Greece’s AI strategy: DAEDALUS, OpenAI school pilot, and a founders’ checklist

“We’re already fighting yesterday’s battle”: Greece’s prime minister gets candid about AI

“Sometimes I feel that we’re already fighting yesterday’s battle, ” Kyriakos Mitsotakis told roughly 250 founders, investors and operators at an Endeavor Greece event in San Francisco. He wasn’t selling utopia. He admitted the blunt, uncomfortable truth: governments are scrambling to catch up while AI reshapes schools, jobs and the social fabric faster than policy can keep pace.

Mitsotakis used the trip, part fact‑finding and part pitch, to do two things at once. He presented concrete moves Greece has already launched (a classroom pilot with OpenAI, and a national supercomputer called DAEDALUS assembled with Hewlett Packard Enterprise and bound for Lavrio) and he raised the messy questions many leaders avoid: What happens when kids grow up with digital companions? How do we protect minors in classroom pilots? How fast will jobs vanish, and who’s planning re‑skilling at that speed?

What Greece has put on the table

OpenAI for Greece. Per OpenAI’s announcement, the government signed a memorandum with OpenAI, the Onassis Foundation and Endeavor Greece to pilot ChatGPT Edu in upper‑secondary schools this academic year, provide credits and mentorship for startups, and run safety‑by‑design workshops and an accelerator aimed at retaining STEM talent and spawning local AI firms.
DAEDALUS supercomputer in Lavrio. The Ministry of Digital Governance reports that DAEDALUS has been assembled and entered final testing with HPE before transport to Lavrio Technology Park. Phase A is financed with €29, 200, 000 (including VAT) from Recovery & Resilience Fund/Public Investment Program money and Phase B has up to €12, 679, 900 (excluding VAT) in EuroHPC Joint Undertaking funding. Officials describe DAEDALUS as the computing core of the government’s “AI Factory Pharos, ” intended to serve startups, universities and public bodies.
Policy incentives and reforms. Mitsotakis highlighted tax and visa incentives to attract founders and returnees, changes to how stock options are taxed, and labor‑law loosening designed to make Greece friendlier to startups. He also said Greece received roughly €36 billion from the EU’s post‑COVID recovery fund to invest in digital infrastructure. These are government claims and should be checked against the specific legislative texts and EU documentation for implementation details and timelines.
Data‑center courting. Mitsotakis noted Microsoft and AWS projects, a Microsoft cluster near Athens and an AWS deal with Greece’s largest utility to build a major data center in a former coal region, as part of an effort to attract capacity and cloud partners. He added that Greece “has not had any significant reaction” to data centers so far; local environmental and community impacts remain a live question elsewhere and deserve scrutiny here as well.

Why these moves matter, and where to add caution

DAEDALUS and an OpenAI classroom pilot are more than PR. A national supercomputer expands local high‑performance computing (HPC) capacity for climate modeling, genomics, simulation and inference workloads, and the OpenAI accelerator can help convert graduate talent into home‑grown startups.

Don’t conflate DAEDALUS with the thousand‑GPU training farms used by frontier model developers. The ministry and HPE have described assembly and testing, and published funding figures, but they have not released detailed hardware specs (GPU counts, interconnect topology, storage scale) or the access and governance rules founders will need to plan around. In practice, high‑end foundation‑model training typically requires sustained clusters with specific GPU types, NVLink fabrics and multi‑petabyte storage, capabilities that aren’t interchangeable with every HPC design.

The OpenAI pilot is explicitly scoped to reduce teachers’ administrative workloads and test personalized tutoring. OpenAI’s announcement highlighted regionally diverse teacher participation and accelerator support, but public, measurable outcomes, time saved per teacher, student‑learning improvements and the pilot’s data‑protection safeguards, have not yet been published in full. Those metrics will determine whether the pilot scales or stalls.

The hard questions Mitsotakis put on the table

“What does it mean for our kids to grow up with digital companions or boyfriends or girlfriends?”

That line landed because it compresses several policy issues: moderation, age‑appropriate filtering, mental‑health risks, and legal liability when synthetic agents replace or augment human relationships. For school deployments, it raises clear questions about parental consent, retention of student data under GDPR, and contractual responsibility, who signs the pilot agreements, and what does the data‑protection impact assessment say?

“Job displacement is something which is going to happen…and no government and no society is prepared for the speed with which it will happen.”

Mitsotakis urged humility. If frontier labs, like OpenAI and Anthropic, say they do not fully understand emergent behaviors, regulators should listen. That does not mean outsourcing policy to builders. Regulators need technical briefings plus independent audits, academic review and clear incident‑reporting channels.

Where the record is thin, what to verify next

DAEDALUS technical and governance specs: peak FLOPS, node types (CPU vs GPU), interconnect, storage, energy use, who allocates compute time, pricing for startups vs academia, and installation timeline.
OpenAI pilot details: number of teachers and students, selection methodology, evaluation metrics, contractual lead (Ministry of Education vs implementing NGO), and the pilot’s GDPR/data‑protection impact assessment.
Data‑center projects: official Microsoft/AWS statements, the identity of the utility partner, environmental impact assessments, and community consultation records.
Macro claims: MSCI reclassification timetable and bond‑yield comparisons require date‑stamped market sources (MSCI, Bloomberg/Reuters, Bank of Greece) because yields and index decisions change over time.
Legal policy changes (tax, labor, visas, and the reported social‑media ban for under‑15s): request the raw legislative texts or ministry press releases to confirm effective dates and eligibility criteria.

What this means for founders and business leaders (practical, immediate asks)

If you’re weighing Greece as a hub or expansion target, it’s worth serious consideration, but only with documents in hand. Here are four concrete, non‑negotiable asks to make of the relevant ministries, HPE or OpenAI before you commit:

Compute specs and SLAs: Request DAEDALUS’s detailed technical brief (GPU/CPU counts, interconnect, storage, peak TFLOPS/EFLOPS) and a service catalogue that spells out access tiers, queueing policy and pricing for startups versus research teams.
Training vs inference policy: Clarify whether DAEDALUS will permit large‑scale model training or is optimised for scientific HPC and inference, and ask for example workflows and expected monthly GPU‑hour quotas.
Data‑protection & pilot governance: For any school or public‑sector pilot, demand the pilot’s data‑protection impact assessment, parental‑consent templates, retention policies and a named data‑controller responsible under GDPR.
Environmental and community assurances: For data‑center projects, request environmental impact assessments, power‑purchase agreements (renewable sourcing), water‑use plans and records of local consultation.

These are the concrete items that separate a well‑intentioned announcement from operational reality.

Balancing enthusiasm with realism

Mitsotakis’ candor stands out. He mixed promotion, visits to Tesla and Sequoia Capital, and a self‑described “homecoming” to Stanford with the admission that governments don’t have all the answers. He said Greece is “paying down its debt at a record pace” and that it borrows more cheaply than the United States. Those are political and market assertions that need date‑stamped verification from MSCI and financial‑market sources before they are treated as settled facts.

His willingness to convene technologists with “the social scientists, the philosophers, the historians” is smart: complex, systemic risks need multidisciplinary review. Convening is only step one. To be credible, Greece, like any country hoping to lead, will need transparent timelines, independent audits, published evaluation results and clear legal texts for the incentives it markets.

Short, honest key questions and answers

Is DAEDALUS built and funded?

Yes. The Ministry of Digital Governance reports DAEDALUS has been assembled and entered final testing with Hewlett Packard Enterprise. Phase A funding is €29, 200, 000 (incl. VAT) from Recovery & Resilience funds; Phase B has up to €12, 679, 900 (excl. VAT) from EuroHPC, per ministry reporting. Detailed hardware and access specs have not been publicly released yet.
What will the OpenAI pilot actually do in schools?

Per OpenAI, the pilot will deploy ChatGPT Edu in upper‑secondary schools this academic year to reduce teachers’ administrative work and test personalized tutoring; the Onassis Foundation is managing implementation and the program includes accelerator support and credits for startups. The pilot’s evaluation metrics and data‑protection documentation should be published publicly.
Has Greece solved the data‑center debate?

Mitsotakis said there has been no significant public reaction so far. That claim needs corroboration with local reporting and the environmental assessments and community consultations for Microsoft/AWS projects before declaring the debate settled.
Are the government’s macroeconomic claims verified?

Those were presented by the prime minister; independent confirmation (MSCI for reclassification, Bloomberg/Reuters or Bank of Greece for bond yields and debt metrics) is required to treat them as established facts.
What should founders demand right now?

Ask for DAEDALUS’s technical brief and SLA, the OpenAI pilot’s evaluation and GDPR terms, full texts of visa/tax incentive legislation, and environmental impact statements for any data‑center project.

Final thoughts for executives and policy watchers

Greece is moving from rhetoric to infrastructure: HPC onshore, a high‑profile AI partnership, and incentives aimed at reversing brain drain. Those are meaningful signals for investors and founders. But the gap between announcements and operational detail matters. Technical specs, governance rules, evaluation data and legal texts will determine whether these initiatives create enduring value or fade as missed promises.

The prime minister’s admission that societies are “not prepared for the speed” of displacement is both a political vulnerability and a pragmatic opening. If Greece pairs its investments with transparent reporting, independent review and concrete social‑policy plans, retraining budgets, published pilot outcomes and data‑protection safeguards, it can be more than a promotional stop on a Silicon Valley tour. It can become a testbed for pragmatic, accountable AI adoption. Founders and investors should watch closely and press for documentation before they commit.