Corning
A glassmaker that draws pure silica into the optical fibre carrying internet and AI data-centre traffic, and also makes screen glass, car exhaust parts and solar wafers.
The scores
Growth3/4 ↑
Nine straight quarters of growth and 32% in optical, but it narrows to one business and one end market rather than broadening.
Business quality3/4 →
Scale in a capital-heavy draw is real and buyers pre-pay for it, but a named rival just bought its way in above the glass.
Financial position3/4 →
Net debt is under half of equity and maturities are spread, but the expansion leans on customer deposits and on incentives netted against the assets.
Execution record3/4 →
Springboard has been raised twice and met, but three presentation changes in six quarters make the record harder to check than it should be.
Per share capture2/4 ↓
Stock compensation doubled, buybacks stopped with $3.0 billion authorised, a $2.0 billion share-sale program arrived, and a customer holds a 15 million-share warrant.
Potential3/4 ↑
Four named counterparties and one dated, volume-stated contract underpin it, but the later years rest on a photonics platform nobody has signed for.
Risk profile3/4 →
Not four: two customers are 28% of optical sales, three are 59% of display, and growth rests on a few hyperscale budgets.
Valuation2/4 →
Not three: the price pays for the whole $40 billion plan and the margin holding through it, leaving 22 times 2030 earnings.
Evidence quality3/4 →
Everything central is filed, but the later years rest on management claims graded rather than verified - the plan, the photonics platform, the agreement coverage.
No upside the evidence defends · if it fails you keep about 56%.
The Verdict
We rate Corning unattractive. It draws glass into the optical fibre that carries internet and AI data-centre traffic, and that business is growing fast with much of the demand signed, so the business is solid. But its biggest new customer, NVIDIA, was paid partly in Corning shares, and part of recent cash came from customer deposits and government money. The price is full, not cheap: it already treats management's whole 2030 sales plan as delivered, so any shortfall would hurt. As a bet it is fragile: the hoped-for light-based chip links business has no sales figure yet, so there is little to win and much to lose if it slips.
The long caseFor photonics to matter, AI chip makers would have to swap the copper links inside their racks for light, and buy Corning's fibre to do it.
The evidence under this - 8 claims
- Trailing earnings are the wrong denominator for a business management says will double: the vendor analyst panel's average 2028 estimate is $5.98 a share, which puts the same price at 26.2 times, and the panel's 2029 average at $6.40 puts it at 24.5 times.
- The GAAP figure is depressed by a genuinely non-cash currency item, and the core figure is inflated by the adjustment for it, so neither number is the economics - which is why the price is also stated against 2030 in C29.
- The shares are 42.3% below their 52-week high, so the multiple is already well down from where the market was willing to pay.
- 2026-09-26editedRe-struck on the research pack refreshed at the 2026-09-26 review (S28) at the 25 September close of $156.74; the earnings per share are unchanged, so only the price moved.
- 2026-10-27editedLeft as the dated 25 September statement it is; C59 re-strikes the price at the 26 October close. refresh_by moved to C59.
- The target has been raised twice in five months and its unit changed with it: a January 2026 upgrade to '$11 billion of incremental annualised sales by the end of 2028' became, in May 2026, a '$30 billion run rate by the end of 2028'. A plan restated in a different unit cannot be checked against its own prior version.
- Management declined to put a revenue figure on photonics, the platform the later years depend on, and Weeks said explicitly that the plan is built by 'adjusting for different probabilities of different timings and different content opportunities' - a probability-weighted internal plan, not a guide.
- Corning cannot reconcile any of it to GAAP: the release states that it is 'unable to provide outlook information on a GAAP basis'.
- He explicitly refused to guide it: 'I don't think we guided anything for next year on Photonics, just to be clear... We'll update sort of each quarter as we go' - so the figure is a platform ambition, not a plan component anyone can check.
- The in-rack link the platform targets is passive copper in the customer's own current documentation (C08), so the displacement has not begun where it matters most.
- Photonics appears nowhere in the segment note, the revenue disaggregation table or the 10-Q's management discussion: there is no line item to measure it against.
The company
Corning melts pure silica into a glass rod and draws it into fibre, then cables it and sells the connectors that join it up. Phone and cable companies and data-centre builders buy it because light carries far more traffic than copper; Verizon alone has contracted over 80 million miles to 2032.
Fibre is now 43.7% of sales and the business the growth comes from. The rest is display and phone glass, car exhaust parts, solar wafers and lab products.
The evidence under this - 3 claims
How it makes money
Corning sells fibre, cable and glass by volume, so its sales follow how fast customers build. More and more of the fibre is sold ahead under long-term supply agreements, some of them prepaid: money received for goods not yet delivered reached $2.7B in June.
Fibre earned $438M of the quarter's $846M of segment profit, 52%, while solar and lab products lost money.
The evidence under this - 2 claims
- 2026-10-27editedQ3 call: deposits are repaid as customers take volume
Where the money comes from, stream by stream
| Stream | Share of sales (Q2 2026) | How it earns it |
|---|---|---|
| Optical communications | 46.0% | Fibre, cable and connectivity sold to telecom carriers and to data centres, increasingly under prepaid multi-year supply agreements. |
| Electronics glass and advanced optics | 27.3% | Display glass for TV and monitor panel makers and cover glass for phone makers, sold by volume to a few large buyers. |
| Automotive | 10.5% | Ceramic substrates and filters for vehicle exhaust systems and technical glass for cars, sold to engine and car makers. |
| Polycrystalline silicon and solar | 9.7% | Hemlock's polysilicon plus solar wafers and modules made in the United States. |
| Life sciences | 5.5% | Laboratory glass and plastic labware and cell-culture products sold through distributors. |
| All other | 0.9% | Emerging businesses too small to report separately. |
Who runs it
Wendell Weeks has run Corning since 2005 and also chairs its board; he joined in 1983 and once ran the fibre business. Ed Schlesinger has been finance chief since 2022. Insiders own about 0.25% of the company, so the people in charge have little of their own money in it. Most of the bonus rides on Corning's own adjusted earnings per share and sales; the long-term awards pay on adjusted cash flow and sales, and none on how the shares do against other companies.
The evidence under this - 3 claims
- Every metric is a core (non-GAAP) figure Corning defines itself, and from April 2026 its new adjustment for hedged exposures raises core growth against the old one (C22, C23), so the bar the bonus is struck against moves with the company's own definitions.
- The proxy says Corning budgets for share repurchases when it sets the annual targets, and nothing pays on shareholder return against peers.
Why it wins, and who could beat it
Why customers come back
Fibre is drawn on expensive towers that take years to reach full output, and Corning has more of them than anyone. Customers are not locked in. But in a shortage there is no second supplier with spare capacity, so buyers who need volume pay ahead to secure it: NVIDIA deposited $1.0B, running to 2029.
CRU, an independent price reporter, said in May 2026 that American fibre plants were running flat out. The shortage is industry-wide, so it shows what scale buys rather than that Corning's is unmatched.
The evidence under this - 3 claims
- Pre-payment is equally consistent with a customer that wants a supplier's capital plan de-risked on its own terms - NVIDIA extracted a warrant over 15 million shares for the privilege of paying in advance, which is not what a captive buyer does.
- The moat is at the glass, and the growth product is increasingly connectivity and assemblies, where Amphenol, America Fujikura and Prysmian all compete and where Amphenol has just spent $10.5 billion.
- Corning's own 10-K says the Optical Communications competitive landscape includes 'industry consolidation, pricing pressure and competition for the innovation of new products' and that 'these competitive conditions are likely to persist'.
- The shortage is not Corning's alone: CRU reports Chinese bare-fibre prices up more than 80% between November 2025 and January 2026, so tight capacity lifts every producer, and it says what scale buys in a shortage rather than that Corning's scale is unmatched.
- CRU expects capacity additions in 'months rather than weeks' - a shortage measured in months is a cycle, and a cycle ends when capacity arrives.
The threat
The prize management wants most, the links between AI chips inside a server rack, is still copper. NVIDIA's own documentation says its flagship rack joins its chips through a copper backplane. And Amphenol, which already sells a rack's copper and power connections, spent $10.5B in January on the former CommScope's fibre cable and connector business.
The evidence under this - 3 claims
- The same customer put $1.0 billion of its own cash down and took equity, which is a costly way to express doubt about the transition.
- Weeks argued on the call that outside observers overstate the drama: 'all of us on the inside know the various trade-offs we can make to still deliver Scale-Up depending on what happens with other component suppliers'.
- Documentation describes what ships today and lags design decisions by a generation; the absence of optics in the current guide is not evidence about the next platform.
The rivals: NVDA, APH, TMO, DHR, AVTR.
Who else does this, rival by rival
| Ticker | Rival | Where it competes | How far along | Priced against it |
|---|---|---|---|---|
| NVDA | NVIDIA's own NVLink passive copper cable cartridge backplane, inside the GB200 NVL72 rackPriced against it: Different product | Photonics platform | On sale | Different product |
| APH | Amphenol, after its $10.5 billion purchase of Vistance's Connectivity and Cable Solutions business (the former CommScope), completed 9 January 2026Priced against it: Same product | Optical Communications | On sale | Same product |
| TMO | Thermo Fisher ScientificPriced against it: One segment only | - | On sale | One segment only |
| DHR | DanaherPriced against it: One segment only | - | On sale | One segment only |
| AVTR | AvantorPriced against it: One segment only | - | On sale | One segment only |
How big it could get
How big this could get
Nobody has added the market up, so it can only be bounded. About 60.2 million American homes still await their first fibre line, and Verizon alone is pushing toward 40 to 50 million homes. RVA, an industry researcher, forecasts about 213 million miles of fibre to link data centres over five years. Turning homes into miles needs a figure nobody publishes.
The evidence under this - 3 claims
- The count is homes marketed a fibre service, which is a carrier's passing rather than fibre bought: the miles of glass behind each passing are not published by anybody reached, so the count cannot be turned into a volume for Corning.
- The subsidised part of the remaining build is smaller than the headline: of the 3.79 million locations the federal BEAD program will fund, 62.9% are in line for fibre, per Broadband Breakfast's reading of the state plans.
Who has signed
Two deals are firm: NVIDIA's $1.0B deposit to 2029, and Verizon's 80 million miles to 2032. Meta's, worth up to $6B, and Amazon's, multibillion, disclose no volume. The catch is NVIDIA's price: a warrant over 15 million Corning shares, which will be counted as a cut to future sales.
The evidence under this - 5 claims
Where the growth is coming from, business by business
Where the growth is coming from
| Business | Share of revenue | Growth: last quarter | Last half-year | Last year | Share in 3 years at its half-year rate |
|---|---|---|---|---|---|
| Rest of the company | 54% | +6.0% | +7.7% | +11.2% | 38% |
| Optical Communications | 46% | +32.3% | +34.1% | +31.4% | 62% |
| Whole company | 100% | +16.6% | +18.3% | +19.0% | 100% |
On the filed split through 2026-06-30, if each business keeps its last half-year's growth, the company grows about 20% next year and 23% in year 3, against 19.4% over the last twelve months. Optical Communications carries the case: it grows 34% against 8% for the rest, and would be 62% of revenue by then. A scenario, not a forecast: no business holds its rate.
How we read it: The business carrying the case: Optical's 34% is 52% of segment profit and enterprise grew 65%, while the rest grew 8% with Solar's ramp inside it. One business, one end market: 3, not 4.
The results
The numbers
Sales grew 19% over the twelve months to Jun 2026, about the same pace as the 19% in the year to Dec 2025.
Operating margin is 15%, up from 7.1% in the year to Dec 2023.
Running costs take 21¢ of each sales dollar, about where it was (21¢) in the year to Dec 2021.
Each dollar put into the business earns 11% a year after tax, about where it was (8.3%) in the year to Dec 2021.
The business brought in $3.92B of cash from operations, 23% of sales, before capital spending, which the filer does not tag.
It owes $5.92B more than the cash it holds, about 1.5 years of the cash its operations bring in.
The share count rose 3.7% in 4.5 years, and pay in shares costs 2.5% of sales.
All the numbers
Filed figures from the company's 10-K and 10-Q, read 2026-10-04. Fiscal years end in December.
| Dec 2021 | Dec 2022 | Dec 2023 | Dec 2024 | Dec 2025 | 12 mo to Jun 2026 | |
|---|---|---|---|---|---|---|
| Size and growth | ||||||
| Sales | $14.1B | $14.2B | $12.6B | $13.1B | $15.6B | $17.0B |
| Sales growth | +25% | +0.8% | −11% | +4.2% | +19% | +19% |
| Profit | ||||||
| Gross margin | 36% | 32% | 31% | 33% | 36% | 36% |
| Operating margin | 15% | 10% | 7.1% | 8.7% | 15% | 15% |
| Net margin | 14% | 9.3% | 4.6% | 3.9% | 10% | 11% |
| What growth costs, per sales dollar | ||||||
| Research and development | 5.7% | 6.3% | 7.1% | 6.1% | 5.1% | - |
| Selling, general and admin | 13% | 13% | 15% | 15% | 14% | 14% |
| Pay in shares | 1.3% | 1.2% | 1.7% | 2.1% | 1.8% | 2.5% |
| Cash | ||||||
| Cash from operations | $3.41B | $2.62B | $2.00B | $1.94B | $2.69B | $3.92B |
| Balance sheet | ||||||
| Cash and short-term investments | - | - | $1.78B | $1.77B | $1.53B | $2.50B |
| Debt | $7.04B | $6.91B | $7.53B | $7.21B | $8.43B | $8.42B |
| Net cash | - | - | −$5.75B | −$5.44B | −$6.91B | −$5.92B |
| Returns and shares | ||||||
| Return on capital | 8.3% | 5.8% | 3.7% | 4.5% | 10% | 11% |
| Shares (diluted) | 844M | 857M | 859M | 869M | 871M | 875M |
What growth costs
Corning's research, selling and office costs barely move with sales, so when its factories fill up each extra sale is more profitable, as the last two years showed. It works in reverse too: in the 2023 slump the same costs squeezed profit hard, and today's margin is only back where it was in 2021. Stock pay and factory spending are growing faster than sales.
The evidence under this - 2 claims
The leverage runs both ways: in 2023 sales fell 11% and the same overheads halved operating margin from 2021's level, so the recent widening is mostly a recovery to 2021's margin rather than a new high.
Counting the $285M of tax paid to net-settle stock awards, $613M went to offsetting stock pay against $777M granted, so the gap is smaller than buybacks alone suggest; and the dividend of about $1B a year was held through the 2023 trough.
The last report (28 Jul)
Quarter ended June 2026, filed figures.
The September quarter beat forecasts, with sales on the company's adjusted basis of $5.08B against $5.03B expected and adjusted earnings of $0.91 a share. Fibre sales rose 44% and data-centre network sales 84%, and the adjusted operating margin rose to 22.7% from 20.9%.
Solar still lost $7M on $436M of sales, and the cash from customer deposits fell by $70M. On the call, managers said Solar improves slowly through 2027, deposits are repaid as customers take fibre, and a $2B share-sale plan is unused housekeeping.
What it did to the thesisIt supported the demand and margin side and left the price side open. The call explained the weak spots without a number to change our view, and the shares remain well above where the plan looks paid for.
The quarter in detail
Sales growth from the previous quarter, filed figures.
Earnings per share $0.78 against $0.76 expected; $0.64 as filed, the gap being what the company adds back.
The evidence under this - 5 claims
Solar still lost $7M on $436M of sales (C58) and contract liabilities fell in the quarter; management's account of both is a description, not a figure, and C09 and C57 argue scale-up optics may stay out of the rack.
What management did with the money
Corning bought back no shares in the first half of 2026, against $133M a year earlier, while stock pay doubled to $250M. On the day this thesis was written it also arranged to sell up to $2.0B of new shares, about 1.5% of the company.
Insiders in 2026Q2: 47 purchases and 60 sales, from their Form 4 filings.
The evidence under this - 2 claims
- Nothing has to be sold: the agreement sets no minimum, the company sets price, time and size limits, and the listed uses of the money include buying shares back and paying dividends, so the program can sit largely unused.
- The $3.0 billion of buyback authorisation that went unused in the first half (C32) is still in place, so the two programs point in opposite directions and neither says which one management intends to use.
- 2026-09-26editedRe-struck at the 25 September 2026 close on the refreshed pack (S28): about 12.8 million shares, still 1.5%.
The price, and where we disagree
In short. Full. The demand for Corning's fibre is real, but the price already pays for its whole 2030 sales plan before any of it is earned, leaving no room for a slip.
The shares returned 96.1% in a year and sit well below their high. They fell on each of the four results days before this one even though every quarter beat forecasts, and then fell further into the 27 October print: the market sells Corning on its next-quarter guidance.
The guide that came with the September quarter is for sales of $5.2B to $5.3B, and the Springboard update is due with it in January.
The evidence under this - 4 claims
- 2026-09-26editedRe-struck on the refreshed pack (S28) at the 25 September 2026 close; the window rolled forward, so the one-year return fell from 121.7% and the distance from the high widened from 38.9%.
- Three of the four falls reversed inside a month: twenty sessions after the January and April prints the shares were 37.0% and 13.6% above the pre-print close, so the day's reaction is not the market's settled verdict.
- The July fall came after the shares had already dropped 44% from the 29 June high of $255.69, so part of it is the end of a run rather than a reaction to the print.
- January 2026 is the exception in kind: Barron's headline says the shares could not extend the momentum of the Meta agreement announced that morning, which is an objection to the price rather than to the guide.
- The April reason given was consumer electronics, a Glass Innovations drag, not optical - so the shared thread is the next quarter's guide, not one business.
How we read the multiple
Buyers pay about $71 for each dollar of last year's profit, against about $46 usually. On Corning's own adjusted profit, which leaves out some costs, it is about $55.
The price looks reasonable only if the $40B sales plan for 2030 arrives in full:
- Fibre sales must grow about 31% a year for four and a half years.
- At today's profit on each dollar of sales, that earns about $7.00 a share in 2030.
- Today's price is already about 22 times that profit, four and a half years early.
Against its own history. Far dearer than its own normal. The business is growing faster than it used to, but the gap prices in years of that growth before any of it is filed.
The evidence under this - 4 claims
- Trailing earnings are the wrong denominator for a business management says will double: the vendor analyst panel's average 2028 estimate is $5.98 a share, which puts the same price at 26.2 times, and the panel's 2029 average at $6.40 puts it at 24.5 times.
- The GAAP figure is depressed by a genuinely non-cash currency item, and the core figure is inflated by the adjustment for it, so neither number is the economics - which is why the price is also stated against 2030 in C29.
- The shares are 42.3% below their 52-week high, so the multiple is already well down from where the market was willing to pay.
- 2026-09-26editedRe-struck on the research pack refreshed at the 2026-09-26 review (S28) at the 25 September close of $156.74; the earnings per share are unchanged, so only the price moved.
- 2026-10-27editedLeft as the dated 25 September statement it is; C59 re-strikes the price at the 26 October close. refresh_by moved to C59.
- Management says earnings will grow faster than sales and that it expects to be at or above its 20% operating margin target, so holding the margin flat at 20.9% may understate the 2030 figure; a 24% margin would give about $8.10 a share and 19.4 times.
- The calculation uses core rather than GAAP profit, which accepts the company's own currency adjustment - the same adjustment this thesis argues is not earnings.
- It also assumes the $40 billion arrives, which is the assumption the whole case turns on and is graded sensitivity: high.
- 2026-09-26editedRe-struck at the 25 September 2026 close ($156.74, S28): about 22 times the 2030 figure, down from about 24; the 2030 arithmetic itself is unchanged.
- The target has been raised twice in five months and its unit changed with it: a January 2026 upgrade to '$11 billion of incremental annualised sales by the end of 2028' became, in May 2026, a '$30 billion run rate by the end of 2028'. A plan restated in a different unit cannot be checked against its own prior version.
- Management declined to put a revenue figure on photonics, the platform the later years depend on, and Weeks said explicitly that the plan is built by 'adjusting for different probabilities of different timings and different content opportunities' - a probability-weighted internal plan, not a guide.
- Corning cannot reconcile any of it to GAAP: the release states that it is 'unable to provide outlook information on a GAAP basis'.
- Holding three businesses flat for four years is an assumption, not a forecast: Glass Innovations carries the display cycle and Corning says it sells more content per device each year, so a flat line there is conservative and pushes the required optical figure up.
- Optical grew 35% in FY2025 and 32% in Q2 2026, so 31% a year is a continuation rather than an acceleration - the arithmetic is demanding because of the duration, not the rate.
- Part of the gap may be closed by businesses that do not exist yet; management describes photonics as a $10 billion platform that would sit inside or beside Optical Communications and is not in the Q2 2026 base.
Priced against its rivals
Only Amphenol sells the same thing to the same buyers as a large part of what it does. A price comparison needs at least two, so the price is judged against its own history instead.
Private or foreign rivals that cannot be priced: Fujikura Ltd. and its subsidiary America Fujikura Ltd. (AFL), Sumitomo Electric Industries, Ltd., Prysmian Group S.p.A., AGC Inc., Nippon Electric Glass Co., Ltd., Schott AG, Heraeus, JENOPTIK Industrial Metrology Germany GmbH, NGK Insulators, Ltd., Ibiden Co., Ltd., LENS (Lens Technology Co., Ltd.), Greiner AG, Eppendorf SE, Sarstedt AG & Co. KG.
The bet: what it pays if it works
This bet is on one business only: the photonics platform, light-based links inside the server rack. Management calls it a $10 billion platform but gives no sales figure, and NVIDIA's flagship rack still uses copper there. Nothing we have defends a figure within two to three years.
The rest of Corning is counted separately and earns all of today's profit, so losing photonics costs no current earnings, only the part of the price paid for it.
The steps it has to take
- Done NVIDIA, whose rack the scale-up link would enter, prepays $1 billion for supply and buys $500 million of stock
- Not started Management gives a photonics revenue figure for 2027 due 2027-01-31
- Not started NVIDIA's rack documentation specifies an optical scale-up link in place of the copper backplane due 2027-12
- Not started Photonics revenue separately quantified in the accounts due 2027-12
The detail behind the label
The evidence under this - 3 claims
- He explicitly refused to guide it: 'I don't think we guided anything for next year on Photonics, just to be clear... We'll update sort of each quarter as we go' - so the figure is a platform ambition, not a plan component anyone can check.
- The in-rack link the platform targets is passive copper in the customer's own current documentation (C08), so the displacement has not begun where it matters most.
- Photonics appears nowhere in the segment note, the revenue disaggregation table or the 10-Q's management discussion: there is no line item to measure it against.
Where we disagree with the market
Management's $40B plan for 2030 is the base case: the analysts' average forecast matches it exactly.
The demand is real and largely contracted, but the price already pays for it, and the biggest customer was paid in shares to provide it.
The costs sit in the quarterly filing, not the results release: the warrant reduces future sales, and $880M of the quarter's cash was deposits and incentives.
The evidence against us: Corning has beaten forecasts eight quarters running. If data-centre sales keep growing above 30%, photonics gets its own line, and margins pass 22% without new deposits, we are wrong.
What analysts say, and the evidence
The evidence under this - 6 claims
- He explicitly refused to guide it: 'I don't think we guided anything for next year on Photonics, just to be clear... We'll update sort of each quarter as we go' - so the figure is a platform ambition, not a plan component anyone can check.
- The in-rack link the platform targets is passive copper in the customer's own current documentation (C08), so the displacement has not begun where it matters most.
- Photonics appears nowhere in the segment note, the revenue disaggregation table or the 10-Q's management discussion: there is no line item to measure it against.
What could go wrong, and how we would know
The risks
- Copper holds inside the rack. If NVIDIA's racks stay copper, the $10B photonics business management describes has nowhere to grow.
- Somebody else's budget. Demand rests on data-centre spending that buyers set each year and can cut; Amazon's has already turned its free cash flow negative.
- A few buyers. Two end customers were 28% of fibre sales in 2025.
- Losing businesses. Solar lost $7M on $438M of sales in the June quarter, and lab products lost $21M.
- More shares. The $2.0B share sale and the 15-million-share warrant both shrink existing holders' slice.
The evidence under this - 7 claims
- He explicitly refused to guide it: 'I don't think we guided anything for next year on Photonics, just to be clear... We'll update sort of each quarter as we go' - so the figure is a platform ambition, not a plan component anyone can check.
- The in-rack link the platform targets is passive copper in the customer's own current documentation (C08), so the displacement has not begun where it matters most.
- Photonics appears nowhere in the segment note, the revenue disaggregation table or the 10-Q's management discussion: there is no line item to measure it against.
- Nothing has to be sold: the agreement sets no minimum, the company sets price, time and size limits, and the listed uses of the money include buying shares back and paying dividends, so the program can sit largely unused.
- The $3.0 billion of buyback authorisation that went unused in the first half (C32) is still in place, so the two programs point in opposite directions and neither says which one management intends to use.
- 2026-09-26editedRe-struck at the 25 September 2026 close on the refreshed pack (S28): about 12.8 million shares, still 1.5%.
What we still do not know
- How much photonics sells. Management declined to put any revenue figure on it, even for 2027.
- How the new currency adjustment would have changed past years. Corning says only that growth would have looked higher.
The evidence under this - 3 claims
- He explicitly refused to guide it: 'I don't think we guided anything for next year on Photonics, just to be clear... We'll update sort of each quarter as we go' - so the figure is a platform ambition, not a plan component anyone can check.
- The in-rack link the platform targets is passive copper in the customer's own current documentation (C08), so the displacement has not begun where it matters most.
- Photonics appears nowhere in the segment note, the revenue disaggregation table or the 10-Q's management discussion: there is no line item to measure it against.
- It cannot be checked: Corning gives the direction and no figure, and says it does not recast, so no reader can reproduce the like-for-like growth rate from the filing.
- The direction is the flattering one - it says the reported 17% and 30% understate the underlying comparison - which is the direction a company has least incentive to overstate and most incentive to mention.
- The same disclosure says the effect on segments other than Glass Innovations would have been immaterial, which bounds it: the definitional change bites on the yen-exposed business, not on the optical growth the case rests on.
What would change our view
- Unattractive→AttractiveThe plan converts, at a price not already paid for 2030
It needs all of:
- ○Fibre sales keep growing above 15% a year with the adjusted operating margin above 19%, so the plan shows in filed profit.checked at the next results, due by 27 Jan
- ○The shares fall to about 38 times filed profit, roughly 12 times the plan's 2030 earnings; 58.33 times on 26 October.about $83.22 today, 46% below the last close
- Reopens the researchThe photonics business becomes visible at all
It needs:
- ○Photonics sales are reported on their own line, and any new capacity comes with a named multi-year supply deal behind it.checked at the next results, due by 27 Jan
What is coming
- 27 October 2026 Q3 2026 results: Core operating margin against Q2's 20.9%, Solar back to profit, enterprise network growth above 30% and contract liabilities still rising, and the Q4 guide.
Every dated event ahead
- 27 October 2026 Q3 2026 results: Core operating margin against Q2's 20.9%, Solar back to profit, enterprise network growth above 30% and contract liabilities still rising, and the Q4 guide.
- November 2026 Q3 2026 Form 10-Q: How many shares were sold under the $2.0B at-the-market program, whether buybacks resumed, and how much of the NVIDIA warrant has been charged against revenue.
- January 2027 Q4 2026 results and the Springboard update: Whether the $20B annualised sales target for the end of 2026 was reached and whether the 2028 and 2030 plan is restated again or in a new unit.
- March 2027 2027 proxy statement: Whether the 2026 pay metrics keep paying on core figures that now include the hedged-exposure adjustment, and whether insider ownership changes.
Since we wrote this
- 27 OctEarningsCorning beat on margin and fibre; Solar still lost moneyCore margin hit 22.7% and enterprise networks grew 84%. Solar stayed in loss, deposits fell, and the shares are still 58 times filed profit, so the view holds.
- 5 OctThesis reviewThe bet now names what it ridesThe owner asked that every bet name the macro trend behind it, after BWXT's bet was missing its chief executive's case for nuclear over gas.
- 4 OctThesis reviewOwner's checklist: revenue streams, cost of growth, management, catalystsThe owner's checklist found the method silent on who runs the company, what growth costs, where the money comes from and what is dated ahead; this pass fills those four blocks from the proxy and filings.
- 3 OctNewsFirst Solar sues a Corning solar subsidiary and JA Solar over patents. View heldFirst Solar sued Corning's American Panel Solutions and JA Solar for patent infringement.
- 29 SepNewsAT&T named as a new multi-year Corning fibre buyer, worth over $3 billion. View heldAT&T signed a multi-year fibre and cable supply agreement worth more than $3 billion with Corning.
- 26 SepThesis reviewOutside evidence of a fibre shortage added. View heldAn independent price reporter says American fibre plants run flat out, and the shares fell on the guidance at three of the last four results, so October's guide matters as much as the quarter.
- 26 SepPrice movePrice figures moved to the September 25 close; view heldThe shares are 5.6% below where the valuation was first struck. At 71.57 times filed profit, and about 22 times what the 2030 plan would earn, the price still pays for the plan in advance.
- 24 SepNewsA $2.0B share-sale programme found. View heldCorning signed it the day the thesis was written, and the thesis had missed it. It is about 1.5% of the shares, none sold yet: it deepens the dilution objection without moving the view.
- 11 SepThesis writtenUnattractive, medium convictionThe fibre demand is real and contracted, but the price already pays for management's whole 2030 plan.
Also checked 2 times with nothing new to act on.
Sources 33
Sources - 33
Every claim above cites one of these, or states the reasoning that derived it. The quote is frozen at the date it was read: a link can rot, and an agent fetch returns a reading of a page rather than the page. Backs counts the whole claim ledger, so it can name a claim this page does not surface.
- S01primarycompany filing2026-07-28GAAP and core income statement, the full GAAP-to-core reconciliation line by line, segment tables, cash-flow statement, the Springboard run-rate targets, the Amazon and NVIDIA partnership descriptions, Q3 2026 guidancebacksC01C02C03C16C20C21C22C23C24C25C26C33C34C42
- S02primarycompany filing2026-07-29Note 2 (the $1.0 billion deposit to 2029 and the $296 million warrant charged against future revenue), Note 12 (the warrant terms), Note 13 (share count), Note 14 (the segment restructuring and the hedged-exposure basis), contract liabilities, the equity statementbacksC01C12C13C14C20C22C23C24C32C42C43C50C54C55
- S03primarytranscript2026-07-28Q2 2026 earnings call transcriptmanagement's answers on optical margin, long-term agreement coverage and pricing; the $10 billion photonics platform and the refusal to guide it; eight analyst questions, of which the sharpest were Wamsi Mohan's on the absence of operating leverage from a $300 million sequential enterprise gain and George Notter's on long-term agreement coveragebacksC26C34C40C41
- S04primarycompany filing2026-02-12Item 1 Competition by segment, the customer concentration table, the patent estate, the government-incentive accounting policy, the debt maturity schedule, the solar risk factors, and Exhibit 10.33 (the Solar Technology LLC lease and construction-agent amendment guaranteed by Corning)backsC04C25C31C33C53C54C55
- S05primarycompany filing2026-01-28the FY2025 figures, the original $8-billion Springboard plan and its January 2026 upgrade to $11 billion, the old constant-currency definition, and the Meta agreement of up to $6 billion - none of which appears in any quarterly filingbacksC15C34C35
- S06primarycompany filing2026-05-06the date and existence of the NVIDIA securities purchase agreement, which ties the unnamed customer in the 10-Q to the partnership named in the results releasebacksC12
- S07primarycompany filing2026-02-11a named rival's own account of the same market: the $10.5 billion CommScope purchase completed 9 January 2026, IT datacom at about 36% of 2025 sales and up about $4,593.7 million in the year, and the approximately $500 billion worldwide interconnect estimatebacksC05C11
- S08primarycompany filing2026-07-31a named customer's own capital spending: trailing-twelve-month purchases of property and equipment of $169.0 billion against $103.0 billion, and free cash flow of negative $7.6 billion against positive $18.2 billionbacksC18
- S09primarycompany release2026-09-08Verizon newsroom - Verizon and Corning announce fiber agreement Verizon Communicationsthe buyer's own statement of quantity, term and purpose - the only contracted volume-and-term pair on the roster of Corning agreements, and the base factor in the prize
Verizon and Corning have reached a multi-billion dollar agreement for 80+ million miles of high-density optical fiber and connectivity solutions from 2027 to 2032. ... This fiber supply accelerates Verizon's push toward 40-50 million broadband passings.
backsC06C10 - S10primarycompany filing2026-06-09the transaction codes behind the largest insider sale in the window - M then S, an option exercise at $27.03 and a same-day sale at a weighted average $186.4607 - and the 908,353 shares still held directlybacksC36
- S11primarygovernment data2026-05-19EIA - Data center server energy use grows across the commercial building stock U.S. Energy Information Administrationthe breadth of the driver measured by somebody outside the industry, and the reason the near-term rate was taken from Amazon's filings instead - this is a projection, not a current rate
Across all cases, servers alone accounted for an estimated 7% of commercial sector electricity consumption in 2025. Data center server electricity use grows to 22%-33% of commercial building electricity use by 2050 across our cases.
backsC19 - S12secondaryvendor data2026-09-11GLW research pack, frozen at content/GLW/research-pack.jsonprice, market value, the 252-day return and the 52-week range, the volatility bucket, the analyst split and estimates, the earnings surprise history, FINRA short interest at the 31 August 2026 settlement, and the prior-fiscal-year median multiplesbacksC35C36C38
- S15primarycompany release2026-09-11NVIDIA DGX Grace Blackwell hardware documentation (undated page, retrieved 11 September 2026) NVIDIAwhat the in-rack link is actually made of today, from the vendor that defines the rack - the single most important fact against the photonics case
DGX Grace Blackwell rack scale systems are rack scale solutions for graphics processing units (GPUs) connected by NVLink through the NVLink passive copper cable cartridge backplane.
backsC08 - S16primarycompany filing2026-09-11the at-the-market program: up to $2,000,000,000 of common stock through Goldman Sachs as agent or principal, a 1.0% commission, the company setting price, time and size limits, and the April 2026 automatic shelf it is sold underbacksC44
- S17primarycompany filing2026-09-11the use of proceeds (general corporate purposes, which may include share repurchases, dividends, debt repayment, acquisitions, working capital and capital expenditure), no minimum offering amount, and 861,388,331 shares outstanding at 24 July 2026backsC44
- S18primarycompany filing2026-09-11the agreement behind C44: common stock 'Having an Aggregate Offering Price of up to $2,000,000,000' through Goldman Sachs as agent or principal, sold under the Form S-3 shelf 333-295316, with no forward-sale mechanism in itread, but no claim rests on it
- S19primarycompany filing2026-09-11counsel's opinion that the shares will be validly issued; boilerplate, read so the audit's reading list is closed, and it changes nothingread, but no claim rests on it
- S20secondaryvendor data2026-09-26FMP end-of-day prices and earnings calendar for GLW, September 2023 to 25 September 2026the closing price either side of each results day from October 2024 to July 2026, the consensus and actual adjusted earnings per share for each print, and the 29 June 2026 high of $255.69backsC45
- S21secondarynews2025-10-28the objection stated on the day of the Q3 2025 print, read from the article's summary in the FMP news feed; reuters.com refuses the fetch tool
Gorilla Glass maker Corning on Tuesday forecast fourth-quarter sales above analysts' estimates, but revenue from its largest business unit that makes fiber optic products fell short of expectations.
backsC46 - S22secondarynews2026-04-28the objection stated on the day of the Q1 2026 print, read from the article's summary in the FMP news feed
Corning forecast second-quarter revenue below Wall Street estimates on Tuesday, as weakness in the consumer electronics market continues to pressure business despite strong demand for data center products.
backsC46 - S23secondarynews2026-07-28the objection stated on the day of the Q2 2026 print; CNBC's same-day piece (403 to the fetch tool) records it as the worst day since 30 July 2002
Corning surpasses expectations with its second-quarter earnings, but the stock falls on in-line guidance.
backsC46 - S24primaryindustry data2026-02-26an independent price reporter's reading that fibre capacity is slow to add, and that Chinese bare-fibre prices rose more than 80% between November 2025 and January 2026
Capacity cannot be restored immediately, and meaningful additions are likely to take months rather than weeks.
backsC47 - S25primaryindustry data2026-05-27CRU Group - CRU introduces monthly optical fibre price reporting in response to market volatility CRU Groupthe moat's mechanism observed from outside: United States preform and draw capacity at full utilisation, and prices rising on it
In the US, even the structurally high prices of BABA compliant optical fibre moved higher, driven by acute fibre shortages amid growing hyperscaler demand for AI data centres, coupled with the limited domestic glass preform and drawing capacity already running at full utilisation.
backsC47 - S26primaryindustry data2025-12-17Fiber Broadband Association and RVA LLC - 2025 North American FTTH Deployment and Market Update Fiber Broadband Association84.6 million United States homes with a fibre service marketed, 11.8 million added in the year, 116.5 million broadband serviceable locations and 60.2 million estimated first-unit passings remaining; and a five-year forecast of 92,000 route miles and 213 million fibre miles of data-centre outside interconnect. Read from the PDF text; fiberbroadband.org refuses curl with 403
Five-Year Outside Interconnect Forecast: 92K Route Miles 213M Fiber Miles ... Number of homes with at least one fiber service marketed (excludes redundant fiber services available to the same home)
backsC48C49 - S27primaryindustry data2026-07International Data Center Authority - Global Energy Report (2026) International Data Center Authoritythe only outside quantity found for what delayed data-centre capacity costs its owner - measured for grid-connection delay, not for fibre
The average lost revenue cost of these delays is over $10,000 per megawatt, per day.
read, but no claim rests on it - S28secondaryvendor data2026-09-26GLW research pack, refreshed at the 2026-09-26 review and frozen at content/GLW/research-pack.json (market tier at the 2026-09-25 close)content/GLW/research-pack.jsonthe 25 September 2026 close, the 252-day return and 52-week range, the vendor's trailing multiples and prior-fiscal-year medians, the analyst split and estimates, the Form 4 roster from 30 September 2025, FINRA short interest at the 15 September 2026 settlement, and the filed quarterly series and filed diluted earnings per share through the June 2026 quarterbacksC36C37C39
- S29primarycompany filing2026-03-20beneficial ownership at 31 December 2025, the 2025 pay plan's metrics and weights (PIP, GoalSharing, CPUs, PSUs, RSUs, the ROIC modifier) and the 2025 and 2023-2025 payoutsbacksC51C52
- S30primarycompany filing2026-10-04Corning's own XBRL facts (SEC company facts), frozen at content/GLW/financials.jsoncontent/GLW/financials.jsonFY2020 to FY2025 and the four quarters to June 2026: net sales, operating income, the ResearchAndDevelopmentExpense tag, SG&A, stock pay and diluted sharesbacksC54C55
- S31primaryindustry data2025-12-31an independent analyst's 2030 value and port count for co-packaged optics, the 2027 start of scale-up shipments, and that current co-packaged products are scale-out switches only; read through ir_fetch
We expect that both Broadcom and Nvidia will introduce scale-up switches, GPUs or XPUs with CPO in 2026. Marvell plans to have it as well, using Celestial AI’s technology. Shipments of these products will start in 2027. The market value for CPO engines, including scale-up and scale-out applications, is now projected to reach $10 billion with close to 100 million of CPO ports shipped in 2030.
backsC56 - S32primaryindustry data2026-05-31the adoption assumption under LightCounting's forecast, its base case of copper inside the rack, in-rack optics named as upside, and optics sales of almost $80 billion by 2031; read through ir_fetch
this forecast is based on an assumption that only 10-15% of all scale-up interconnects will migrate to NPO/CPO by 2030-2031. We also expect that the scale-up networks will continue to rely on copper interconnects within the racks and the optical interconnects will be only used for multi-rack connections.
backsC57 - S33primarycompany filing2026-10-27GAAP and core income statement, segment tables, balance sheet at 30 September 2026, cash-flow statement, adjusted free cash flow reconciliation and the fourth-quarter outlookbacksC58
- S34secondaryvendor data2026-10-27the 26 October 2026 close of $127.74, an observed traderead, but no claim rests on it
- S35primarytranscript2026-10-27management's account of the Q3 split signals: margin, Solar, deposits and funding, growth plan, the equity shelf and Chinese fibre
We said in the May IR event we expected optical scale-up to begin to accelerate in 2028. We continue to believe that strongly.
backsC14C60