What’s changing in frontier tech – from geopolitics to AI and energy

The world looks different from how it did in January.

The geopolitical backdrop has hardened. Governments are thinking more explicitly about technological dependence; while AI, space and other frontier technologies are becoming more tightly bound to infrastructure and security.

In this edition of the Frontier Technologies & Innovation wrapper – drawing on conversations among big tech, government and civil society at the Annual Meeting in Davos this January – we explore what that shift means.

Also, scroll down for a conversation with Varun Sivaram, founder and CEO of Emerald AI, a start-up that turns data centres into flexible support for the power grid.

For the past two years, the dominant frame for frontier technology has been competition. The focus was on who has the most capable model, the largest valuation or the fastest iteration cycle. But a more consequential contest is now coming into view – one centred on the energy, infrastructure, supply chains, standards and strategic relationships needed to deploy these technologies at scale.

At Davos in January, for example, the near-term discussion on quantum focused not just on when useful computing arrives, but also on who sets post-quantum security standards and who is prepared for the transition. In space, governments are putting greater weight on sovereign launch and connectivity capacity as concerns over concentration and strategic dependence grow. In cybersecurity, more powerful tools are meeting a more fragmented geopolitical environment, leaving highly interconnected systems exposed to uneven levels of risk.

For AI, the shift is most visible in the infrastructure debate. The question is no longer simply who has the most advanced model but who can power it, scale it, govern it and secure it. The burden is also landing in places the innovation conversation rarely reaches: in rising household electricity bills near new data centres, in freshwater demands in already water-stressed regions, and in grid constraints that are starting to shape national competitiveness.

Across these domains, the strategic lesson is similar. In an interdependent system, the goal is not self-sufficiency in everything, but becoming indispensable in something.

Some of the infrastructure constraints described above are also creating openings for a new generation of companies. Emerald AI, founded in 2024 and backed by NVIDIA’s venture arm - plus investors including Google chief scientist, Jeff Dean - is one of them. It builds software designed to make AI data centres more flexible power users; turning them, in effect, into grid-responsive assets rather than fixed sources of demand.

We spoke with Varun Sivaram, Emerald’s founder and CEO and a senior fellow at the Council on Foreign Relations, about what that looks like in practice. As he put it, the company’s central thesis is simple: if energy is becoming one of AI’s biggest bottlenecks, AI itself may help relieve part of that pressure by making demand more flexible.

➡️ The pain: AI is becoming an infrastructure problem

The International Energy Agency expects global electricity demand from data centres to more than double by 2030. In the United States, meanwhile, the transmission system is already becoming a major constraint on new data-centre buildout, with long waits to connect in some regions.

That is the environment in which Emerald is operating. Sivaram argues that the issue is not simply a shortage of generation, but the mismatch between when the grid is most stressed and how inflexibly large data centres are normally expected to behave. “Power and energy is the single biggest bottleneck to the AI revolution in the US and in the West,” he told delegates assembled in Davos.

The company’s aim? That some of that bottleneck can be eased if AI workloads become more responsive to grid conditions, rather than running at maximum intensity regardless of what is happening on the system.

➡️ The fix: making AI data centres power-flexible

Emerald’s core argument is that AI data centres are unusually well suited to flexibility. Some workloads can be deferred, some can be shifted geographically and some demand can be balanced with on-site resources such as batteries or generation. "AI data centres have a transformative potential to be flexible in their power consumption. They're remotely controllable and can respond within milliseconds," founder Varun Sivaram says.

Emerald’s software helps data centres use electricity more flexibly. Urgent tasks are handled immediately, while less time-sensitive work can be delayed or slowed down. Work can also be shifted between different locations depending on where power is most available. In addition, data centres can tap into batteries or their own energy sources when needed. Together, this helps reduce pressure on the power grid while keeping systems running smoothly.

The premise is not to interrupt AI use but to manage it more intelligently in the moments when the grid is under real pressure.

➡️ Road ahead: From pilot to proof point

Emerald says it has now completed four demonstrations at commercial data centres. The first was in Phoenix, Arizona, in collaboration with Oracle, NVIDIA, Salt River Project and EPRI. On a day of peak grid stress, the company says it reduced AI power consumption by 25% for three hours, a result Sivaram presents as evidence that data centres can support grid reliability without creating visible disruption for users.

The company is now trying to move from demonstration to commercial scale. Sivaram said Emerald plans to announce Aurora, a power-flexible AI facility in Manassas, Virginia, with partners including NVIDIA, Digital Realty, PJM and EPRI. The project has been framed as an early test of whether grid-responsive computing can move beyond pilot mode and become part of the design of future AI factories.

"I want to take AI from being the villain that communities don't want anywhere nearby because it raises your power prices, to being a hero where every community, every utility, wants to bring AI data centres to their grids."

China's robotics sector moves to capital markets: Unitree Robotics has filed for an IPO on Shanghai’s STAR Market, seeking to raise 4.2 billion yuan. The company’s revenue rose 335% in 2025, with humanoid robots accounting for 51.5% of revenue in the first nine months of the year. The significance goes beyond one company: a successful listing would create an early public-market benchmark for the humanoid robotics sector and offer a concrete signal of where China’s industrial policy is placing its bets.

Europe's satellite ambitions come into focus: Germany’s proposed €10 billion military satellite network is raising concerns in Brussels about duplication with the EU’s IRIS² secure-connectivity programme. The deeper issue is not only institutional overlap but a growing tension between national urgency and collective European planning as space is increasingly treated as strategic infrastructure.

Post-quantum security becomes a near-term policy priority: China is expected to have national post-quantum cryptography standards within roughly three years, Reuters reported, with finance and energy seen as early priority sectors. That follows the United States finalising its first post-quantum standards in 2024.