Pinned Tweet
Bottleneck Hunt #5: $GLW
Every AI data centre is basically just a massive amount of fibre. The GPUs only work as one computer if they're all wired together, and there's not enough fibre to go round.
TrendForce reckons the world is about 180 million fibre-km short this year, plants are running at 92% and stockpiles have gone from 4 months in 2024 to 1 month. It also takes 18 to 24 months to build new capacity so it's not getting fixed any time soon.
Corning is one of the largest names within it. They've been making glass for over 100 years and now have multi-billion deals with Meta (up to $6bn through 2030), Amazon, Nvidia, Verizon and AT&T.
Bull case is the numbers are really good. Q3 sales were up 19% to $5.08bn, optical was up 44% to $2.38bn and AI product sales more than doubled. Margins are up 3.1 points to 22.7%, and customers are literally paying cash upfront to fund the new capacity. They're aiming for $40bn a year in sales by 2030.
Bear case is it's not a monopoly like BWXT. Fujikura is spending $1.9bn to triple its US production. Corning is also spending about $2bn on capex this year, the glass side is only growing 1% and the price already assumes the plan works. It dropped 12% after beating in Q2 and dipped again after Q3. It's about $114 now, which is roughly 26% down in a month.
My verdict is the shortage is actually visible but Corning is one of a few that can make it at scale, not the only one. I like the business more than I trust the price right now so it's going on the watchlist.
356
If I had to build a portfolio only out of bottlenecks, here’s what I’d pick.
$ASML, the only company in the world that makes EUV machines, so every leading edge chip goes through them.
$HWM, which makes the turbine blades, and customers already see its capacity as sold out to 2031.
$LEU, which runs the only licensed HALEU enrichment in the US, the fuel next-gen reactors need.
$PKE, the sole qualified source of a key material for the Patriot PAC-3 missile.
$UMAC, one of the few US makers of drone motors, at a time when the Pentagon needs them built outside China.
Chips, power, nuclear, drones and missiles, this is everywhere I think that the future is heading.
2
218
Government stakes in stocks: green light or red flag?
CASE FOR
The government becomes your customer, your backstop and the reason the permits get signed.
CASE AGAINST
Politics changes every four years, and a shareholder that can also regulate you comes with strings.
It’s not a one-off anymore: Intel, MP Materials and Trilogy Metals all have one now.
With moderation I think it's okay, but when positions get too large emotions can play a larger role.
97
Every vaccine is tested with crab blood. $CRL
THE BOTTLENECK #6
Before a vaccine, an insulin pen or an IV drip goes anywhere near a patient, it has to be tested for bacterial toxins. The FDA-approved standard for that is the LAL test, which is made from horseshoe crab blood, and around 70 million of these tests are done every year.
The crabs are caught, partly drained of their blue blood and released, which is about as niche as a supply chain gets.
WHO MAKES MONEY
Charles River ($CRL) is one of the FDA-licensed LAL makers. It also sells a synthetic version called Trillium that doesn’t need any crab at all.
THE CATCH
The synthetic route now has a proper rulebook. A dedicated recombinant testing chapter was approved in July 2024 and published in 2025. It sits alongside the old LAL standard rather than replacing it, but it means this bottleneck is slowly being engineered away.
MY TAKE
What I like is that Charles River sits on both sides, so whether labs stick with crab blood or switch to the synthetic one, they’re selling the test either way. The risk is that it turns into a commodity once anyone can make the synthetic version.
99
Why I'm not buying Corning yet and what would change my mind.
Optical growth staying above 30%, that $2bn of capex turning into actual free cash flow (Q3 was $661m), and the stock finally not going down on good news.
If those line up I'll look at it again.
89
$BTC is useless.
It produces nothing, pays nothing and needs nothing, so it’s only worth what the next person thinks it’s worth.
The second the market decides it’s worthless, it is.
Gold has thousands of years of being wanted, farmland grows food, and Nvidia sells chips people are queuing for.
Bitcoin has a chart. And hope.
1
156
One number from the Corning research that stuck with me.
An AI data centre needs about 36x more fibre than a normal server rack.
There's a month of inventory left and it takes up to 2 years to add capacity, which is why customers are signing 6 year deals and paying cash upfront.
97
Next Bottleneck Hunt is coming, and I've got a shortlist but I want yours first.
Reply with one company you think is a choke point the market isn't paying attention to, and why.
The best one gets the full breakdown next week with credit to you.
94
Being a monopoly doesn't make something a good investment.
$BWXT is the only company that makes US naval reactors and it trades at about 27x earnings.
$GLW is a real choke point and it's still down 26% in a month.
The bottleneck tells you it's a good business, the price tells you if it's worth buying.
1
328
$NBIS is down about 51% in a month.
My first buy was around $124 and it's at about $116 now.
They posted a massive revenue beat last week, up 68% on the quarter, and it still sold off.
My view on it hasn't changed, but I sized it so a drop like this doesn't hurt.
If a 50% drop would make you panic sell then the position is too big, not the stock.
1
1,091
Nvidia reports tonight and I've held it on and off since $49 so I'm in a weird spot, up over 3x and still nervous.
It's down about 29% in a month, options are pricing roughly a 10% move either way, and they guided $108bn revenue with the Street at about $109bn.
No idea what it’ll do.
I'm just watching whether data centre revenue and gross margin hold up with memory costs going up.
100
Corning beat last month and the stock still went down.
You can be completely right about a shortage and still lose money on the stock, because the price is what decides it.
121
Tomorrow's bottleneck is something every AI data centre needs and the world is about 180 million fibre-km short of this year.
It beat on earnings last month and the stock still fell, so it's a good one to pick apart.
8:30 tomorrow.
149
$NVDA reports on Tuesday.
The first ever report without Mr Huang as the CEO.
What are your thoughts on the outcome?
161
Jensen resigning from $NVDA isn’t bullish for other chip companies.
1. Nothing about demand has changed. Capex comes from budgets, not from who runs these companies.
2. Switching costs are huge and slow. Usually a 12-18 month procurement cycle.
3. It’s a scandal, not a product failure. A +10% move is a huge overreaction.
4. If a regulatory sector wide action was made against china, other chip companies would fall into that too.
Don’t get me wrong this is bullish in the short term for the likes of $AMD and $AVGO, but a total rotation out of Nvidia would be a foolish mistake long term.
166
Jensen Huang has resigned as $NVDA CEO.
Effective immediately.
Kress is interim.
Stock is -12% on key-person risk.
The thesis was never on one man but it sure helped.
CUDA, CoWoS and HBM are still the choke points, and none of them resigned today.
It all relies on November 17th now.
$NBIS Q3: Revenue BEAT 🟢
$980.8M vs $907M expected (+8%)
+68% QoQ | +571% YoY
Revenue ran ahead of the Street, and capacity is still the bottleneck, not demand.
The print that decides the stock is exit ARR vs the $7–9B year-end target.
318
Multiple banks and investors have upgraded and downgraded $NVDA today.
Earnings could be quite a gamble, who’s loading up now?
142




