Uncategorized

BOSTON, MA (September 7, 2026): As per Recon Analytics’ annual iPhone pre-launch survey of the U.S. consumers, 62% of intended iPhone 18 base model buyers would upgrade to iPhone 18 Pro, Pro Max or Fold if Apple Skips the iPhone 18 base model during the fall event on September 9, 2026. The survey shows no base model launch would not cost Apple the sale volumes. In fact, it would help Apple upsell the “Pro” models lifting the consumer base to higher ASP models.

Among the respondents who planned to buy the iPhone 18, a follow-up question asked what they would do if that model is not launched in September 2026, 39.1% said they would buy the Pro instead and 14.1% would jump to the Pro Max, and 8.5% would take the Fold instead, making the combined 61.7% trading up rather than delaying their purchase. On the other hand, 23.9% would simply wait for the base model launch, and 9.3% would fall back to last year’s iPhone 17. Brand-defection to other Android brands’ risk is close to non-existent with 1.5% of this subgroup said they would switch to another brand.

“Apple’s move to launch ‘Pro and Fold’ models only during the Fall event will help Apple shift its model mix further to premium models lifting its ASP. In key markets like USA and China, the ‘Pro Max’ outsells ‘Pro’ model during the launch cycle, but in case of no base model, ‘Pro’ is likely to gain significant share of base model users. While iPhone Fold remains the center of attention, we expect the model to account for only a fraction of overall sales volume.”
— Hanish Bhatia, Vice President, Device Intelligence Group

Source: Recon Analytics’ Annual iPhone Pre-Launch Survey

When it comes to overall interest during the last three launch cycles, 20.2% respondents said they plan to buy an iPhone 18, up from 14.9% during the iPhone 17 launch cycle, but still down from 26.1% during the iPhone 16 launch cycle. While the overall interest is higher compared to last year, the sales volumes during the launch cycle remains a factor of availability and launch timings. We expect the sales volume to be lower this year as device prices and economic uncertainty increase and subsidies decrease leading to more consumers holding on to their existing devices. However, the shift in overall mix will lead to higher Pro and Pro Max sales.

When it comes to model mix, the premium shift that opened with the iPhone 17 series has held into the iPhone 18 with 68.5% of buyers choosing a Pro or Pro Max model this year, essentially unchanged from the iPhone 17 and well above the iPhone 16’s 47.7%. This highlights the continued consumer appetite for premium “pro” series iPhone models.

“While the consumer enthusiasm remains high for the iPhone 18 launch, US carriers are taking a more targeted approach when it comes to device subsidies. While the top line subsidy figures continue to increase, fewer high value consumers are eligible to take the full benefit of the carrier promotions. Therefore, we expect overall device sales to be marginally lower this year.”
— XJ Wang, Analyst and Director, Device Intelligence Group

Source: Recon Analytics Annual iPhone Pre-Launch Survey

Annual iPhone Pre-Launch Survey Summary:

Recon Analytics surveyed 8,897 U.S. consumers ahead of the iPhone 18 launch, 6,884 ahead of the iPhone 17, and 7,256 ahead of the iPhone 16, using an identical tracking question each cycle to enable a clean generation-over-generation comparison.

For paid version of the full report including consumer demographics and carrier preference, please reach out to [email protected]

About Recon Analytics:

Recon Analytics is an analyst firm serving the technology, telecommunications, AI, and Airline industries, combining large-scale proprietary consumer surveys with public and financial data to deliver independent, data-driven market intelligence. Recon Analytics’ Device Intelligence Group, covers global smartphone shipments and sell-through, supply chain analysis, bill-of-materials teardowns, component-level demand across smartphones, PCs, tablets, wearables, XR, as well as the consumer device purchase journey. Visit us at: Device Intelligence – Recon Analytics

Media Contact:

Hanish Bhatia, Vice President, [email protected]

XJ Wang, Analyst & Director, [email protected]

The fight over AI data centers is reported as a fight about fuel. Gas against solar, and nuclear against both. On 2026-08-07 Recon Analytics put six specific power arrangements to 6,255 US adults in the AI Pulse survey and had every respondent rate all six, so the comparisons sit inside the same person. The fuel turns out to matter less than one thing almost nobody is arguing about: whether the electricity touches the respondent’s own grid.

Take net support to mean the share who support an arrangement minus the share who oppose it. Dedicated solar and wind, built for the data center and kept off the local grid, is the only arrangement with real backing, at 49.8% support against 22.5% oppose, net +27.3. Put that same solar and wind on the respondent’s grid and net support falls to +7.9. Dedicated natural gas kept off the grid sits at +3.0; the identical gas added to the grid drops to -13.6. Same fuel, same people, one difference, and it is worth 19.4 points for renewables and 16.6 for gas. Americans may not use the phrase behind the meter, and they respond to it anyway.

Nuclear came last of the six, at 21.6% support against 48.7% oppose, net -27.1. It was fielded only in the on-grid form, so the honest comparison is against the other on-grid options: solar at +7.9, gas at -13.6, the existing grid as it stands at -22.1, and nuclear below all of them. A developer who pairs a data center announcement with a new reactor on the local grid is spending the plant’s goodwill, not borrowing it.

There is no free option in the list. Leaving the load on the grid as it is today, building nothing, was rejected at 22.7% support against 44.8% oppose. Americans will not absorb the demand quietly and will not welcome most of the ways of meeting it. One arrangement carries a mandate: new generation, dedicated to the facility, off the local system.

Set against that, the risk to AI usage itself is close to invisible. On the same wave, 1,486 of 6,255 respondents, 23.8%, said what they had heard about AI data centers made them use AI tools less or stop using them altogether. Read alone, that is a demand shock. It does not survive contact with the same respondents’ other answers.

Three checks. Only 3.0% of the 3,166 people asked report that their AI use fell over the past three months, against 3.1% across the prior six months. Of the 1,504 who report increasing their use over the same three months, 22.0% also say data center news made them use AI less. And environmental concern, named as a negative effect of AI, stood at 20.7% on the data center wave against 20.75% pooled across the nine waves before it. The concern is real, it has been stable for nine waves, and it has not changed how much AI people use.

The claim is mostly guilt, and it runs opposite to intuition. Among the 4,601 respondents who use AI at all, the lighter the usage, the more likely someone is to say they cut back: 30.9% of the lightest users against 16.8% of the heaviest. Guilt runs the other way, 6.7% of the lightest users against roughly 19% of the heaviest. Heavy users do not quit. They feel bad and keep going. That holds inside every age band and every income band we cut, so it tracks usage rather than demographics, though young Americans cut back more than older ones at every level of use.

The 489 who said they stopped altogether are a mixed group. Some genuinely stopped and can name the tool they left. But 43.1% of them also report that they never use AI, so for a large share the data center story is a reason given for something that was already true.

That split decides which lever an American reaches for. The people who say they cut back oppose a nuclear plant on their grid at 61.8% and support dedicated clean power at 47.8%. The people who feel guilty and keep using AI are the mirror image, at 49.6% and 65.9%. Light users want the demand reduced. Heavy users want the supply built. The two groups agree that something has to give and disagree about which end of the wire it comes from.

Which points the exposure away from the consumer entirely. Among respondents who approve AI purchases for their organization, 32.2% follow data center news very closely, against 11.5% of ordinary users. A guilty consumer cancels nothing. A buyer cancels a contract, and the buyer is the one paying attention.

For anyone siting capacity, the concession that buys public tolerance is dedicated generation, and it is worth roughly as much as switching fuels. Cleaner power on the same grid buys a fraction of what any power on a separate one does. For anyone selling AI, the exposure is a brand liability sitting with the heaviest and most valuable users, who carry it and still log in every day. Resourcing this as churn risk aims budget at a decline the data does not show. The number worth watching is not how many Americans quit. It is how many of them get a vote on what gets built.

Source: Recon Analytics AI Pulse survey, 6,255 US respondents, fielded 2026-08-07. Comparison figures pooled from waves fielded 2026-02-06 through 2026-07-31. The data center module was fielded once, so these are single-wave readings. Percentages are of the answered base stated at each figure.

Recon Analytics | reconanalytics.com

AI is repricing American work along a line that separates execution from judgment, and it runs through the middle of professions rather than between them. Execution work carries out someone else’s design: the standardized, codifiable tasks. Judgment work draws on knowledge and context to decide what to do when the rules run out. AI is driving the price of execution toward zero, and judgment, its complement, has never been worth more. We call that revaluation the judgment premium. Its cost side, the execution discount, is applied to the work young people learn on, and it is quietly dismantling the way into the most skilled careers in the country.

This overview summarizes the findings of our full report, which draws on the Recon Analytics AI Pulse Study, a weekly survey that has put its employment-impact question to more than 211,000 US adults since September 2025, alongside federal occupation data.

What AI Devalues, and What It Rewards

The official job counts draw the line clearly. In the year to May 2025, the occupations that shrank were the ones whose daily content is execution: customer service representatives lost 130,000 jobs, the largest decline of any American occupation, with data entry, bookkeeping, receptionists, and quality-assurance testing contracting alongside. The judgment occupations beside them, software developers, lawyers, and data scientists, all grew.

The rewards are just as measurable. In the professions adopting AI fastest, 18% of workers say AI skills have already earned them a raise, and nearly as many say AI skills helped them get a job or position. Those rewards concentrate among people with the expertise to direct the technology. We have watched the same dynamic inside our own firm: seasoned AI users with deeper context and domain experience routinely get better outcomes from the same tools, because knowing what to ask and what good output looks like is judgment, and the tools amplify it.

The Repricing Lands on the Young

Worry about AI is nearly uniform. Roughly three in ten workers fear AI-driven job loss, a number that barely moves by age, role, or income, and drifted only three points across the year we have measured. What actually happens to workers is anything but uniform.

Entry-level workers report AI-driven job disruption at up to two and a half times the rate of the most senior workers, and the gap holds inside every profession we can measure. A 23-year-old engineer faces just over twice the risk of a 55-year-old engineer; a junior legal worker faces two and a half times the risk of a senior one. Field choice cannot explain it. What remains is level, and the entry level is where the execution work lives, because apprenticeship has always meant giving the codifiable work to newcomers. Payroll-data research from Stanford economists finds the same pattern in an entirely different instrument: a 16% relative employment decline for workers aged 22 to 25 in the most AI-exposed occupations, while experienced workers in the same occupations held or grew.

The entry level is now the highest-variance position in the American labor market, because the repricing cuts both ways.

Source: Recon Analytics AI Pulse Study, September 2025 through August 2026. Respondents answering the employment-impact question, by age.

Among workers over 50, more than three quarters report no AI effect on their employment at all. Among workers aged 22 to 25, nearly half report an effect, split among disruption, AI-attributed raises, and AI skills helping them get a job or position, making them the most likely of any cohort to report an AI employment outcome in either direction. A new graduate in a technical field is simultaneously the worker AI is most likely to displace and among the workers AI is most likely to elevate. The difference is whether they end up selling execution or learn, fast, to sell judgment over an AI execution layer.

The Pipeline Problem

Judgment is acquired, and it has always been acquired the same way: years of execution work done under expert eyes. AI now competes with junior workers for exactly those tasks. The economy is still paying up for judgment while quietly cutting the production of it, and a profession that stops training juniors is drawing down an inventory that appears on no balance sheet. The full report sizes the affected entry-level population across the four fields where our sample supports fine-grained cuts.

Three audiences should read this differently. Workers: sell judgment, use AI as the execution layer beneath it, and start earlier than feels natural. Employers: treat the junior pipeline as infrastructure and protect it deliberately, or inherit a seniority shortage no salary budget can fix. Policymakers: track outcomes by career stage rather than sentiment, because for entry-level workers in skilled professions the disruption is already a year old.

About This Overview

This is the executive overview of The Judgment Premium (Recon Analytics, August 2026). The full report includes the complete occupation-level analysis, our sizing of the affected entry-level population, the two-leg reward decomposition and its measurement notes, income and employer-deployment cuts, the reconciliation of worker-level and occupation-level data, engagement with the public positions of Geoffrey Hinton, Jensen Huang, and Dario Amodei, four dated and falsifiable predictions we will score publicly, and full methodology. Findings draw on the Recon Analytics AI Pulse Study (211,207 employment-impact answers through August 5, 2026), BLS Occupational Employment and Wage Statistics (May 2025, released May 2026), Brynjolfsson, Chandar, and Chen, “Canaries in the Coal Mine” (Stanford Digital Economy Lab, 2025), and Garrett Touchet, “America’s AI Hope Deficit” (Recon Analytics, June 2026). The full report is available to Recon Analytics clients.

The wireless industry retired its customer metrics at the end of 2025, and this is the second quarter under the new schedule. T-Mobile publishes accounts and nothing below them, AT&T moved postpaid phone inside a segment built around convergence, Verizon took the Consumer and Business phone split off its tables, and Comcast and Charter publish mobile lines. No two of those units are comparable without rebuilding them from the financials up, and two quarters in, this is the regime rather than a transition anyone will reverse.

We are grateful for these changes. When the carriers take the numbers away and go dark, most of the market goes blind, and our capabilities and experience shine even more brightly. A retired metric does not stop existing; it stops being handed to you, and being handed the answer was never the job here. I have spent 25 years building the instruments this industry lacked, from customer lifetime value to the handset replacement cycle to effective price per minute, message, and megabyte, and rebuilding one a carrier takes back uses the same muscle. That is the whole of the second quarter: five operators obscuring the composition of their growth behind four different units, and in a convergence market composition is the entire contest.

Begin with T-Mobile’s chosen unit, because it teaches the most. An account deactivates only when its last line leaves, so a household that drops a tablet, a watch, and a second phone still counts as fully retained. T-Mobile reported 0.99% postpaid account churn, a floor under line loss and never a ceiling, and the floor sinks as the strategy succeeds: lines per account run near 3.44 on our estimate, T-Mobile having retired the customer count that would let anyone check it. Its own definition shows why the unit cannot bear the weight, since an account blends phones, 5G gateways, fiber, tablets, hotspots and wearables into one item. Acquisition diluted it further, roughly 2.2 million accounts bought across Lumos, Metronet and UScellular, many broadband-only and carrying about one line, and T-Mobile attributes the 7 basis point rise in account churn to exactly that.

Verizon hides the opposite half, and it has a reason to. It reported 184,000 consolidated postpaid phone net additions and its best Consumer second-quarter phone result in five years. A Consumer and Business split puts two numbers in front of the market every quarter and a miss in either gets punished on its own, while the aggregate lets growth in the stronger segment cover a decline in the weaker and still read as a win. That is why the split left the tables.

Underneath it the household base is shrinking. Across the four quarters to June 2026 Verizon added 891,000 retail postpaid phone lines while its retail postpaid account base fell 409,000, from 34.6 million to 34.2 million, including 132,000 in the second quarter alone. Accounts have fallen in five of the six quarters since March 2025. The lines went up and the households went down.

Revenue per household went down too, and Verizon explained that one in advance. ARPA fell from $170.79 to $168.35. Dan Schulman said on May 21, 2026 that promotional amortization was running almost 190 basis points of headwind and would turn as soon as the third quarter, and that the decision not to raise prices again would lap by then. A disclosed 1.9-point headwind is larger than the 1.4% ARPA decline, so that line sits inside an accounting cycle rather than proving an exodus, and Comcast reported convergence ARPA down 1.5% in the same quarter on broadband pricing pressure. What sits outside the cycle is the line itself: revenue per postpaid connection fell about 3% while lines per account rose from 3.63 to 3.70, so each added line earns less than the line already there.

Both halves of the Verizon engine improved. Gross additions rose 2.8% and churn fell to 0.92%, and the 193,000 year-over-year gain splits roughly 141,000 from lower churn against 75,000 from more gross additions, less about 23,000 absorbed by the larger base. Schulman said in May that at least half would come from churn and possibly more; it arrived near three quarters. Retention did the heavier lifting, succeeding at the line while failing at the account.

Verizon’s profit number is the most impressive thing in the quarter, and it is a cost story. Total operating revenue fell 0.7% to $34.3 billion and adjusted EBITDA still grew 7.2% to $13.7 billion, lifting margin from 37.1% to 40.1%, both records. Adding roughly $920 million of EBITDA on revenue that shrank about $240 million means close to $1.2 billion of cost came out of the quarter. Free cash flow grew 24.4%, adjusted EPS rose 6.6% to $1.30, guidance went up a second consecutive quarter, and the buyback target rose to as much as $4.5 billion. Two things temper it. Reported EPS fell 22.0% to $0.92 on roughly $655 million of rationalization and severance charges. And part of the margin is arithmetic rather than discipline: equipment revenue fell nearly 20%, over $1.2 billion, as the upgrade rate went from 3.6% to 2.6%, and selling fewer subsidized handsets flatters margin on its own. A fall iPhone launch could reverse that half.

AT&T added 432,000 postpaid phone lines, its best wireless quarter in years, and FirstNet accounts for roughly 66,000, putting consumer phone near 366,000. Nearly all of it came from inside the fiber footprint, and AT&T’s own disclosure shows it. Converged customers on fiber, households buying both AT&T fiber and an AT&T consumer phone line, grew 178,000 in the quarter. An average US household holds 2.4 people aged six or older, so at two lines each those households brought about 356,000 phone lines against 366,000 consumer additions. Two rather than 2.4 leaves room for the households that already had AT&T mobile before the fiber sale.

the house on the account. A family that has already chosen AT&T for both products is the likeliest family to put everyone on one plan. Above that line, growth outside the fiber map is negative. Using the broader convergence figure would overstate the case: converged customers across all advanced home internet grew 302,000, but 124,000 came on Internet Air, the fixed wireless product, which sells to households that typically already have AT&T phones and brings broadband revenue rather than phone lines. Fiber pulls mobile, and mobile pulls fixed wireless. Inside 38.6 million fiber locations the phone business and the fiber build are one machine, and outside them AT&T is waiting for the truck. It publishes phone churn three ways and splits none by converged status. Converged households churn less than single-product households, so both the fiber build and the FWA sale are buying retention as well as growth, and what stays off the page is not whether convergence works but how much it is worth.

Cable is one strategy at two operators, and this quarter does not separate them. Comcast added 448,000 domestic wireless lines, its best on record, reaching 10.2 million, and Charter added 406,000 to reach 12.5 million. On the broadband side they lost 167,000 and 172,000, within 5,000 of each other and both worse than the prior quarter. Which one looks better on broadband flips with the comparison base, Comcast ahead through 2024, Charter through most of 2025, Comcast again in 2026, so a single quarter’s year-over-year gap says more about the year-ago quarter than about this one. What is durable is the plan both chief executives are executing. Converged households churn less than single-product households, so selling wireless into the internet base insulates it, and Charter pushed hardest in the quarter, guaranteeing $1,000 of first-year savings to households moving two or more lines from Verizon, AT&T or T-Mobile. Comcast’s convergence ARPA runs near $85 against $168.35 per postpaid account at Verizon and $152.91 at T-Mobile, funded by offloading about 90% of mobile traffic onto its own Wi-Fi across 65 million converged passings. No carrier reaches that by cutting price. The cost is churn neither discloses, which we put near 2% monthly against the sub-1% the carriers report.

Each carrier executed its chief executive’s plan almost flawlessly, and that is the risk. AT&T’s growth is real and it stops at the edge of the fiber map. Verizon’s retention engine and its cost discipline are real, and the household base both work on keeps shrinking. T-Mobile’s margin is real and the metric that would price it is no longer published. The danger is not that the plans fail. It is that they succeed exactly as designed, and each has retired the one number that would show what success costs.

Source: carrier filings and calls pulled July 26, 2026 for the quarter ended June 30, 2026. Verizon 2Q 2026 Financial and Operating Information and earnings release, plus investor conference remarks of May 21, 2026; T-Mobile Q2 2026 Investor Factbook and supplementary data; AT&T 2Q 2026 earnings release and trending schedule; Comcast 2Q 2026 earnings release and earnings call of July 23, 2026; Charter 2Q 2026 results release and trending schedule.

Americans love their phones and hate the deal that puts the phone in their hands for free. On Recon Analytics Mobile Pulse, about 337,000 wireless interviews in the trailing twelve months through June 26, 2026, the phone itself is the most loved thing in wireless at +23.8 cNPS; the handset deal that delivers it is the least loved at -11.8, below billing support, and no combination of carrier and customer group scores the deal positive, including the customers actively receiving a subsidized phone. The same offer is a top-three reason to choose a carrier for 24% of switchers and, among customers leaving within three months, a top-three reason to leave for 29%. Handset financing is demanded and hated at the same time, by the same people, and behind the averages the machine now runs backwards: the fast upgrader, once the most expensive customer to subsidize, now pays for the speed and trades back a nearly new phone, while the customer who waits for the old phone to die takes the full subsidy and hands back a device worth almost nothing.

Hated From Day One, Even When the Phone Is Loved

The resentment attaches to the deal, and the cleanest evidence holds the hardware constant. Among customers whose phone is under a year old, the outright cash buyer scores +19.3 with 5.6% saying they are leaving within three months; the financing customer with identical-age hardware scores +5.0 with 12.1%. Same phone, and the meter alone costs 14 points of satisfaction and more than doubles imminent exit. The grievance starts immediately: new financing customers run negative through their first six months, bottoming at -6.4 cNPS, with 41% to 42% detractors inside two quarters, worn down by ports, trade-in shipments, bill uploads, and reward-card cycles before the first dropped call.

The Agreement Outlasts the Complaint

Financing customers say they are leaving within three months at a 10.2% rate yet churn roughly 0.7% a month while the balance runs, about 2% in a quarter, a five-fold overstatement. Bring-your-own customers say 5.7% and churn near 1.4% a month, the highest of any group and the word closest to the truth, because nothing stands between intent and action. The unhappiest customers stay the longest; the group most content on value leaves the most; and the customer who financed a phone, paid it off, and stayed scores +25.5, the most satisfied in the market.

The Resentment Survives a Raise

Income does not buy the deal peace. The money complaint concentrates where it should: against paid-off peers at the same income, the financing customer’s value-for-price deficit runs about 25 points below $50K and narrows to 12 points above $150K. But overall satisfaction among financing customers is essentially flat from $25K up, value-for-price is negative in every band, and stated exit peaks at the top of the scale: 12.7% of financing customers between $100K and $150K say they are leaving within three months, 10.8% above $150K, against roughly 4% of paid-off customers at every income. The affluent financing customer complains less about the price and queues at the door more; the lower-income customer names the price and stays, the balance a harder leash for the household that cannot write the payoff check. And the exposure concentrates exactly where the carriers can least afford it: the share of customers financing rises from 37% below $50K to nearly half above $100K, while bring-your-own falls from 11% to 5%, so the deferred-exit inventory sits deepest in the young, affluent, multi-line household the machine was built to win.

Why Carriers Subsidize Instead of Encouraging BYOD

The subsidy is an acquisition and financing machine, and bring-your-own feeds it nothing. The free phone wins the switcher: attractive device deals are the strongest switching lever a carrier controls by decision rather than a decade of capital. It selects the customer worth winning, the young, multi-line, high-income postpaid household. The balance it creates converts stated leaving into staying, 0.95% blended monthly churn against 1.4% for bring-your-own. And its conditions upsell the premium plan and device protection, about $62 a month of service ARPU including protection against $49. A subsidy happens discretely, when a phone changes hands: on the roughly four-year rhythm the actuals now show, a nine-year customer takes three subsidized phones, and what they cost depends on what comes back. Weighted by where the base sits, the subsidized customer carries about $1,450 of device cost and returns about $1,900 of net lifetime value against $1,516 for bring-your-own, repaying the phones through the relationship. The bring-your-own customer is cheap to serve and profitable from the first day, and worth less by the end, because nothing binds them and nothing upsells them. That is why three sophisticated carriers kept paying for six years.

Behind the average, the machine runs backwards. Upgrading fast is now sold, through fees, payoff thresholds, and mandatory turn-ins, and at July 2026 buyback prices, up to $999 paid for a year-old iPhone 17 Pro against its $1,099 launch price (SellCell, 11 July 2026), the fee plus the returned phone more than covers the early upgrade. At AT&T, $120 of annual Next Up Anytime fees and the returned flagship exceed the credits paid and the balance waived, roughly $200 earned per annual cycle on the device motion alone; T-Mobile buries the same charge in its top plan tier, and Verizon sells the fastest lane as outright rent at $50 a month with the phone inside the fee. Every carrier now profits on the fast upgrader; they differ only in how visibly they charge. The expensive lane is the crowded one: the 44% of the market that replaces a phone only when it dies hands back a dead or damaged device worth perhaps $150 against a near-maximum credit, collects three full-freight subsidies, and nets out at roughly the bring-your-own customer’s value. The deal’s entire surplus comes from the middle of the cycle curve.

Why the Carriers Are Looking for Alternatives

Every input that made the machine pay is decaying at once. Flagship retail roughly doubled over the decade while the subsidy grew by about a third, from roughly $400 to $450 per device in the contract era to about $600 net today, so the deal covers a smaller share of a pricier phone and the customer feels it. The next cycle is steeper: DRAM contract prices more than doubled in the first half of 2026, NAND rose 70% to 75% (TrendForce contract-price releases, 31 March and 14 May 2026: trendforce.com/presscenter/news/20260331-12995.html and 20260514-13044.html), and Apple’s fall devices are expected to cost hundreds of dollars more. The customer, meanwhile, holds the phone close to four years, up from about two a decade ago: phones three or more years old grew from 21.8% in our survey base in the third quarter of 2023 to 28.6% in the second quarter of 2026.

Battery is the wear that shows first, only 15% of buyers named the new model’s release among their reasons, and the deal has become the financing method to remedy a failing device rather than for the multi-media event that creates the purchase. The agreement’s grip weakens as eSIM, one-tap portability, and rival balance buyouts cut the cost of leaving, and the generosity earns no affection to fall back on as it is taken for granted.

Since the current device deal structure is creating resentment, the carriers are trying a different approach. AT&T’s Build-A-Plan strips the subsidy from the price at $15 a line; OneConnect trades the phone deal for fiber. Verizon fields Simplicity at $30, paid upgrade tiers that sell the speed outright, and the Verizon One bundle, with acquisition and retention spending down about 35% (Verizon Q1 2026 earnings call, 27 April 2026). The convergence bundles hold a household at roughly 0.6% monthly churn on an asset that appreciates, though they are for the new-new customer only, on the order of 10% of the decision pool, and online-only, which narrows the door further. From the customer’s side the bundle does to the mobile line what the subsidy did to the phone: OneConnect is $90 all in for a line and gigabit fiber, and crediting the fiber at its $65 retail price leaves the phone line at about $25, while Verizon One’s line lands between nothing and ten dollars once the home connection is valued at retail. The near-free line replaces the free phone, now anchored to a home connection no rival can buy away with a balance payoff. T-Mobile holds the heaviest promo engine while its financing customers resent it least, and waits. A subsidy is the best retention tool only while nothing better exists, and 2026 is the year better may have arrived.

Will Customers Break the Habit They Love to Hate?

The exit exists; we don’t know yet if the customer is accepting it. If buyers take the clean low price, the subsidy narrows to a switcher-only tool. If they keep taking the free phone, the model carries forward on a margin device inflation is erasing, funded by the customers who wait for their phones to die. The deciding rule is the one this note runs on: actions speak louder than words. Customers call the device deal the thing they most dislike about their carrier, a quarter of them switch partly to get one, and nearly half replace a phone only on failure, whatever the offer. The industry has now priced the way out. The answer will not surface cleanly in the filings, where the subsidy amortizes into service revenue and rising below-the-line fees blur the picture; it will show first in the advertising weight behind the new plans, in the panel’s financing and bring-your-own mix, and in whether the clean plans hold or grow their old subsidies back one feature at a time. Watch what they buy.

Why do business customers stay with their telecom operator? Ask around the industry and you’ll hear the usual answers: contracts, bundles, the hassle of migrating service, the shortage of alternatives in some markets. All of that is real, and all of it shows up in our data. But when Recon Analytics asked business decision-makers directly why they haven’t switched despite being unhappy with their primary provider, one of the most common answers given is good account manager relationship.

Between November 5, 2025, and May 6, 2026, Recon Analytics asked 1,355 mobile and 1,439 internet U.S. business service decision-makers why they haven’t left their current primary mobile or internet provider despite being unhappy with their provider. Among large businesses (1,000+ employees), 36% told us a good account manager relationship is one of the reasons they haven’t switched mobile providers. Midsize businesses (20-999 employees) came in at 35%. Small businesses (1-19 employees), at 9%. For not switching primary internet provider 32% of large businesses, 28% of midsize, and 11% of small said good account manager relationship. When it came to the most important reason for not switching primary internet provider large and midsize businesses selected account management relationship the most often.

Why a Person Beats a Price Sheet

Here’s the part the switching-cost spreadsheet misses. Leaving an operator isn’t just canceling a service. When there’s an account manager involved, it means telling someone you know, someone who took your calls and fixed your problems, that you’re walking away. Most of us will do a lot to avoid that conversation. Once a personal relationship is established, breaking it stops feeling like a procurement decision and starts feeling like letting a friend down. That discomfort buys an operator something no discount can: forgiveness.

A billing error or a coverage gap that would send an anonymous customer shopping becomes something to work through, because the customer isn’t mad at a logo. They have a name to call, and they’d rather give that name a chance to fix it than walk away.

The Relationship Only Works If You Show Up

Now for the uncomfortable data. Between March 4 and April 1, 2026, we asked 1,581 business decision-makers whether they have a dedicated account manager from their mobile carrier and how often they actually work with them. Coverage looks fine on paper: 90% of large businesses have one, and so do 79% of midsize businesses. Cadence is where it breaks down. Among businesses with an account manager, 30% of large accounts and 35% of midsize accounts work with that person only once a year, or only when the contract comes up for renewal.

An account manager who appears once a year, right when the customer is deciding whether to stay, isn’t building a relationship. That visit is a sales call, and the customer knows it. Loyalty is built between contracts through regular conversations that are not entirely focused on reupping a contract.

What Operators Should Do With This

If you run a business-to-business unit at an operator, customer retention programs need to be based on regular customer communications. The midsize tier deserves the most attention: those businesses value the relationship at nearly the same rate large enterprises do, yet they’re covered more thinly and contacted less often. Setting a quarterly-touch floor for every large and midsize account with an assigned manager is about the cheapest churn insurance available. Building a relationship helps to create value in the commercial relationship and prevents the relationship from devolving into being totally about getting the lowest price on mobile and internet services.

Small business needs a different answer, and honestly, a dedicated account manager isn’t it. There are far too many small businesses generating far too little revenue each to justify one, and most small businesses agree: over half of those without an account manager told us they don’t want one. But that doesn’t mean the only thing an operator should send a small business is the bill. A scheduled check-in call from a pooled business team twice a year, a quarterly service review with an actual name attached, even a real person reaching out when usage patterns change: none of that requires enterprise economics. The goal is the same one the account manager serves upmarket. Give the customer a human reason to pick up the phone before they pick up a competitor’s offer. Make the small business feel like a person, not just a number.

Of course, stated reasons for staying are attitudes, and churn is a behavior; some businesses that praise their account manager will leave anyway. But the pattern held across mobile and internet, across two question forms, and across a six-month field window. Business retention in telecom is bought with regular conversations. By mid-2027, I expect at least one nationwide operator will formalize a quarterly business-review program below the enterprise tier and market it as a feature, because the first one to do it will be selling something the survey says customers actually want and stay for.

Methodology: Recon Analytics survey of US business telecom decision-makers. Switching-reason questions were asked of respondents who rated their provider’s overall complete experience 0-6 on a 0-10 scale and had not changed providers; fielded November 5, 2025 through May 6, 2026 (mobile n = 1,355; internet n = 1,439). Account manager coverage and contact frequency were asked of all respondents, March 4 through April 1, 2026 (n = 1,581). Percentages computed on valid answered bases.

The public comment period has now ended in the docket opened by the Federal Communications Commission to help the agency determine whether and why the agency should preempt California state laws that prohibit former telcos from retiring copper plant in favor of fiber, fixed wireless or other high-speed connections.  The central question being asked in the FCC’s proceeding is whether California can (and should) force a telephone company to keep 130-year-old copper wire in the ground indefinitely, and call that consumer protection?

The answer is no, it cannot, and it should not. The FCC has already told states they may not order carriers to maintain the copper network forever, and in June it cleared AT&T to retire copper voice service at more than 184,000 California locations. The state is fighting to undo it, in the Ninth Circuit and now in this docket. On the facts, and on what Californians themselves say they want, the FCC has this right and California has it backwards.

Start with the copper. The copper telephone loop is the first strung technology in the 1800s, older than powered flight, older than the transistor, and older than the interstate highway. It was a marvel in its era. That era is over. A pair of copper wires carry a few megabits on a good day, corrodes in the rain, and is worth enough as scrap that thieves cut it out of the ground. Fiber carries thousands of times the capacity over the same route and costs less to run once it is in. Fixed wireless and satellite now reach the hardest addresses for a fraction of what new copper costs to deploy. There is no measure left on which copper wins.

Californians know it. Over the past year, Recon Analytics asked more than 40,000 of them which provider they would choose for home internet. Fixed wireless drew 45%, fiber 24%, and satellite 4%: nearly three-quarters chose a technology that is not copper. Cable took 12%. Copper DSL drew under 8%, and even that overstates it, because a third of the people who named AT&T’s copper option already have AT&T fiber. They are not asking for copper; they are staying loyal to AT&T. Strip the brand loyalty and demand for copper as a technology rounds to nothing. Given a real choice, Californians pick the future, and they pick it decisively.

So what is California protecting? The case for the mandate is that copper is a lifeline: it carries its own power, works in a blackout (as long as one has back-up battery power for their cordless phones connected to the copper), and the people still on it skew older, poorer, and more rural. The fear underneath is that allowing companies to retire copper means some consumers will have zero connectivity at home, no voice, no Internet.  This is simply not the case.

Consider the FCC’s National Broadband Map.  Of California’s 10.25 million homes and small businesses with broadband service, the map counts 39,024, or 0.4%, with DSL as their only wired or fixed-wireless option, and every one of those 39,024 sits within reach of satellite. Not a single California household would be left with zero connectivity solutions.  None. The people losing a DSL line are not falling off the grid; they are being moved off the slowest thing on it, toward the cable, fiber, fixed wireless, or satellite options that already reach them.

The blackout worry deserves the same straight answer. It describes copper as it was engineered, not copper as it has aged; decades of deferred maintenance, water damage, and theft have left much of the old plant failing in the very storms where it is supposed to shine. The replacements meet it with battery backup, and the FCC did not simply switch the copper off. It requires a carrier to put an equal or better service in the customer’s hands first, with working 911 and access for people with disabilities, before the old line goes dark. The protection travels to the new technology. What the rule ended was the requirement that it be delivered over Victorian wire.

Look closely and the mandate protects nothing. It freezes a sliver of customers on the slowest network in the state, and it does so at real cost. Keeping copper energized is not free. The money comes out of the same budget that builds fiber, and every dollar spent nursing a corroding line is a dollar that does not connect a home to gigabit service. California’s own consumer advocates have already sketched the way out, agreeing to let a carrier leave copper behind where it commits to build fiber in return. This sort of 1:1 approach belies California’s grave misunderstanding of the communications solutions available to its citizens.  Fiber is a fantastic technology but not in every location.  Fixed and mobile wireless are also fantastic solutions in many geographies but not in all.  And then there’s satellite, including satellite D2D.

California will tell the Ninth Circuit that the FCC has acted arbitrarily and capriciously by granting AT&T’s request to transition 180,000 customers from old copper to a newer, more capable connection.

There is nothing arbitrary about the FCC’s effort to transition the country from antiquated analog networks to an all IP infrastructure.  And Californians have spoken.  They want fiber, wireless, and satellite, and almost no one is actually stuck on copper. The arbitrary act is the one that commands an industry to maintain technology of the telegraph age, forever, and dress it up as looking out for the little guy. The record is closed and the call is easy. The future is fiber, wireless, and satellite. It is not more than a hundred years of copper, and no state should be able to order anyone to pretend otherwise. Let the copper go.

By: Roger Entner, Analyst and Founder

The AWS-3 reauction closed at $3.57 billion, and the number is inflated. Strip out EchoStar, which bid to raise prices rather than to win, and the auction raises about $2.0 billion. EchoStar walked away with two trivial licenses in Guam. Its purpose was never to win spectrum; it was to push the clearing price past the roughly $2.9 billion that erased its own default penalty, and Verizon and T-Mobile paid the difference. That is the story of this auction.

How the three MNOs bid

AWS-3 was a fill-in auction. The three nationwide carriers were not building new footprints; they were topping up capacity in the markets where they are tightest. They are tightest in dense cities, where traffic outruns spectrum, which is why the money concentrated in a handful of big-market licenses and why those licenses cleared so high.

Verizon was the only bidder that treated AWS-3 as strategic, and it spent where it is shortest. It entered with roughly 248,000 bidding units, held demand across about 194 markets through round 50, and closed with 82 licenses for $3.16 billion, 88.5% of everything raised: $924 million for New York, $776 million for Chicago across two blocks, $158 million for Boston. Those are the markets where its urban capacity is most strained. Verizon decided that depth was non-negotiable and absorbed the clock to get it.

AT&T bid like a company looking to fill in its portfolio in a disciplined way. It opened on 39 markets, closed on 8, and won 10 licenses for $121 million with a $66.5 million Charlotte anchor. The company’s more significant spectrum play is its purchase of EchoStar’s 600 MHz and 3.45 GHz spectrum, which gives it a deep, nationwide footprint that drops straight into its converged fiber-and-wireless build.

T-Mobile carried the second-largest eligibility into the room, about 184,000 units, enough to fight Verizon market for market, and chose not to. It held that eligibility flat through round 30, then bled it down to roughly 14,000 by the close, dropping every expensive market as the price rose. It won the most licenses by count, 102, but spent only $278 million at a $1.7 million median, almost all small metros. A carrier that already leads on network adds cheap capacity at the edges and refuses the war up top.

What each carrier will do with the spectrum, and why it bid that way

The three carriers’ spending ranks their spectrum positions. The AWS-3 they won is paired mid-band that sits next to the AWS spectrum all three already run, so it deploys fast, by aggregating onto existing 5G radios and sites rather than building anything new. The question for each was not whether it could use the spectrum but how badly it needed it, and the answer ran inverse to how well-supplied each already is.

Verizon needed it most and will deploy it fastest. It carries dense urban traffic and a growing fixed-wireless load, and its mid-band depth arrived late, with C-band. AWS-3 adds paired uplink and capacity in exactly the cores where C-band is most strained, and it lights up by aggregating with the AWS and C-band Verizon already operates there. The $3.16 billion buys near-term capacity relief in New York, Chicago, Boston, and the other top metros, which is why Verizon, and only Verizon, treated the auction as a must-win.

T-Mobile needed it least and bid accordingly. It holds the deepest mid-band position in the industry, the 2.5 GHz layer it inherited from Sprint, which already blankets its markets with capacity. Urban AWS-3 was redundant to a network already long on mid-band, so T-Mobile let the expensive licenses go and spent $278 million on cheap small-market licenses, incremental capacity and optionality at the edges where adding a layer costs little. It will fold those into existing sites where they help and hold the rest as low-cost insurance.

AT&T had already solved the problem elsewhere. Its nationwide capacity need is being met by the roughly $23 billion purchase of EchoStar’s 600 MHz and 3.45 GHz, deep low-band and mid-band that drops straight into its build. That made AWS-3 marginal, so AT&T cherry-picked 10 markets where it had a specific gap and the price was right, Charlotte the anchor, and spent $121 million filling them, deploying onto infrastructure it already runs.

The corollary is simple. This auction was a revealed map of who is short of urban capacity and who is not. Verizon is short and paid, T-Mobile is long and passed, AT&T bought its way out separately and only topped up. None of it is a new network; all of it is capacity poured into markets these carriers already serve. That is what fill-in means, and every real bidder treated the auction that way. EchoStar did not.

EchoStar gamed the same system twice

In the 2014-15 AWS-3 auction, Dish, now EchoStar, bid through two designated entities, SNR Wireless and Northstar Wireless, and claimed roughly $3.3 billion in small-business bidding credits. The FCC found Dish controlled the entities, voided the credits, and the licenses were disgorged. About 197 of them are what Auction 113 reauctioned. The first game used the designated-entity rules to try to buy spectrum at a discount it was not entitled to, and the taxpayer carried the cost when it failed.

The second game used the auction’s own mechanics, and the incentive was inverted. As the defaulting party, EchoStar owed the shortfall if the reauction raised less than what was owed, with the liability extinguished once the auction cleared roughly $2.9 billion. So EchoStar had a dollar-for-dollar interest in a high price and no interest in winning anything. It posted Verizon-scale eligibility, roughly 249,000 bidding units, the kind of commitment that signals a serious national bidder, and used it to carry phantom demand across as many as 199 markets through round 40. That demand pushed the clock higher across the entire license map, for every real bidder in the room.

What it did when the price cleared is the tell. EchoStar was still holding 80 markets of demand at the round the auction crossed the threshold that erased its liability. The next round it cut to 23 markets. The round after that, to zero. It spent more than 45 rounds inflating the clock, and the moment the running total guaranteed its penalty was gone, it dropped its entire position and stopped bidding for the final stretch of the auction. It rode the price to the exact level it needed and got off, closing with the two cheapest licenses on the board, a pair of Guam licenses at $614,000 each.

Guam was not a target; it was the residue. EchoStar was demanding nearly every license to lift the clock, and on Guam the only genuine local bidders, Docomo Pacific and PTI Pacifica, had dropped out by round 35. With no one left to take the licenses, EchoStar’s own demand stuck them to it. It will not build them. A company exiting facilities-based wireless has no use for two island licenses, so expect them sold to a local operator or left to sit against a buildout clock, the same kind of obligation that started this whole story.

The pattern across both games is the same: use the structure of the auction to extract value it could not win bidding straight, and leave someone else holding the cost. The Treasury the first time, Verizon and T-Mobile the second. None of the second game broke a rule. EchoStar bid inside the auction’s own mechanics, and that is exactly what makes it effective and repeatable. The design permitted a defaulting party to bid up the reauction of its own forfeited licenses to cover its own penalty, and EchoStar used the design as written.

What would have happened if EchoStar had stayed home

Strip EchoStar’s demand out of the bid file, let every license clear at the round its demand would have settled without it, and hold the other bidders constant. The reauction raises about $2.0 billion instead of $3.57 billion, a $1.6 billion difference. And $2.0 billion is below the roughly $2.9 billion that extinguished EchoStar’s liability. Without its own phantom bidding, the reauction of EchoStar’s defaulted licenses falls close to a billion dollars short of what EchoStar owed, and EchoStar pays the gap. EchoStar bid to rescue itself, and the carriers funded the rescue.

Where the $1.6 billion came from:

What would have happened if EchoStar had stayed homeThese without-EchoStar figures are a first-order model that holds the other bidders’ behavior constant. The direction and the scale are robust; the exact dollars are estimates.

EchoStar came to the auction as a seller

The bidding makes sense once you see what EchoStar now is. It is exiting facilities-based wireless, having agreed to sell its AWS-4, H-block, and AWS-3 spectrum to SpaceX for roughly $19.6 billion. A company selling its spectrum and leaving the business had no strategic reason to win AWS-3 licenses. Its only reason to be in the room was financial, to clear its penalty, and that is exactly how it bid. Any accidental wins could be sold with the remaining AWS-3 licenses it still holds.

America has two half-connected segments and a third that is not connected at all, and the policy conversation gets all three wrong. The two half-connected segments are mirror images, each on a single network: the household with a phone and no home internet, 14.8 million strong at 11.5% of the country in the 2024 American Community Survey five-year estimates, and its mirror, the household with home broadband and no cell phone, three to six million adults the industry does not track. The third segment is connected to nothing, 11.5 million households, 8.9%, with no internet subscription of any kind, and most of them are offline by choice. The dominant reason they give is not price and not availability; it is that they do not want the internet, and that answer has grown through every subsidy era. Much of the disconnection in this country is a decision, not a deprivation, and a free society owes that decision the respect of its dignity, not a conversion campaign. The two half-connected segments fail the policy frame differently: it prices the connection and forgets the endpoint. Lifeline and the programs after it hand a low-income household a working phone with the service included; no program of any size hands that household a computer. A home with no computer gets nothing from a wired connection the phone in its pocket does not already deliver. Call it the missing screen.

The screen is where the adoption story actually turns. In the 2024 estimates, 12.2 million households own a smartphone and no other computing device, and 5.8 million own no computing device at all: 18.1 million households, 14.0% of the country, whose largest screen is a phone. The long arc that looks like a triumph, no-subscription households falling from 25.3 million in 2017 to 11.5 million in 2024, was closed largely by the phone, not the wire. Cellular-data-only subscriptions rose from 8.9 million to 14.8 million over the same span, so more than 40% of the headline gain was a phone plan. Households with no computing device at all fell from 15.2 million to 5.8 million while smartphone-only-device households climbed from 4.7 million to 12.2 million. America’s poorest households did not buy computers. The phone became the computer, and every statistic that counts a cellular plan as a connected household ratified the substitution.

The households living this way are not the deprived customers the policy frame imagines. In the Recon Analytics Pulse over the trailing twelve months, from June 19, 2025 to June 19, 2026, smartphone-only customers rate their wireless provider at +24.4 cNPS, about seven points above the +17.3 that households with home internet give their carriers. They are more satisfied with their one network than two-network households are with either of theirs. They skew poor, 80% under $50,000 in household income and 55% under $25,000, and prepaid, 28% against 18% for dual-service households, and value brands led by Straight Talk, Cricket, and Metro by T-Mobile hold roughly 40% of the segment. A pitch built on what they are missing bounces off people who do not feel they are missing anything.

The no-internet-at-home segment is the poorest of the three, and its +11.5 cNPS, though below the smartphone-only score, is still solidly positive. Its income tells the policy story: 37.5% report household income under $10,000, against 26.7% of the smartphone-only group and 10.1% of dual-service households; 63% sit under $25,000 and 87% under $50,000. This is Lifeline’s customer base, the household for whom the marginal dollar competes with groceries. And yet affordability no longer explains why they stay offline. The NTIA’s offline households name “don’t need it, not interested” as their main reason 55.7% of the time, up for a decade, while “too expensive” fell from 18.8% in November 2019 to 15.4% in November 2023 and “not available” sits at 2.8%. Through the entire ACP era, the largest connectivity subsidy in American history, the don’t-need share rose and the too-expensive share fell. Price is not the residual barrier. The cleanest refuser is the older household, two-thirds of offline seniors give the don’t-need answer, that ran a full life without a computer and sees no reason to buy one now. A connectivity policy that cannot accept that answer is a conversion campaign, and the honest public metric is universal offer, not universal adoption.

For carriers, the prize is not the refuser. It is the device-blocked household and the second network it will eventually add, and one variable decides which brand gets it. Cross a smartphone-only customer’s current mobile brand against the home internet brand they would choose, and the brand on the bill, not the network underneath, predicts the answer. Metro customers stay in the T-Mobile family 56% of the time, T-Mobile customers 50%, Verizon 47%, AT&T 44%, Straight Talk 35%. The control case proves it: Cricket is owned by AT&T and runs on AT&T’s network, but the bill says Cricket, and only 16% of Cricket’s smartphone-only customers pick an AT&T-family home product. They scatter to cable. The network transfers nothing; the brand on the bill transfers almost half. Call it one-bill gravity, the cheapest customer-acquisition channel in telecom, where the relationship already exists and only the second product is missing. The carriers are now packaging it: Verizon One bundles a line with home internet at $70 a month, and AT&T OneConnect hard-bundles fiber and wireless from $90, both selling the second product into an existing bill.

Wireless is taking share among exactly these households. Among those earning under $50,000, 9.0% named a wireless connection, fixed or prepaid, as their home internet provider over the trailing twelve months, and 10.8% in the most recent quarter, a gain of about 0.8 points a quarter. Halve that slope to respect fixed wireless capacity limits and state low-income mandates, and the share still clears 12% by the end of 2027. The affinity points there too: forced to choose, 46% of smartphone-only households name a fixed-wireless or prepaid-wireless product, and Starlink’s 5% pushes wireless past half, against 21% for cable and 12% for fiber. T-Mobile and Metro hold the largest prize, 300,000 to 600,000 conversions worth $110 to $320 million in annualized service revenue at the entry tiers these households actually pay. Two things would break it: a fixed wireless capacity wall in the urban, lower-income ZIP codes where the segment lives, since spare capacity concentrates in rural markets; and cable pricing its entry tiers below the floor wireless can match, the dynamic that let cable win the subsidized ACP-era conversions four to one over T-Mobile’s own products. The wildcard above both is policy: a device benefit, an ACP successor that funds the screen rather than the connection, would attack the missing screen head-on and hand the segment to whoever bundles the computer with the wire.

The mirror segment runs on the same physics. The adult with home broadband and no phone is overwhelmingly a senior who pays her own bill: 84% of phoneless respondents over 60 pay the household internet bill, against 71% of phoneless 18-to-29-year-olds who do not, household members who will get a phone at household formation regardless. The senior is the buyer, she rates her aging cable or DSL line at -1.1 cNPS, and cable already bills 59% of these households. Cable owns the right first-line product in by-the-gig mobile, attachable with no new relationship and no second bill, and Consumer Cellular just sharpened the senior pitch with SpeakEasy, a brand for adults 75 and older selling a flip phone from $14.95 a month and a smartphone from $19.95, device included. It is the whole thesis in miniature: the first-line sale is won by whoever shows up with a working phone, not a SIM card and a setup guide. The segment is small, 200,000 to 500,000 lines worth $35 to $180 million a year, but for cable it is found money inside accounts already served.

The through-line is simple. The country did not close the digital divide so much as paper over it with a five-inch screen, and the policy and competitive playbooks both still aim at the connection when the constraint is the endpoint and the relationship. Subsidize the computer, not just the connection, and respect the dignity of the household that has weighed the offer and said no. Sell the second wire through the brand already on the bill, not the network under it. The half-connected are not waiting to be rescued. They are waiting to be sold the right thing, by the company they already pay.

Based on the Recon Analytics Pulse, both survey populations, July 2022 through June 19, 2026, roughly 1.84 million unweighted responses; the trailing-twelve-month window runs June 19, 2025 to June 19, 2026 (smartphone-only n=18,751; no internet at home n=8,786; no mobile service n=3,168). Population sizing from US Census Bureau ACS five-year table S2801, 2017 and 2024 releases; non-use reasons from the NTIA Internet Use Survey, November 2023; telephone status from the CDC NHIS, second half of 2024. cNPS is promoters minus detractors per Recon Analytics’ standard. Forced-choice responses measure affinity, not realized conversion.

For a decade, airlines treated in-flight Wi-Fi as a cost to be minimized or an ancillary to be sold by the megabyte, and the industry argued endlessly about the technology: air-to-ground versus satellite, Ku-band versus Ka-band, how much bandwidth a widebody really needed. That argument is over. Low Earth Orbit won. Starlink Aviation delivers roughly an order of magnitude more capability than the geostationary systems it is replacing, on bandwidth, latency, and coverage combined, and there is no GEO roadmap that closes the gap.

Adoption moved the way technology transitions always move, slowly and then all at once. More than 40 airlines had committed to Starlink by mid-2026, many of them in the prior 90 days, and the US mainline market has now sorted into two LEO camps: Starlink, which holds Southwest, United, Alaska, Hawaiian, and, as of late May, American; and Amazon Leo, which holds Delta and JetBlue. The technology contest is settled. The interesting contest moved to two other places at once: the airline loyalty ledger and the capital markets.

What the leaders understood

Here is the finding the laggards missed. A broadband-class customer experience, with free fast Wi-Fi as its most visible component, correlates with materially stronger satisfaction and loyalty economics than a metered, ancillary-by-ancillary model. The behavioral data is more telling than the stated intent. Frequent flyer membership, measured by the carrier each passenger most recently flew, runs from a high near 67% among the full-service network carriers down to 36% at the ultra-low-cost end. The meter-everything model produces the weakest loyalty penetration in the industry by a wide margin.

Wi-Fi is not the sole cause of that gap, and the full report is careful about what the data can and cannot prove. But the two carriers leading our Wi-Fi customer-cNPS table also sit near the top of the loyalty data, and the carrier at the bottom of one sits at the bottom of the other. Spirit ran that same metered model, and on May 2, 2026 it became the first major US airline to fail in 25 years. Free Wi-Fi did not kill Spirit, and free Wi-Fi alone will not save a weak carrier. But the model that treats every passenger touchpoint as a fee to be maximized is the model that loses the loyalty war, and the loyalty war decides which carriers survive a bad year.

Three of the four largest US carriers are now designing the free-Wi-Fi product as a loyalty gate, and at least one has begun citing the enrollment effect on earnings calls. The full report quantifies the per-aircraft economics: how many incremental enrollments a narrowbody generates each year, what share convert to co-brand cards, and why the same math that closes cleanly at LEO speeds does not close at GEO speeds.

The IPO raises the price

None of this changed when SpaceX went public. The price did. On June 12, 2026, SpaceX priced its IPO at $135 a share for a $1.75 trillion valuation, raised roughly $75 billion in the largest market debut on record, and opened well above the offer price. The IPO does not alter Starlink’s technology or its lead. It alters Starlink’s financial position and the bargaining table where airlines sit.

A public SpaceX has a lower cost of capital, an acquisition currency, and a shareholder duty to extract pricing power that a private supplier courting reference customers did not have. The pre-IPO window that produced the most favorable terms airlines will see for years has effectively closed. American signed Starlink on May 26, just over two weeks before the listing, which is the timing thesis of this report playing out in real time: it locked its terms while SpaceX was still private. The leverage that remains is narrower and time-boxed, and it sits with the carriers still negotiating renewals.

Winners, losers, and the re-rating

If LEO won, the question for anyone holding a position in the supplier stack, including the investors pricing it, is who survives the transition. Viasat is the most structurally pressured name. SES-Intelsat is a durable second in the multi-orbit niches where redundancy genuinely matters. Panasonic survives by retreating to its seatback-screen and software moat and ceding the connectivity layer. Amazon Leo is the swing variable: its FCC deployment milestone was waived on June 5, 2026, and its readiness now turns on launch cadence rather than a binary regulatory test. The full report maps each name to a four-to-five-year outcome and explains why a later Starlink-only carve-out, not the full-company listing, is the cleaner connectivity catalyst for investors.

The decision the facts force

The facts are the same for airline and telecom executives, the GEO players, and the investor. The decision they force is different.

For airline executives, this is a procurement-timing argument: when to lock long-dated LEO terms, why capacity-per-aircraft guarantees matter as much as price caps, how to reframe free Wi-Fi as a loyalty-acquisition channel rather than a sponsorship line, and how to build the dual-source optionality that keeps a public supplier honest at renewal.

For the telcos, the convergence is the whole story. The constellation that won the cabin is the one now reaching the phone in the passenger’s pocket. Starlink’s direct-to-cell service runs on the orbital capacity SpaceX sells to aviation, so the IPO that raised Starlink’s pricing power raised it across both markets at once. The aviation timing thesis is the telco timing thesis. American locked its terms while SpaceX was private; the operator that signs its direct-to-device capacity after the first public earnings call pays for the wait. The defensive read matches the GEO implication: single-sourcing the supplemental-coverage layer to the supplier with the most pricing power is the exposure, which is why AST SpaceMobile exists. The offensive read is larger. When one satellite serves both the cabin and the cellular dead zone, in-flight connectivity and direct-to-device collapse into a single capacity market, and the operator that treats its satellite relationship as a procurement line cedes the convergence.

For the GEO players, defending the connectivity layer is the wrong instinct. No roadmap closes the capability gap, and pricing head-to-head against Starlink is a losing trade. The defensible move runs the other way. Airlines want dual-source optionality precisely because a public SpaceX has every incentive to push pricing at renewal, which turns redundancy into a product. Viasat and SES-Intelsat can position as the disciplined second source, concentrating capital where LEO economics are weakest: maritime, government and defense, oceanic and polar routes. Panasonic’s seatback retreat applies the same logic one layer up, owning the experience the passenger touches rather than a commoditized pipe.

For investors, it is a supplier-stack re-rating map: which names to underwrite, which to avoid, and which catalysts to watch.

Data: Recon Analytics US Airline Pulse, pulled June 4, 2026, coverage November 1, 2025 to June 4, 2026, 84,612 respondents. SpaceX financial and IPO figures per the public S-1 and Reuters reporting.