The author is a Reuters Breakingviews columnist. The opinions expressed are her own.

By Karen Kwok

PARIS, July 10 (Reuters Breakingviews) - The Sun King would have recognised France’s latest courtship ritual. Emmanuel Macron may not be Louis XIV, but the French president is nonetheless trying to pull the aristocracy of AI into his nation's orbit.

More than 8,000 AI founders, chipmakers and financiers this week descended on Paris for the RAISE Summit. The surroundings encouraged grand ambitions. Before the main event opened at the Carrousel du Louvre, guests in black tie wandered through the Palace of Versailles' Hall of Mirrors and dined beneath the chandeliers. Alumni of France’s elite engineering schools were everywhere. So, too, were freely poured glasses of champagne and conversations continuing late into the gardens under the fireworks.

The talk itself was more austere. Rather than simply celebrating ever-larger models and computing clusters, executives' focus centered on making AI cheaper. Chip designers and data-centre operators promised to squeeze more work from advances in silicon and construction design efficiencies. AI startups explained how companies could avoid wasting tokens on repetitive tasks.

The debate even took in another key AI theme - the power supply. A temporary outage each day disrupted the conference during a Paris heatwave, cutting microphones and connectivity. At one point, French AI darling Mistral boss Arthur Mensch had to continue without amplification. The symbolism was hard to miss: an industry debating whether it consumes too much energy briefly struggling to keep its own lights on.

Cost is becoming AI’s awkward reality. The United States' biggest technology companies like Meta Platforms META.O, OpenAI and Anthropic can keep throwing capital at models, chips and data centres. Most businesses and countries cannot. Companies may see encouraging results from AI experiments, but deploying the technology across an organisation requires accessible data, retrained workers and rebuilt software. The spending often arrives well before the revenue or productivity gains.

France offers one possible response. It cannot match Silicon Valley’s private capital, but its tradition of "grands projets" favours infrastructure, state coordination and national champions. Its "grandes écoles" produce highly trained engineers. Adding together the major announcements since early 2025, France has attracted roughly €180 billion of prospective AI and data-centre investment based on the more concrete initial phases. SoftBank Group’s 9984.T proposed French buildout starts at €45 billion, while Brookfield BN.TO and Nebius NBIS.O have committed some €38 billion. France's government fund BpiFrance has promised to invest €10 billion in the next five years in AI.

The sums are less solid than the palace walls. They span several years, include servers and power infrastructure, and may rely on debt, partners and future customer demand. Still, France’s nuclear fleet gives it something increasingly valuable: access to large amounts of relatively low-carbon, nuclear electricity with which to power AI and data centres.

That explains the prominence at the Paris gathering of AI’s less glamorous plumbers. Semiconductor challengers, networking specialists and “neoclouds” - specialist cloud providers that acquire access to high-end AI chips - occupied much of the conversation. South Korea’s Rebellions as well as U.S.-based d-Matrix and SambaNova, are redesigning processors around inference - the process of running a trained AI model to generate outputs for real-world use — to generate more tokens with less power consumption. Fellow American Cerebras Systems CBRS.O argues that its giant wafer-scale systems can run models more efficiently than conventional graphics processors.

A second group is helping companies shop around. Nvidia-backed Baseten and Fireworks AI offer platforms that combine models, customise them for particular tasks and route workloads among different cloud providers. The goal is to avoid becoming captive to one expensive supplier. Increasingly capable "open-weight" models, which make public their core mathematical parameters and include those developed in China like MiniMax, are part of the mix.

The shift is not only about price. Washington’s temporary restriction on foreign access to Anthropic’s advanced Fable and Mythos models reminded executives that dependence on a single American provider can carry political risk. The controls subsequently disappeared, but the scare strengthened the argument among Paris attendees for retaining more control over models, data and infrastructure - even if it initially costs more. Executives also argued that most corporate tasks do not require the most expensive and smartest model available.

The harder question is who captures the value. Roughly 80% of attendees at the Paris gathering were non-French. Foreign investors may build the data centres, Nvidia NVDA.O and other overseas chipmakers may supply the processors, and American cloud groups and model developers may still collect much of the revenue. France could end up contributing the land, electricity and hospitality, while others own the intelligence running inside.

As national AI playbooks go, this Gallic combination of state support, infrastructure and political theatre still positions the country as one of the more AI-facing European players. The test is whether today’s cloud barons leave behind French companies and durable profits - or just server farms and empty champagne bottles.

Follow Karen Kwok on LinkedIn and X.

CONTEXT NEWS

RAISE Summit at the Louvre, Paris, took place from July 8 to July 9.