In partnership with
It's time to set up your holiday streaming campaign
Are you ready to reach holiday shoppers on CTV? For many streamers, the season is already underway.
Last year, according to Roku internal data, holiday content searches on Roku saw an 86% increase during October. And in November, those searches rise an average of 39% weekly. This surging holiday spirit is mirrored by consumer activity, as about half of Roku streamers shopped Black Friday and Cyber Monday deals in 2025.
Meeting those shoppers is easier than you might think. Begin with the seasonal outcome you hope to drive: awareness, online sales, leads, store visits, or another measurable goal. From there, define your audience, upload your creative, set the budget and campaign dates, and establish measurement before launching. You can even create a shoppable ad unit that lets streamers buy with a click of their remote.
Our holiday playbook offers a month-by-month guide to reaching shoppers during this all-important time of year.
First-time spenders get a $5,000 match on the first $5,000 spent with code BEEHIIV5K. Terms apply.
iPrompt Signals DEEP DIVE
COMPANION TO ISSUE 36 · 6 NOVEMBER 2026
Who Owns the Megawatts
A permit pause doesn't switch off a single server. It decides whose pipeline waits, and who gets paid meanwhile.
The directive from Austin ran to a few short paragraphs. On 21 September the governor of Texas told state agencies to halt data-centre permits until the grid operator had audited every data centre in its connection queue. No end date. A compliance update due 19 October. The governor was a Republican, and Texas accounts for about a quarter of US data-centre construction. On Tuesday, 41 local ballot questions on data centres went to voters, and in 40 the outcome favoured limits. His Democratic successor campaigned for something stronger.
Notice what that order does and doesn't touch. Not one server already running goes dark. What stops is the queue.
So the useful question isn't how much AI America will build. It's whose pipeline is in the queue, how long each can afford to wait, and who collects rent meanwhile. This piece separates what's actually restricted from what's proposed, explains why exposure matters more than business model, runs a low, middle and high case on what a year's wait costs one builder, and ends with a four-step check for any holding.
1. WHAT'S RESTRICTED, WHAT'S LOCAL, WHAT'S ONLY PROPOSED
Only two statewide restrictions bind today. Everything else is local, failed, or political. Both binding measures pause new approvals; neither touches capacity already energised.
Where | Measure | Status | What it touches |
Texas | State agencies halt data-centre permits until ERCOT audits every data centre in its interconnection queue | Enacted: governor's directive, 21 Sep 2026; no end date | Projects waiting to connect |
New York | Executive Order 62: freeze on new data centres of 50MW or more | Enacted: executive order, 14 Jul 2026; 'up to one year' | New large builds |
Georgia, state | HB 1012, a statewide moratorium bill | Failed: the 2026 session adjourned without passing it | Nothing statewide |
Georgia, local | Moratoria in more than a dozen counties and cities | Local orders | Sites in those areas |
41 local measures | 20 restrictions or moratoria passed; 6 specific developments rejected; advisory and conditions votes mixed | Voted, 3 Nov: 40 of 41 outcomes favoured limits | Particular towns and sites |
Federal | Calls for a national moratorium | Proposal only; would need a presidential signature | Nothing yet |
Sources: Office of the Texas Governor · Jones Day (NY EO 62) · Inside Climate News (Georgia bills fail) · Axios Atlanta (local moratoria) · Ballotpedia · CNBC (moratorium calls)
That gap matters. Goldman's estimate that moratoria could affect about 45% of planned US capacity expansion is a forecast of what could follow Democratic wins at state and local level, not a measure of what's in force. Texas's governor-elect has promised a moratorium, but the state legislature stays Republican, which limits what she can do alone.
Sources: Morningstar (Goldman) · CNBC Investing Club (Bernstein) · CNBC (Texas governor-elect) · Quartz
🌱 NEW TO INVESTING? HERE'S WHAT THIS MEANS An interconnection queue is the waiting list to plug into the grid. A finished building full of servers earns nothing until the utility connects it. Texas has frozen the paperwork for everyone on that list. The ones already connected are off it, and that's suddenly worth something. |
2. BUSINESS MODEL IS SHORTHAND. EXPOSURE IS THE TEST.
It's tempting to sort the trade into three seats and stop there.
Seat | Who | What a pause does | Number to watch |
Owners | Microsoft, Amazon, Alphabet; landlords such as Equinix and Digital Realty | Running space gets scarcer; new building can slow | Bookings, and direct pricing data |
Builders | CoreWeave, Nebius, IREN; developers in the queue | Revenue waits while spending on unconnected sites sits idle | Active against contracted power, and where the gap sits |
Buyers | AI labs such as Anthropic | Pay more for whatever capacity is free now | Compute cost per dollar of revenue |
But the seats describe business models, not exposure. Microsoft builds; CoreWeave already runs about 1.5GW and has more than 1GW contracted outside the US, including 360MW in Indonesia due online in about 18 months. What decides the damage is how much of a company's not-yet-running pipeline is unpermitted, unfunded, or inside a restricted jurisdiction. A big pipeline that's permitted, funded and in Utah is a different animal from a small one stuck in a Texas queue.
Nobody reports that number cleanly. Until they do, the seats are a starting hypothesis, and the most useful thing an investor can do is ask for the location of the gap.
3. WORKED EXAMPLE: WHAT A YEAR'S WAIT COULD COST A BUILDER
CoreWeave is the cleanest public case, because it discloses both active and contracted power. Only the guidance lines in the first table are disclosed. Everything in the second table is an assumption, so each gets a low, middle and high value. The point isn't the answer. It's seeing which assumption moves it.
Disclosed (company guidance) | Figure |
Contracted power, including what's already active | 4.2GW (11 Aug) |
Active power, year-end 2026 target | more than 1.85GW |
Active power added during 2026 | about 1GW (850MW at end-2025) |
Capex, 2026 | $35bn–$39bn |
Q3 interest expense guide | $860m–$940m (about $3.6bn a year) |
Q3 adjusted operating income guide | $200m–$260m (about $0.9bn a year) |
Assumption or result | Low | Middle | High | Basis |
Capacity delayed a year | 0.15GW | 0.35GW | 0.6GW | 15%, 35% or 60% of a year's ~1GW of additions |
Revenue per running GW, per year | $6bn | $8bn | $10bn | Blended estimate: exit run rate ÷ year-end active power; not a figure for any single deployment |
Capital spent before power arrives, per GW | $8bn | $15bn | $25bn | Shells and power gear only, up to chips already delivered |
Cost of capital | 8% | 10% | 12% | Blended financing rate; unsecured notes priced at 9.625%–9.75% this year as a proxy |
Revenue pushed back a year | $0.9bn | $2.8bn | $6.0bn | Delayed GW × revenue per GW |
Capital sitting idle | $1.2bn | $5.3bn | $15.0bn | Delayed GW × capital spent before power |
Carrying cost, per year | $0.1bn | $0.5bn | $1.8bn | Idle capital × cost of capital (the definition used throughout) |
Carry ÷ annualised interest guide | 3% | 15% | 50% | Scenario comparison, not a forecast; see below |
Carry ÷ annualised adjusted operating income | 10% | 57% | 196% | Scenario comparison, not a forecast |
What moves the answer. Two assumptions do most of the work: how much capacity slips (a fourfold range across the cases) and how much was spent before the power arrives (threefold, highest if chips are already sitting in a warehouse). With the rate, they move the carrying cost almost twentyfold. Neither is disclosed, so those are the two questions to put to management.
Carry isn't the same as new interest. Carrying cost is what the idle capital should be earning. It turns into extra cash interest if the delay forces new borrowing (to keep paying suppliers while revenue waits, say), postpones repayment of existing debt, or extends its term. Some interest on assets under construction may already sit inside the $860m–$940m guide, or be capitalised rather than expensed, so the percentage rows may overstate the increment. They compare a hypothetical carrying cost with annualised guidance; they aren't forecasts, and they aren't bounds.
Deferred isn't guaranteed. CoreWeave counts backlog 'subject to the satisfaction of delivery and availability of service requirements'. A long enough delay could give a customer room to renegotiate or walk, on terms outside investors can't see.
Sources: CoreWeave Q2 release
WHAT THE EXAMPLE SHOWS, AND WHAT IT DOESN'T It shows why a builder's shares can drop 10% on news that cancels no contracts: delay can turn a timing problem into a financing problem. The middle case equals about 57% of annualised adjusted operating income; the high case, about twice it. Whether either is manageable depends on liquidity, cash generation, debt maturities and access to new financing, which this model doesn't test. Backlog isn't cash: it's future revenue, conditional on delivery. It doesn't show CoreWeave's actual exposure. It doesn't say how much of its next gigawatt sits in Texas, New York or a town that just voted, or how much has already been spent on it. Those two numbers decide which column is real. Tuesday's results are the first chance to ask. |
4. WHAT THE OWNERS GAIN, AND WHY IT'S HARDER TO SEE
The builder's cost shows up in lines you can read. The owner's gain is quieter. It turns up in pricing, and pricing moves slowly.
Equinix is the clearest tell, though not yet proof. Its third-quarter annualised gross bookings hit a record $533m, up 35%, which the company put down to strong demand and strengthening pricing, and it raised guidance for a third straight quarter. All of that came before the vote. Bookings combine contract volume and price, so they can't isolate rent increases on their own. The signal to look for is the pair: bookings holding up and management putting figures on pricing. Morningstar's caveat is fair, too: new capacity remains capital-heavy for Equinix as well.
For the hyperscalers the gain is partly what they don't spend. Axios framed Wednesday's rally that way: delayed commitments mean healthier cash flow and lower financing costs. Evercore's Krishna Guha argued a slower build could 'plausibly extend the AI investment cycle'. An owner whose own pipeline isn't stuck gets paid twice, once in rent and once in capex it no longer has to rush. If its pipeline is stuck too, it's a builder with a better balance sheet.
Sources: Equinix Q3 release (SEC) · Morningstar EQIX · Axios
5. THE BUYER'S SQUEEZE, IN ONE FILING
Anthropic's amended prospectus shows the buyer's seat from the inside. Its third-quarter operating loss of $567m had two main causes: about $5bn of compute delivered under short-term commitments that, in the filing's words, currently costs significantly more than comparable multi-year compute, and a $2bn non-cash donation-matching charge. The adjusted line shows compute pressure more cleanly. Adjusted operating income still rose, to about $2.5bn from $2.2bn, but the adjusted margin fell from 19% to 15%, which the filing mainly puts down to short-term compute.
Anthropic, 30 Sep 2026 | Figure |
Q3 revenue | $16.96bn, up 47% on Q2 |
Compute bought on short-term terms in Q3 | about $5bn |
Purchase and lease commitments | about $895bn, substantially all compute and infrastructure |
Of which subject to termination rights or conditions before they start | about $213.8bn |
Due within 12 months | about $69.8bn, of which about $4.3bn carries those rights |
Cash and short-term investments | $131.8bn |
Two readings follow. First, the buyer is already short of capacity, before most of the gate has closed. Second, the termination rights cut both ways. They give Anthropic room to step back from capacity that arrives late or not at all, and they make part of its suppliers' backlogs less firm than the headline suggests. When the price range lands, the roadshow question isn't only valuation. It's how much of the money raised goes straight to whoever owns the megawatts.
Sources: FinanceFeeds (roadshow)
6. CHIPS: DELAYED, OR LOST?
Thursday treated chipmakers like builders. I'm not convinced that's right, but I can't rule it out. AMD guided fourth-quarter revenue to about $17bn, up 66%. Advantest says customer demand will exceed supply into 2027. Those are statements about demand, and demand didn't change this week.
Volume is different. A GPU without a connected building is inventory. If even part of Goldman's 45% scenario happens, US deliveries get pushed out and some builds move abroad or to states such as Utah and Iowa, where developers are already looking. Both take time. My read is timing, not lost demand. Nvidia on 17 November is the first test of whether customers have started saying the same.
Sources: AMD release · Advantest Q2 note · Morningstar (Goldman) · FourWeekMBA (developers move) · Gotrade (Nvidia date)
7. WHAT TO WATCH
When | Checkpoint | What would change the picture |
10 Nov · after close | CoreWeave Q3 | Where the not-yet-active gap sits; any cut to 1.85GW, and whether management blames permits |
Week of 9 Nov (reported) | Anthropic price range | Proceeds against the ~$70bn of commitments due within a year |
17 Nov · after close | Nvidia Q3 | Any US customer deployments named as delayed |
No fixed date | ERCOT audit | Completion lifts the Texas halt, unless the incoming governor extends it |
12 Jan 2027 | Texas legislature convenes | Any statute on data-centre permits or who pays for grid upgrades |
Mid-Feb 2027 | Equinix Q4 | Bookings against $474m a year earlier, with figures on pricing |
By Jul 2027 | New York review | Standards that end EO 62's freeze, or harden it |
Sources: INDmoney (Q2 call guidance) · Anthropic S-1/A (SEC) · Gotrade (Nvidia date) · Office of the Texas Governor · Equinix Q4 2025 release · Jones Day (NY EO 62)
8. A FOUR-STEP EXPOSURE CHECK FOR ANY AI HOLDING
1 Find the two numbers. Active against contracted power. For a landlord, leased against under construction. For a chipmaker, how much of the order book is tied to named US sites.
2 Locate the gap. Which states and towns is the not-yet-running capacity in, and is it permitted and funded? Texas and New York are the binding ones today; local orders matter site by site.
3 Price the wait. Capital already spent on that capacity, times the company's cost of capital, gives a rough yearly carrying cost. Then ask whether the delay forces new borrowing or holds up repayment of existing debt; that's what hits cash.
4 Read what the customer owes, and when. Revenue recognition and payment are different things. Look for when payment obligations start (signing, delivery or acceptance), any acceptance milestones, and the cancellation or termination terms, usually in the commitments and revenue notes.
Run that on one holding and you'll know which seat it really sits in. Honestly, the vote didn't settle whether AI is a bubble. It changed who pays while the question stays open.
Stay curious — and stay qualified.
Disclaimer: This article is for informational and educational purposes only and does not constitute financial advice. iPrompt Signals is not a registered investment advisor. The worked example uses labelled assumptions and is illustrative, not a forecast. Always conduct your own research and consult a qualified financial professional before making investment decisions.