- Analyzed Investing
- Posts
- AI’s Hungry Ghost: Power, Tariffs, and the Blackwell Bottleneck
AI’s Hungry Ghost: Power, Tariffs, and the Blackwell Bottleneck
Markets are worshipping at the AI altar. The real gods are the grid, trade policy, and time-to-rack.

Business news doesn’t have to be boring
Morning Brew makes business news way more enjoyable—and way easier to understand. The free newsletter breaks down the latest in business, tech, and finance with smart insights, bold takes, and a tone that actually makes you want to keep reading.
No jargon, no drawn-out analysis, no snooze-fests. Just the stuff you need to know, delivered with a little personality.
Over 4 million people start their day with Morning Brew, and once you try it, you’ll see why.
Plus, it takes just 15 seconds to subscribe—so why not give it a shot?
Intro
Another Monday, another round of RATE-CUT bingo on financial TV. Pass. The forces that actually move cashflows this quarter aren’t on the FOMC calendar—they’re buried in grid auctions, tariff chess, and the physics of getting silicon from a fab to a humming rack. AI is still the center ring, but the drama has shifted from model demos to megawatts, metal, and middleware. Watch where incentives, infrastructure, and import rules collide—that’s where multiples expand or evaporate.
Silicon Hype Meets Copper Wire
The AI trade is colliding with a simple constraint: electricity. Capacity markets across key regions are flashing red for 2026–2027 and beyond as data centers soak up incremental demand. Prices have leapt, and operators are projecting double-digit GW additions just to stand still. That’s a neon sign that power scarcity is now an earnings variable, not a think-piece.
States are responding with a new message to hyperscalers: if you want the juice, prepay the tab. Regulators are pushing big users to shoulder more of the interconnection and generation cost, insulating retail ratepayers and pushing project economics back onto the builders. Translation: CAPEX and timelines are drifting up and right.
Washington is trying to grease the skids with permitting tweaks for non-emitting generation and data-center builds. It won’t conjure transformers or gas turbines out of thin air, but it shortens one piece of the gantt chart.
Meanwhile, the ground truth is sprinting ahead of the headlines: multi-hundred-megawatt AI campuses keep landing in deregulated markets, where cheap land and loose caps lure site-selection teams. Deregulation doesn’t solve interconnection physics, but it does explain why so many site maps now pin the same states.
Blackwell: From Launch Slides to Loading Docks
Nvidia is still the scoreboard leader, but the “surprise me” phase is over. The latest beat came with thinner “wow,” and even the bulls admit the ramp is gating on supply chains and deployment cycles, not just demand. From tape-out to revenue, a modern AI system can take a year to hit the floor, which means CFOs are juggling bookings while facilities, power, and networking catch up. The cadence is normalizing.
Under the hood, Blackwell’s ramp may be the fastest in Nvidia’s history—and the most logistically complex. CoWoS capacity, HBM stacks, boards, optics, racks—every link has to show up on time. Earlier chatter had GB/B200 timing slipping across quarters before volumes inflected; later reports showed the ramp accelerating once parts started flowing. Net: the silicon is real, the backlog is realer, and the long pole is often everything around the chip.
And don’t ignore the power side of “time-to-rack.” Efficiency gains don’t lower demand if the workload explodes—Jevons Paradox in a server suit. Utilities are slow to expand because they’ve been burned by hype cycles before. That lag is now an investor variable, not an operational footnote.
Tariff Chess: EVs, Chips, and the New Industrial Policy
While AI eats compute, trade policy is busy eating margins. The U.S. slapped triple-digit tariffs on Chinese EVs and widened Section 301 actions that phase through 2026. That didn’t end the story; it started a live-fire test of how far industrial policy can tilt supply chains before demand routes around it.
Across the Atlantic, Brussels has leaned into provisional duties on Chinese EVs and is haggling toward “price undertakings” to defuse escalation. This isn’t just car wars; it’s a template for how the West will handle clean-tech overcapacity—by price floors, quotas, and paperwork that behave like tariffs without saying the word. Supply chains won’t snap; they’ll reroute, raise working capital, and turn lead times into a competitive moat.
Even beyond EVs, tariff détente and flare-ups are moving targets. The upshot for investors is simple: country-of-origin risk is now a core factor. Business models that depended on frictionless imports are being repriced. Those that can dual-source or near-shore earn scarcity premiums—especially when they can also secure megawatts.
Strategy Spotlight: Next-Gen Data Infrastructure
If the real constraint on AI isn’t ideas but infrastructure, then owning the picks-and-shovels beats guessing the next model demo. The Next-Gen Data Infrastructure thematic strategy is built for this moment. It targets companies across cloud compute, data storage, data-center operations, and high-bandwidth networking—the plumbing that converts GPUs and electricity into usable AI throughput.
Structurally, it’s long-term, rebalances every 30 days, and holds a diversified basket of roughly 20 public names with established track records in the space. Allocations can be equal-weighted or tilted by fundamentals, letting the portfolio lean into relative strength without turning into a single-name bet. In a world where grid capacity is scarce and interconnection queues are the governor on growth, this strategy sits where the cash register rings: selling capacity, bandwidth, and uptime. It isn’t a promise of smooth sailing—every cycle brings capex indigestion and supply-chain snags—but it aligns with this week’s core thesis: the edge isn’t the hype around AI; it’s the concrete, copper, and code that make it work.
The Big Picture
Power is the new P/E. Capacity pricing and interconnection queues are quietly setting AI equity winners and losers.
Time-to-rack rules the tape. Demand isn’t the bottleneck; deployment logistics are. Expect more “beat, lower-wow” quarters.
Policy is pricing. Permitting tweaks and state cost-shifts change build math more than any keynote slide.
Tariffs are structural, not a headline. EV duties and quasi-tariff “undertakings” are rewiring routes and working capital.
Avoid the decoy debates. The market’s fixation on rates misses where margins are made: grid access, lead times, and origin risk.
Educational content only; not investment, tax, or legal advice. Markets change—so should your priors.