The pitch writes itself, and I have now heard three versions of it. Minnesota should go win something. Land a national AI institute, a federal consortium, a flagship anchor tenant, and the cluster grows up around it — the way it grew up around Austin after Austin won the Microelectronics and Computer Technology Corporation in 1983 and SEMATECH in 1988. Get the trophy, get the decade.
I am going to argue for a Minnesota AI play, and I am going to start by conceding that the precedent everyone reaches for did not hold.
Both of Austin’s trophies are gone.
The claim I want to defend is narrower and more durable. Minnesota should stop thinking about AI economic development as a race for compute and start thinking about it as a race for assurance — the capacity to establish, on evidence, that a given AI system does what its seller says it does. Compute is a capital-expenditure auction, and Minnesota does not win auctions — the state’s own business lobby explains why in some detail, quoted below. Assurance is a research-and-labor good. It lives in people, published methods, accredited practice, and institutions that certify competence, which makes it the one category of regional advantage that cannot pack up and move when another state writes a bigger check.
Two disclosures first, because both should change how much weight you give this. My firm would make money in a Minnesota AI-assurance market. I already sell verification of a narrow kind — every citation in every filing that leaves this office runs through a gate built for exactly that purpose — and a state where somebody has to certify that a model behaves is a state where my kind of work gets more valuable. I am also a candidate for statewide office in 2026, which gives me an obvious second reason to want Minnesota to look like it has a future. Discount accordingly and check the numbers; several of them I computed myself out of two federal spreadsheets, so you can rerun them.
Both of Austin’s trophies are gone
Start with what actually happened, in full, because the version that gets told at economic development lunches stops about twenty years early.
MCC was the American answer to Japan’s Fifth Generation Computer Systems project — a for-profit research consortium funded by twelve member companies and run by Admiral Bobby Inman. When it went looking for a home, the Texas State Historical Association records that “fifty-seven communities submitted bids,” narrowed to four finalists: San Diego, Atlanta, North Carolina’s Research Triangle, and Austin. Austin was selected in July of 1983. The winning package included a subsidized lease at the Balcones Research Center next to UT-Austin, low-cost loans, and reduced mortgage rates for relocating personnel, and the University of Texas and Texas A&M agreed to substantially upgrade their computer science and electrical engineering departments.
The academic accounting is more specific, and the specificity matters later. MCC received more than $20 million in incentives, of which two of the major ones, per the IC² Institute’s 2016 assessment, were a facility and lab at UT-Austin “leased for $10 a year (financed by university and private statewide contributions) and the creation of 32 $1 million endowed chairs in engineering and science.”
Five years later Austin did it again. SEMATECH — the semiconductor consortium incorporated in 1987 with fourteen founding members — “chose Austin over 137 competing cities,” and “the consortium received $62 million in incentives from the state of Texas.” The state money was not the whole of it, and DARPA still describes the federal side on its own site: “Beginning in 1987, the SEMATECH consortium received funding from the Federal Government to help revitalize the U.S. chipmaking industry,” and “A decade after its founding, in 1997, the consortium was standing on its own without annual funding from the Government.”
Now the part that gets left out.
By the turn of the century the personal computer and the internet had made most of MCC’s research agenda irrelevant. In June 2000 the MCC board voted to dissolve the consortium. It had fifty-eight employees left, down from almost four hundred at its mid-1980s peak. Formal dissolution papers were filed in 2004.
SEMATECH did not dissolve. It moved. In May 2007 the trade press reported that International SEMATECH was relocating its headquarters from Austin to its operations at the University at Albany, against a five-year, $300 million commitment from the State of New York. Five years later the Austin American-Statesman reported another research program and dozens more jobs going to Albany, calling it “another major step in its gradual exodus from Austin” — and noting that Texas had signed a $40 million incentive deal in 2004 meant to keep the headquarters and the jobs, which the governor’s office was by then weighing whether the consortium had honored.
So: the first trophy was liquidated and the second was outbid. If you are going to cite Austin as the model, cite all of it — a hostile fact-checker will find the rest in about ten minutes, and the version of this argument that survives him is the better argument anyway.
What left Austin was the part that had been bought
Here is the turn, and it is the whole essay.
MCC was a corporation. SEMATECH was a corporation. A corporation is a lease on an activity — it has a headquarters, and a headquarters can be picked up and set down in Albany when Albany offers three hundred million dollars for it. Thirty-two endowed chairs in engineering and science cannot be picked up. An upgraded electrical engineering department cannot be picked up. Neither can the engineers. Of SEMATECH’s four hundred technical employees, roughly two hundred and twenty were assignees sent to Austin by member firms for six to thirty months — and the IC² researchers record that many of them ultimately stayed in the city.
The scholars who studied this most closely reach a conclusion that is less flattering to the trophy theory than the trophy theory admits. Elsie Echeverri-Carroll and Michael Oden, in the Kauffman-funded 2016 assessment for UT-Austin’s IC² Institute — a genuinely balanced document, not a chamber of commerce pamphlet — put it this way: “Although the location of Sematech and MCC raised the floor of the talent market, Austin’s entrepreneurial success after the location of the research consortia has been mainly tied to the continuous migration of talent, mostly from other cities in Texas and Silicon Valley.”
That sentence concedes and asserts in equal measure. It concedes that the consortia mattered — they “raised the floor of the talent market.” It asserts that the mechanism which built the economy was the sustained inflow of skilled people, on top of entrepreneurship infrastructure the same report credits substantially to George Kozmetsky rather than to either consortium. The consortia were catalysts inside a multi-decade buildout. They were not the machine.
That is the useful lesson, not the discouraging one. Both of the things Texas bought went away, and everything Texas built stayed — the departments, the chairs, the people, and the reputational fact that the Greater Austin Chamber could now get a California CEO on the phone. Texas paid $62 million for a headquarters that left in nineteen years, and five years before that had bought a corporation that liquidated. Out of the two deals it got, as a byproduct, thirty-two endowed chairs, a permanent upgrade to two engineering schools, and a generation of engineers who did not go home. Reverse-engineer a strategy out of that and you do not chase the headquarters. You go straight at the byproduct.
Minnesota loses auctions, and the compute race is an auction
In July I argued in this section that the durable value in this industry migrates down the stack — to compute, and beneath it to power. That carries a corollary Minnesota economic development has not absorbed: competing for durable value in compute means competing on capital expenditure, land, transmission, and tax treatment. That is an auction, and auctions go to the bidder with the lowest cost basis and the fastest permitting.
Which describes almost no part of Minnesota’s position. The Minnesota Chamber of Commerce — a business advocacy organization with no reason to flatter my thesis — reports in its 2026 Business Benchmarks that “Minnesota’s economy is no longer keeping pace with the nation,” that the state ranks “44th for overall tax competitiveness, with the 2nd highest corporate rate and 6th highest personal income tax rate,” and that from 2020 to 2024 it ranked 41st in net domestic migration, a net loss of 47,930 residents. The Chamber does list data centers among the “fast-growing sectors” Minnesota is “well-positioned to attract.” Maybe. But a data center is a building with a substation attached, biddable precisely because nothing in it could not be built somewhere else next year for a slightly larger abatement. That is a relocatable asset by definition, and it is why the compute race is the wrong primary play for a high-cost state whose labor force, on the Chamber’s own numbers, has grown 0.2 percent a year since 2019.
Minnesota already won the recognition and did not get the money
The other reason to stop talking about starting from zero is that Minnesota did not start from zero. It competed, half-won, and almost nobody here talks about the half it lost.
On October 23, 2023, the Economic Development Administration designated Minnesota MedTech Hub 3.0 one of the 31 inaugural Tech Hubs authorized by the CHIPS and Science Act. The EDA’s own release describes it as “a consortium led by Minneapolis Saint Paul Economic Development Partnership” that will make the state a global center for Smart MedTech “by integrating artificial intelligence, machine learning, and data science into medical technology.” That is not an AI-adjacent designation. That is an AI designation, awarded federally in October 2023, on the strength of assets Minnesota already had.
The same release told the designees what came next: “Designated Tech Hubs are now eligible to apply for the next phase of the Tech Hubs Program that will invest between $50-$75 million in each of 5-10 Designated Hubs.”
Phase 2 landed on July 2, 2024. EDA announced $504 million in implementation funding, in grants “ranging between $19 million and $51 million,” to 12 of the 31 designated Hubs, across projects serving fourteen states: Colorado, Florida, Georgia, Illinois, Indiana, Montana, Nevada, New Hampshire, New Mexico, New York, Ohio, Oklahoma, South Carolina, and Wisconsin.
Minnesota is not on that list. Wisconsin is.
A note on sourcing. Both EDA documents quoted above were retrieved through Internet Archive snapshots of the identical eda.gov URLs, because eda.gov serves a bot challenge to automated retrieval. Original URLs and archive timestamps are in the sources block.
This is not a story about a federal snub. Twelve of thirty-one got funded and nineteen did not; reasonable panels make defensible choices. It is that Minnesota has already been told, by a federal agency running a competitive process, that it holds a real AI-in-medtech asset base — and that the recognition by itself bought nothing. A designation is a finding of fact. It is not a program. Something has to be built underneath it.
Minnesota drafted a license for a profession whose science it has not funded
Somebody already started, in the one place nobody was watching.
On March 23, 2026, House File 4544 was introduced by Representatives Koegel, Rymer, Norris and Bahner, “relating to commerce; establishing a license for artificial intelligence independent verification organizations; establishing an advisory council; authorizing rulemaking; requiring reports; proposing coding for new law in Minnesota Statutes, chapter 325M.” It was read for the first time and referred to the Committee on Commerce Finance and Policy. A Senate companion, S.F. 4636, was introduced the same day by Senators Frentz and Lucero off the identical Revisor draft and referred to Commerce and Consumer Protection. Neither chamber recorded any action after those referrals, the 2026 regular session adjourned in May without touching either bill, and neither was enacted — the Revisor’s table of statutes affected by the 2026 session shows exactly two changes in chapter 325M, and neither is this. Nothing here is law, and what follows describes a proposal that died in committee.
It proposes a state licensing regime for private AI auditors. A licensed “independent verification organization” — an IVO — would be, in the bill’s words, “an entity licensed by the commissioner pursuant to section 325M.51 to assess an artificial intelligence model’s or artificial intelligence application’s adherence to standards reflecting best practices for the prevention of personal injury and property damage.”
The architecture is unusual. The applicant files a proposed plan identifying the risks to be mitigated, and for each one the plan must include “a proposed definition of acceptable levels of risk” and “metrics that are measurable and can be used to determine whether the acceptable level of risk defined by the IVO produces beneficial outcomes.” The Commissioner of Commerce may license the applicant on finding that “the applicant demonstrated independence from the artificial intelligence community” and that “every element of the applicant’s proposed plan is adequate to ensure that artificial intelligence models or artificial intelligence applications verified pursuant to the plan mitigate one or more risks to an acceptable level.”
So the certifier writes the standard, and the Department of Commerce grades it — through an advisory council the bill requires the commissioner to appoint and to delegate licensing to.
Verification would be voluntary — the bill says so in terms: “Nothing in this section requires an artificial intelligence model or artificial intelligence application to seek IVO verification.” Which raises the question of why anyone would buy it. Proposed section 325M.54 answers in three lines:
In a civil action asserting claims for personal injury or property damage caused by an artificial intelligence model or artificial intelligence application, there is a rebuttable presumption against liability if: (1) the artificial intelligence model or artificial intelligence application in question was verified by a licensed IVO at the time of the plaintiff’s alleged injury; (2) the plaintiff’s alleged injury arose from a risk that the IVO was licensed to verify and for which the IVO did verify the artificial intelligence model or artificial intelligence application; and (3) the artificial intelligence model or artificial intelligence application is within the specified market segment, if any, for which the IVO was licensed to conduct verification.
I try cases. A rebuttable presumption against liability is not a footnote — it is a burden shift in the defendant’s favor in every personal-injury and property-damage case within its reach, and it is the most commercially valuable thing in this bill by an order of magnitude. It is what makes the certificate worth paying for.
Now put the halves together. The certificate shifts a burden in a Minnesota courtroom. The standard it is measured against is written by the company selling it. The adequacy of that standard is judged by the Department of Commerce. And no Minnesota research institution is funded to develop, publish, or contest the methods by which any of it would be measured.
None of that is an attack on the bill. The instinct behind it is sound and better than most of what circulates federally: voluntary rather than mandatory, keyed to demonstrated risk categories rather than to a developer’s revenue, and applied to open and closed systems alike — which means it survives the subtraction test I proposed in Protectionism in a Safety Costume where the federal frameworks do not. Nothing here argues for restricting anything. It argues that a credential is worth exactly as much as the science it certifies against, and Minnesota drafted the credential first.
That gap is not a drafting problem better language can fix. It is the open technical problem, and I have written about it twice in the last week: the measurable distance between what an AI developer intended and what anyone can prove, and the fact that a model checkpoint is an untrusted executable whose computational graph can carry a backdoor no file-format fix reaches. Neither is solved; both are the live frontier of an active research field. An auditor licensed to certify against an unsolved problem is a credential with nothing underneath it, and the state that issues it owns the consequences when a jury asks what the certificate actually measured. Which is the argument for building the science in the state that wants the auditors.
Assurance is payroll, not capital, and that is the entire point
The strongest objection to everything above is not that assurance is unimportant. It is that assurance may never become a thing anyone buys separately.
State it in its best form. Verification could collapse into the model vendors themselves, the way software security largely collapsed into the cloud providers — sold as a feature rather than bought as an independent audit. Frontier labs already publish system cards and run internal red teams, and a customer who trusts the vendor’s evaluation will not pay a third party for a second one. On that reading the assurance market is a regulatory artifact that exists only where a statute conjures it, and building a regional capability around it means training a workforce for a job that gets absorbed. That is a serious argument and it is not obviously wrong.
It loses anyway, for a reason older than this industry. Third-party assurance does not exist because regulators are fond of it. It exists wherever the buyer of a system cannot inspect it and the seller has an interest in the answer — financial statements, structural steel, elevators, clinical trials, and now this. Self-certification holds up right until the first contested injury, and then the question in the courtroom is who checked and who paid them. Minnesota’s own bill has already conceded the point by requiring that an IVO demonstrate “independence from the artificial intelligence community.”
And the largest jurisdiction that has actually legislated this has written the answer into binding law, in language worth reading twice. The EU AI Act defines a “conformity assessment body” as “a body that performs third-party conformity assessment activities, including testing, certification and inspection,” and a “notified body” as one of those bodies notified under the Regulation. Then Article 31 sets out what such a body must have. Not what it must buy. What it must staff:
The notified body shall have permanent availability of sufficient administrative, technical, legal and scientific personnel who possess experience and knowledge relating to the relevant types of AI systems, data and data computing, and relating to the requirements set out in Section 2.
That is a staffing requirement written into binding law, sitting beside an obligation in the same article to “take out appropriate liability insurance for their conformity assessment activities” — an obligation that lifts only where the Member State assumes the liability itself. Independence, competence, permanence, insurance: the specifications of a professional services firm, not of a data center.
Which is the structural claim in one line. A fab is capital and an audit is payroll. Capital relocates in a board meeting. A workforce relocates one household at a time, slowly, and mostly it does not. That asymmetry is not a slogan — it is the observable difference between what Texas lost and what Texas kept.
Minnesota has both tiers of this already, and uses neither for it
Assurance work is not one job. It is two, and Minnesota is organized in exactly two tiers.
The upper tier develops the methods, and it is more substantial than the state gives itself credit for. The University of Minnesota Twin Cities spent $1,409,710,000 on research and development in fiscal year 2024 — 21st in the country among the 681 institutions that received the full form of the National Center for Science and Engineering Statistics’ most recent Higher Education R&D survey, and twelfth among public universities on that survey’s own public-private classification. It is not the largest research university in the country and it is not close; Johns Hopkins spent nearly three times as much. It is one of twenty-one public universities that spend over a billion dollars a year on research, and the National Academy of Inventors ranked it 34th worldwide in utility patents granted for 2025 — 91 patents, and a top-50 placement every year since that ranking began in 2013. That is a real research plant, and none of it is pointed at the problem of proving what an AI system does.
The lower tier applies the methods, and this is where it gets practical, because most verification work is technician work. Somebody runs the eval suite, maintains the test corpus, documents a model artifact’s provenance, checks a hash against a signature, and writes the finding up in a form that survives cross-examination. That is a two-year credential and a career, not a dissertation. Minnesota State describes itself as “the third largest system of state colleges and universities in the United States and the largest in the state with 26 colleges, 7 universities, and 54 campuses,” serving “more than 290,000 students each year.” Fifty-four campuses is a distribution network for exactly that credential, reaching parts of the state no data center will ever reach.
The two systems are separate by design, and the separation is usually called a defect. One is statutory: Minnesota State exists under Minn. Stat. ch. 136F and can be restructured by ordinary legislation. The other is constitutional. Minn. Const. art. XIII, § 3 reads, in full: “All the rights, immunities, franchises and endowments heretofore granted or conferred upon the University of Minnesota are perpetuated unto the university.” Minnesota courts have built a body of doctrine on that clause about the Regents’ autonomy from ordinary legislative direction; I will not characterize case law in an essay, and the constitutional text is the point. For an assurance strategy the split is no defect. Research capacity the legislature cannot command, alongside a teaching network it can direct, is a workable division of labor for building a discipline — provided somebody notices both halves are required.
And here is the tell. The Chamber’s report, in the same breath as its optimism about advanced sectors, records that Minnesota “ranks 49th in high schools offering computer science and 44th in college graduates earning STEM degrees.” Forty-ninth. A state whose legislature has drafted a license for AI auditors is forty-ninth at teaching high school students the subject those auditors would have to know.
The inconvenient number: this state is spending a stock it is not replacing
The strongest case against acting is that Minnesota is fine — it has the patents, the medical device industry, the research university. Why manufacture an urgency?
Because the Chamber’s report carries both numbers, and the second one governs. Minnesota “produces the 5th-highest number of patents per capita in the nation.” And: “Patent activity has dropped since 2014, ranking Minnesota 47th in patent growth over the last decade. Research and Development (R&D) output ranks 16th nationally, but R&D growth slipped to 45th in R&D between 2016 and 2021.” Fifth in level, forty-seventh in rate. Sixteenth in level, forty-fifth in rate.
The federal data says the same thing about the University, and I computed this part myself so you can check it. NCSES Table 13 carries each ranked institution’s spending back to fiscal 2010, where the institution reported that year. Sort it by the fiscal 2010 column and the University of Minnesota Twin Cities ranks 13th — a rank within that table, subject to the two limits set out in the sources below. Sort it by fiscal 2024 and it ranks 21st. Across those fourteen years the University’s research spending grew about 79 percent, from $786 million to $1.41 billion, while academic R&D nationally grew about 92 percent, from $61.3 billion to $117.6 billion. The University did not shrink. It grew — just slower than the field, in nine of those fourteen years and across the span as a whole — and it slid eight places doing it.
That is what living on a stock looks like from the inside: every single year is fine, the absolute numbers keep rising, and the ranking quietly slides. It argues for moving now, and it is equally an admission that the trend runs against the thing I am arguing for. A state that is 5th in patents and 47th in patent growth has capacity and no momentum, and capacity without momentum is a depreciating asset — the same thing I said about a frontier model, for the same reason.
So the case is not that Minnesota should go win a trophy. Trophies leave. It is that assurance is about to become a purchased service here whether or not anybody plans for it — a bill is already drafted to create the sellers and a liability presumption to create the demand — and the inputs to supply it sit in a billion-dollar research university with constitutional standing and a fifty-four-campus college system, neither of which has been asked. Connecting those does not require winning an auction.
And the profession’s stake, since I write this for other lawyers: every verification capability built here is one my clients can eventually buy and a standard my cross-examinations can eventually run against. The alternative is that when the first Minnesota jury is asked what an AI certificate actually proved, the only people in the room who can answer are the ones who sold it.
What would prove me wrong
Four conditions, stated tightly enough to be checked against.
Assurance never becomes a distinct purchased service. If four years from now the model vendors have absorbed verification into the product — system cards, internal red teams, contractual warranties — and no independent assurance industry of any size exists outside the jurisdictions that legislated one into being, then the market I am telling Minnesota to supply does not exist and I was training a workforce for nothing. The observable is whether anyone pays a third party for this in a state with no statute requiring it.
HF 4544-style licensing passes and works fine with no public research base. If Minnesota enacts an IVO regime, private certifiers file plans, the Department of Commerce evaluates them competently, and five years of litigation produces no case in which the adequacy of a certifier’s own standard becomes the contested issue, then the research base I say is missing was not load-bearing. I would concede that in public.
Somebody else builds this and it relocates anyway. The title of this essay is a claim and it is falsifiable. If another state builds real AI-assurance capability — trained people, published methods, accredited programs — and a larger state then buys it away the way New York bought SEMATECH’s headquarters, my distinction between what can be purchased and what has to be built collapses. Watch whether the people move, not whether an organization’s letterhead does.
The compute race turns out to be winnable here. If Minnesota lands durable AI infrastructure on capital-expenditure terms — not a data center that arrives with a twenty-year abatement and leaves when a better one appears, but something that stays after the abatement expires — then my premise that this state cannot win auctions was wrong and the auction was the better play.
The disclosure belongs next to the prediction rather than in the footer, so, again: I would make money in the world I am describing, and I am running for statewide office in a state I have just spent several thousand words arguing is falling behind. Read the numbers rather than me. They are all below, and the ones I calculated are reproducible from spreadsheets anybody can download.
Sources
- NCSES, Higher Education R&D Survey, FY 2024, Table 13 (from the survey landing page) — workbook downloaded and parsed directly. It supplies U. Minnesota, Twin Cities at rank 21 with $1,409,710 thousand (FY2024) and $786,074 thousand (FY2010); Johns Hopkins at rank 1 with $4,129,264 thousand; all-institution totals of $61,286,610 thousand (FY2010) and $117,554,145 thousand (FY2024); and 681 ranked institutions. The 681 is the standard-form population: NCSES’s methodology for this survey states that “[f]or the FY 2024 cycle, the short-form population included 244 institutions that reported R&D expenditures between $150,000 and $1 million during FY 2023. The remainder of the institutions (681) received the full version of the survey.” Every short-form institution is therefore under $1 million and none can affect a rank near the top. Public-private classification and the count of public institutions above $1 billion come from Table 5 of the same publication, which groups every institution by state and institutional control; it classifies Pittsburgh as public.
- The FY2010 rank of 13th is my own calculation, and it carries two limits, both of which cut against precision rather than for it. First, it comes from sorting Table 13’s FY2010 column, which contains only institutions present in the FY2024 ranking — a school that reported in FY2010 but not in FY2024 is absent from the comparison. Second, and larger: 105 of the 681 have no FY2010 figure at all, including the institution ranked 14th in FY2024. That one is University of Maryland, and NCSES explains why in its own methodology — UMD College Park and UMD Baltimore “began reporting as one unit, University of Maryland, in FY 2019,” and the combined unit reported $1,096 million in FY 2019. Read the 13th as a rank within this table on this table’s terms, not as a published FY2010 ranking.
- Texas State Historical Association, Microelectronics and Computer Technology Corporation [MCC] — “fifty-seven communities submitted bids”; the four finalists; selection in July 1983; the incentive package; “In June 2000 the MCC board voted to dissolve the consortium,” fifty-eight employees remaining against “almost 400 workers” at the mid-1980s peak; dissolution papers filed 2004
- Elsie Echeverri-Carroll & Michael Oden, Preliminary Assessment of the Factors That Led Austin to Become a High-Tech Entrepreneurial City (IC² Institute, UT-Austin, Kauffman Foundation-funded, July 31, 2016) — raw PDF text extracted. At 3: SEMATECH “chose Austin over 137 competing cities,” “$62 million in incentives from the state of Texas,” and (n.4) the $10-a-year UT lab lease and “the creation of 32 $1 million endowed chairs in engineering and science.” At 17: the SEMATECH assignee figures and “Although the location of Sematech and MCC raised the floor of the talent market, Austin’s entrepreneurial success after the location of the research consortia has been mainly tied to the continuous migration of talent, mostly from other cities in Texas and Silicon Valley.”
- DARPA, SEMATECH — federal funding “Beginning in 1987”; “A decade after its founding, in 1997, the consortium was standing on its own without annual funding from the Government”
- SEMATECH shifting HQ to Albany (May 2007) and Kirk Ladendorf, Sematech moving dozens of jobs from Austin to Albany (Austin American-Statesman, Sept. 1, 2012) — the 2007 headquarters relocation, New York’s five-year $300 million commitment, “another major step in its gradual exodus from Austin,” and the $40 million 2004 Texas incentive agreement. These are trade-press and newspaper accounts, not primary budget documents; the fact and year of the move are corroborated across independent outlets, the dollar figures are as reported.
- EDA, Biden-Harris Administration Designates Tech Hub in Minnesota to Boost Development of Smart Medical Technology (Oct. 23, 2023) — the designation, the “Smart MedTech” framing, and the $50–$75 million-to-5–10-hubs statement of what Phase 2 would be
- EDA, Tech Hubs Phase 2 Portfolio Fact Sheet (July 2, 2024) — $504 million, grants “ranging between $19 million and $51 million,” 12 of 31 hubs, and the fourteen states served. Both EDA documents are Internet Archive snapshots of the identical eda.gov URLs (originals:
eda.gov/news/press-release/2023/10/23/Minnesota-MedTech-Hub-3.0andeda.gov/sites/default/files/2024-07/EDA_TECH_HUBS_Phase_2_Fact_Sheet.pdf), because eda.gov serves a bot challenge to automated retrieval. - Minnesota Legislature, H.F. 4544, 94th Session (2026) — bill title, authors, March 23, 2026 first reading and referral to Commerce Finance and Policy; proposed §§ 325M.50–.54, including the IVO definition, the plan requirements (“a proposed definition of acceptable levels of risk”), the licensing findings, the voluntariness provision, and the § 325M.54 rebuttable presumption. Status verified against the Revisor’s House status page and the Senate status page for the companion, S.F. 4636 (Frentz and Lucero; same Revisor draft number 26-07903; referred to Commerce and Consumer Protection, no further action). Non-enactment confirmed against the complete 2026 session-law chapter list and the Revisor’s table of statutes affected by the 2026 session, which shows only §§ 325M.33 and 325M.40 changed. Introduced, died in committee, not enacted.
- Minn. Const. art. XIII, § 3 — quoted in full from the Revisor’s raw text
- Minn. Stat. ch. 136F (2025) — “MINNESOTA STATE COLLEGES AND UNIVERSITIES,” the statutory basis of the Minnesota State system
- Minnesota State, Extraordinary Impact — “the third largest system of state colleges and universities in the United States and the largest in the state with 26 colleges, 7 universities, and 54 campuses”; “more than 290,000 students each year.” This is Minnesota State’s characterization of itself and I have not checked the national ranking against an independent source.
- Minnesota Chamber of Commerce, 2026 Business Benchmarks Report — “Minnesota’s economy is no longer keeping pace with the nation”; “44th for overall tax competitiveness, with the 2nd highest corporate rate and 6th highest personal income tax rate”; 41st in net domestic migration and 47,930 net residents lost 2020–2024; “5th-highest number of patents per capita”; “47th in patent growth over the last decade”; R&D output “16th nationally,” and “R&D growth slipped to 45th in R&D between 2016 and 2021”; “49th in high schools offering computer science and 44th in college graduates earning STEM degrees”
- University of Minnesota, University of Minnesota ranked 34th worldwide in patents by the National Academy of Inventors (Feb. 15, 2026) — 34th for 2025, 91 patents, top 50 every year since 2013
- Regulation (EU) 2024/1689 (the AI Act) — art. 3(21) and 3(22) definitions of “conformity assessment body” and “notified body”; art. 31(9) liability insurance and art. 31(11) the permanent-personnel requirement
- Companion essays in this section: The Model Will Be Free. The Electricity Won’t., Protectionism in a Safety Costume, The Gap Between Intended and Proven, Executable Code You Are Calling a Model, and The Commons Your Practice Runs On
Commentary on technology policy and the economics of the legal and technology sectors — the opinions, predictions, and stated falsifiers are the author’s. Not legal advice, not investment advice, not a comment on the constitutionality or merits of any pending legislation, and not an endorsement of or opposition to any bill. H.F. 4544 is described as introduced and never enacted; nothing here should be read as a statement of enacted Minnesota law. The author’s interests are disclosed in the text: this firm would benefit commercially from a Minnesota AI-assurance market, and the author is a candidate for statewide office in 2026. No client information appears in this article. Questions about anything here: Send us a message or 612-470-6529.