument's default output requires a human to clean it before a model can use it, it is not digital-first.
• Remote operability. Cloud laboratories are useless if the instrument requires a person standing next to it to change conditions.
• Closed-loop capability. The ability to accept a next-step instruction derived from the previous result, without a human in the loop for each cycle.
• Interoperability with the existing network. Compatibility with PCLNet node infrastructure, and plausibly with NSF's broader data stack — the Integrated Data Systems and Services program, the National AI Research Resource, and the American Science and Security Platform all appeared in NSF's July dataset initiative as resources applicants were expected to leverage.

If your instrument concept is strong on physics and weak on all five of these, you have a build problem, not a writing problem, and you have roughly one solicitation cycle to fix it.

Two timing problems to plan around

Problem one: the appropriations calendar. A forthcoming NOFO is subject to the same constraint as every other new federal start right now. The continuing resolution that kept the government open carries a new-starts prohibition and a rate-of-operations limit that run to December 11, 2026 — which we broke down in what actually happens to grants between October 1 and December 11. "Soon" in an October announcement can mean December or later for a program that requires new budget authority. Plan for a solicitation that posts late and moves fast, not one that posts on schedule.

Problem two: NSF's own obligation behavior. NSF closed FY2026 having obligated $6.4 billion in research funds against $8.3 billion the prior year, across 6,218 awards versus 8,647 — with roughly $1.7 billion carried into FY2027 to fund administration priorities. The full accounting is in NSF's FY2026 close-out.

Read those two facts together and SINI's $75 million looks less like new money and more like where the carryover goes. That is good news for instrument builders and bad news for everyone whose core-program proposal is sitting in a queue. The agency is not short on cash; it is reallocating, and the reallocation favors programs that appear in White House science initiatives. SINI appears in one.

What to do in the next thirty days

If you build or integrate instruments: identify the PCLNet node operators in your domain and open a conversation now. SINI awardees enhance the network; the network's existing operators are the people who can tell you what the network is missing. A letter of collaboration from a node is worth more than a paragraph about potential compatibility.

If you run a core facility: audit your flagship instrument against the five digital-first criteria above and write down which ones fail. That list is your proposal's project plan.

If you are a for-profit instrument company: note that PCLNet's predecessor was open to for-profits at full eligibility, not just as subawardees. That is unusual for NSF infrastructure money and worth confirming the moment the solicitation text appears.

Everyone: fix your keyword alerts. The single highest-return action available this week is making sure that when a solicitation titled something with "Super Intelligence" in it posts in December, your monitoring actually catches it. The $75 million is real. The name you are searching for is not the name NSF is using.