🤚 The Open-Palm Illumination
In the latest Moonshots dispatch, Peter H. Diamandis and the panel sit down with Ramez Naam to discuss a refreshingly impolite constraint on the artificial intelligence banquet: electricity. Not vibes. Not regulatory theatre. Not another executive announcing that the future is “agentic” while a deck silently cries in the corner. Actual electrons, moving through actual wires, into actual data centers that increasingly resemble private cities for GPUs with better climate control than most apartments.
The episode’s central provocation is simple: the AI economy is no longer merely a software story. It is an energy story wearing a hoodie. The panel frames the grid as both bottleneck and opportunity, pointing toward underused transmission capacity, better grid software, sodium-ion batteries, fusion optimism, and even the exquisite possibility of wave-powered data centers. Because apparently the cloud has looked at land-based infrastructure and decided it would also like a yacht.
The sober outside evidence is not subtle. The International Energy Agency reported in 2025 that global electricity demand from data centers is projected to more than double by 2030, reaching around 945 terawatt-hours — slightly more than Japan’s entire current electricity consumption. The IEA also said AI will be the most significant driver of that growth, with electricity demand from AI-optimized data centers projected to more than quadruple by 2030. This is the kind of number that makes a facilities manager stare into the middle distance as if remembering a war.
👐 The Two-Handed Reality Check
The fashionable AI narrative says intelligence is scaling because models are improving, chips are multiplying, and venture capital has once again discovered the ancient ritual of confusing burn rate with destiny. But the less glamorous truth is that the new frontier runs through substations, transformers, permitting queues, interconnection studies, and transmission lines installed during an era when “the cloud” meant weather.
Data centers are unusually demanding customers. They do not ask for electricity the way a polite office park does, sipping modestly from the grid between conference calls. They arrive with megawatt appetites, reliability requirements, backup systems, cooling loads, and expansion plans written in the dialect of empire. In the United States, the IEA says data centers are on course to account for almost half of electricity-demand growth between now and 2030. That is not a sidebar. That is the maître d’ quietly informing the kitchen that the GPU table has ordered everything.
This is where Naam’s energy framing becomes useful. The problem is not only generation. It is also delivery. A grid can have theoretical capacity stranded by congestion, slow interconnection processes, outdated operational assumptions, or limited visibility. Grid-enhancing technologies — such as dynamic line ratings, topology optimization, and advanced power-flow control — aim to move more electricity through existing infrastructure without waiting a decade for every new line to receive its ceremonial paperwork tiara.
That does not mean “free power hidden in the couch cushions.” It means the grid is often operated conservatively because reliability matters and blackouts are generally considered poor branding. Unlocking capacity requires hardware, software, regulatory permission, utility coordination, and the patience of a saint employed by a public service commission. Still, the point stands: AI’s energy bottleneck may be softened not only by building more generation, but by making the existing grid less aristocratically inefficient.
🌿 The Gentle Awakening
The sodium-battery discussion matters because lithium-ion batteries, while excellent, are not the only courtier at the storage ball. Sodium-ion batteries use more abundant materials than lithium systems and are being explored for grid storage and lower-cost applications where weight is less sacred than economics. They are not magic. They generally trail lithium-ion on energy density. But for stationary storage, where the battery does not need to impress anyone by fitting into a phone, cost and supply-chain resilience can matter more than elegance.
This is the quiet energy lesson behind the AI spectacle: the future may be less about one miraculous technology and more about stacking many unglamorous improvements until civilization accidentally becomes competent. Better batteries. More flexible demand. More transmission. Smarter siting. Faster permitting. Better cooling. Cleaner generation. Software that understands the grid as a living system rather than a museum exhibit with invoices.
The wave-powered data center idea is the most cinematic, and therefore must be handled with tongs. Marine energy is real; the U.S. Department of Energy describes marine energy as power captured from waves, tides, currents, and thermal gradients. But commercial wave energy remains difficult, with harsh ocean conditions, maintenance challenges, and economics that have repeatedly slapped prototypes with briny realism. Floating or coastal compute powered by the sea is an intriguing concept, not yet a universal escape hatch from terrestrial grid constraints.
Still, there is a broader point beneath the sea foam. AI companies are beginning to think like energy companies because they have no choice. When compute becomes strategic infrastructure, electricity becomes strategy. The winners will not merely buy better chips; they will secure power, cooling, land, interconnection, and regulatory goodwill. The model weights may be digital, but the moat is increasingly made of copper, concrete, water rights, and someone’s ability to get a transformer delivered before the heat death of procurement.
👑 The Crown Verdict
The Moonshots episode is strongest when it refuses to treat AI as a purely abstract intelligence race. Yes, model capability matters. Yes, software orchestration matters. Yes, everyone will continue saying “agents” until the word files for workers’ compensation. But the deeper race is physical: who can turn energy into useful computation at scale, reliably, cheaply, and without convincing the local grid operator to develop a stress-induced twitch.
For investors, this means the AI trade is not just chips and model labs. It is power electronics, transmission, cooling, batteries, nuclear restarts, gas turbines, renewables, grid software, and the exquisitely underappreciated bureaucratic art of interconnection. For policymakers, it means AI industrial policy without energy policy is just a velvet binder full of wishes. For the public, it means the chatbot’s charming little answer about risotto may eventually be linked to a data center requesting the electrical dignity of a small nation.
The premium conclusion is not that AI will run out of power and collapse into a tasteful puddle. It is that the AI boom has graduated from laptop mythology into infrastructure reality. The next phase will reward whoever can make intelligence cheaper not only in tokens, but in watts. And if the ocean eventually hosts floating compute palaces powered by waves, we will applaud politely while checking whether the maintenance contract includes storms, salt, and Poseidon’s hourly consulting rate.
Until then, the grid remains the most important AI platform nobody invited onto the keynote stage.
Inspired by 200GW Hiding in Grid, Sodium Batteries 10x Cheaper, Wave-Powered Datacenters w/ Ramez Naam | EP #280 by Peter H. Diamandis.
Your transformer is showing. Interconnect wisely.