Rogfast tunnel engineers build a 16.6-mile highway 1,280 feet under the North Sea

If you ever needed proof that scale still works, Norway’s Rogfast tunnel is it. The project is designed to become the world’s longest and deepest subsea road tunnel, and the numbers are the whole point: a 16.6-mile highway with a deepest point about 1,280 feet below the sea, under the North Sea. MIT Technology Review’s reporter goes to understand how engineers make something like that happen, including what it feels like to be roughly 1,000 feet beneath the North Sea in a dark, dank tunnel while imagining millions of tons of seawater pressing overhead.

Why this matters beyond engineering nerd joy: the tunnel is not just a flex. It’s a reminder that even when it can feel hard to “get anything done,” ambitious engineering can still ship. That’s the same mental model decision-makers need right now, because the world’s biggest tech build is the AI era, and AI is running into physical limits.

The newsletter immediately pivots from underwater infrastructure to above-ground constraints. It asks a question many operators are grappling with: if the AI boom is putting unprecedented pressure on the electric grid, why do so many companies rush toward adding new power plants? The alternative proposed in the piece is to look at flexibility inside the system that already exists. The core idea is that data centers could limit how much power they draw during high-demand stretches. If they can do that, they may not have to wait for large grid upgrades, and they may not need to build their own off-grid generation just to cover peak demand.

This is where the story gets operationally interesting. Flexibility is not entirely foreign to grid operators. In plain terms, power systems have always needed to manage peaks and troughs. But the newsletter points to “a new generation of software” that could make the process faster, smarter, and more precise for the AI era. The stakes are straightforward: AI workloads are elastic in some ways, but electricity demand spikes are real. If the grid is the bottleneck, then “how” data centers manage load is not a technical side quest. It can be a timeline differentiator for expansion, reliability, and cost.

The newsletter also frames the economic and social context that underlies these constraints. AI requires memory chips, and the rest of the tech stack feels the shock. In the “must-reads” section, it reports that SK Hynix has overtaken Samsung as South Korea’s most valuable company, and it’s now the world’s most valuable memory chipmaker. It also notes that AI’s need for memory chips is set to skyrocket device prices. Another quote of the day ties consumer outcomes to supply constraints: “Even before we start reaping the benefits of AI in our devices, we are already paying the bill,” Francisco Jeronimo, an analyst at IDC, tells CNBC, adding that consumers are covering the costs of the ongoing memory shortage. Translation: every part of the compute supply chain is trying to catch up, and “catch up” almost always means competing for constrained resources.

That’s why Rogfast is relevant even if you never plan to drive a tunnel. Engineering complexity is not one problem. It’s many problems at once: geology, pressure, materials, safety, timelines, and the sheer logistics of building something under the sea. The reporter is literally counting down from around 1,000 feet beneath the North Sea as a human reminder that infrastructure is physical, not theoretical. For boards and executives, the parallel is simple: the constraints are real, but so are the playbooks.

The rest of today’s brief underscores how quickly the market can punish delays and how fast incentives can move. Tesla’s Autopilot allegedly crashed into a Texas home and killed a woman, and the driver said the Tesla Model 3 was in self-driving mode. Tesla’s AI trainers “don’t trust its self-driving tech,” according to Reuters. SpaceX is mentioned receiving the lowest possible ESG rating from MSCI, a triple C, with Russia also receiving the same score after invading Ukraine. And there are stories about Polymarket reportedly paying creators to post fake betting videos and funding insider-trading fears related to an Iran deal. None of these are about tunneling or power management directly, but they all rhyme: systems fail when incentives and controls misalign, and when regulators, risk teams, and operators fall behind reality.

One more thing: the newsletter ends with mobile money supercharging sports betting addiction in Kenya. A lorry ride becomes a gambling moment because a smartphone pings with the outcome of a bet sent via mobile money. The piece describes a sharp rise and decline in fortunes for Bill Kirwa, where a 3,500 shilling bet (about $35) turns into nearly $8,500, then losses mount over time. The reason this belongs in an executive briefing is not because it’s “tech.” It’s because it shows how fast scaling distribution mechanisms can outpace safeguards and planning. That’s also the risk for AI compute and grid capacity. When demand grows faster than controls, the system gets stressed, and the costs show up somewhere, eventually.

So what should decision-makers take from all of this? Rogfast answers a confidence problem: complicated engineering can be delivered even under extreme constraints like those at the bottom of the North Sea. The grid and AI section answers the urgency problem: when demand spikes, waiting for new infrastructure may be slower than smarter flexibility. Put those together, and you get a clear strategic stake for anyone running infrastructure, managing capacity, or overseeing AI-driven demand growth. The winners will be the teams that treat physical limits as design inputs, not after-the-fact excuses.