Dear readers,

This Wednesday, we move further away from code and toward control over data itself. A growing argument from Greek academic Nikos Panagiotou frames AI not just as a technological race, but as a geopolitical one shaped by what he calls algorithmic diplomacy: a contest over which datasets shape the systems that increasingly define how information is ranked, interpreted, and understood.

The implications are structural. If data is the new oil, then the pipelines are training corpora, compute infrastructure, and the invisible editorial choices embedded in machine learning systems. For smaller countries, this creates a difficult reality: influence is no longer only about building technology, but about participating in the systems that decide what becomes visible in the first place.

Meanwhile, capital keeps flowing across the region’s AI and infrastructure layer, from DeepInfra’s $107 million Series B to a wave of seed and early-stage bets in compliance APIs, hybrid work platforms, and agentic ecommerce systems. But beneath the momentum, layoffs in parts of the regional IT sector raise harder questions about outsourcing models under AI pressure.

Finally, this week’s Founder Take explores what actually slows startups down in an AI-first era: not speed, but focus, and the persistent pull of fundraising away from real users.

Enjoy this week’s read!
Bojan Stojkovski
Editor-in-Chief, IT Logs

In the 20th century, control over oil defined global power. In the 21st century, control over data increasingly plays that role. But a deeper and a more profound change is also about the systems that transform data into meaning: how it is ranked, filtered, translated, and stabilised into knowledge that people and machines can act on.

This is where a new concept begins to take shape: algorithmic diplomacy. Not diplomacy in the traditional sense of embassies and treaties, but a subtle form of statecraft focused on the infrastructure of AI. If algorithms decide what the world sees and reads, then the strategic question becomes who decides what feeds those algorithms?

The three pillars of AI power: compute, data, and talent

Algorithmic diplomacy, as he describes it, is the deliberate state-level management of AI training data to project influence, protect sovereignty, and shape global narratives. It is not censorship or propaganda in the classical sense. It operates earlier in the chain, at the level where reality is statistically shaped before it is ever expressed.

“If algorithms decide what the world sees and reads, then we must ask who decides what feeds the algorithm.” he warns.

That question itself reframes how power is used. Instead of focusing only on platforms or outputs, attention shifts to datasets, selection processes, and the invisible editorial decisions embedded in what is included or excluded from training corpora.

The imbalance, he argues, is not just financial but strategic. Without sufficient compute capacity or access to large-scale datasets, countries become dependent on external systems for interpretation and decision-making.

“If you lack computational power, if you lack resources and skills, you risk becoming a consumer of systems you cannot influence.” he says.

This dependency is not visible in traditional geopolitical terms, as it does not involve borders or military presence. However, it manifests through models that mediate search, summarisation, translation, and increasingly, reasoning itself.

One of his more striking claims is that global politics is beginning to fragment into competing data blocs, or ecosystems of information aligned with geopolitical and technological alliances.

This leads to a more subtle divergence. Different datasets produce different defaults. Different defaults produce different interpretations of the same event. Over time, this creates systems that do not merely disagree on facts, but on how facts are structured.

“In practice, we may not only disagree on facts, but on how we understand them.” Panagiotou points out.

The consequence is epistemic fragmentation: multiple coexisting but incompatible informational realities shaped by underlying training distributions.

Small states and structural asymmetry in AI systems

The implications for smaller countries are particularly severe. “What can small states do when they lack compute, data, and scale? It is very challenging. Very challenging.” he asks, repeating the phrase deliberately, underscoring its structural weight.

In this context, sovereignty becomes less about territory and more about participation in the systems that define visibility and relevance.

He also points to a broader institutional problem: weak international coordination mechanisms. Without stronger frameworks, asymmetries between large and small actors are likely to deepen rather than stabilise.

The boundaries between public and private data are increasingly blurred through partnerships and infrastructure sharing. He cites examples where sensitive datasets, including health and administrative systems, are integrated with private AI infrastructures, raising unresolved questions about ownership and control.

He also references ongoing legal disputes between media organisations and AI companies over training data usage. “This is not theoretical anymore. Content is being used to train systems that operate at global scale, often without clear consent frameworks.” he says

The result is a shift in journalism’s role: from gatekeeper of information to raw material for systems that reinterpret it.

Energy, infrastructure, and the return of physical constraints

Although much of the discussion focuses on information, Panagiotou repeatedly returns to physical infrastructure. “AI is not only software. It is electricity, infrastructure, and physical limitation.” he notes. 

The expansion of data centres introduces energy as a geopolitical constraint. The demand for compute is reshaping debates around energy policy, including renewed attention to nuclear power as a scalable low-carbon option. “This is not ideological, it is a question of capacity.” he says.

Infrastructure, once a background condition, becomes a central variable in technological competition. A more speculative but increasingly relevant concern is the rise of synthetic data in model training, he adds.

“If models begin training on synthetic outputs generated by other models,” he explains, “we risk entering a recursive loop.”

He summarises the risk in a single line that anchors the discussion: “At that point, the question is no longer what the world is, but what the model remembers the world to be.”

This is not framed as collapse but as drift - a gradual detachment between lived reality and statistically reconstructed reality.

When asked whether this dynamic leads inevitably to conflict, Panagiotou is measured. “At the moment, we are in a phase of strategic competition. Cooperation is not impossible, but it requires recognition of shared risk.” he says. 

He draws a historical parallel: “We may need something similar to the SALT treaties from the Cold War, where competing powers agreed to limit certain categories of development.”

He explains the analogy plainly: SALT worked because both sides understood mutual vulnerability. Whether similar logic can emerge in AI governance remains uncertain.

“At present,” he adds, “incentives favour acceleration.”

Autonomy, control, and system complexity

A final concern emerges around system autonomy, not in the sense of intent, but capability.

“These systems are becoming more autonomous in their behaviour. Not in intention, but in operational complexity.” he says. 

Panagiotou also returns repeatedly to a central idea: visibility is now constructed, not merely observed. “If algorithms decide what is visible, then the question is not only who builds them, but what world they are trained to see.”  he says,

He concludes with a definition of algorithmic diplomacy that captures the core of his argument:

“This is a negotiation not only between states, but between datasets that encode different versions of reality.”

Most dictation tools choke on technical language. Wispr Flow doesn't. It understands code syntax, framework names, and developer jargon — so you can dictate directly into your IDE and send without fixing.

Use it everywhere: Cursor, VS Code, Warp, Slack, Linear, Notion, your browser. Flow sits at the system level, so there's nothing to install per app. Tap and talk.

Developers use Flow to write documentation 4x faster, give coding agents richer context, and respond to Slack without breaking focus. 89% of messages go out with zero edits. Free on Mac, Windows, and iPhone.

Try Wispr Flow free

More tech rumors? Reach us at [email protected]

IT Logs:   What helps startups win faster: moving quickly or building something truly deep?

IT Logs:  Do you think AI-first startups will clearly beat those just adding AI on top?

IT Logs: What is slowing down founders the most: hiring, fundraising, or regulation?