North American Container Glass: 16 Years Without a New Plant

Sixteen years. That's roughly how long the US container glass industry went without a genuinely new furnace built from the ground up, from the mid-2000s until the Waterloo, New York conversions and the Valdosta, Georgia greenfield started changing the maths in the early 2020s. Sixteen years of the same furnace campaigns, the same three incumbents fighting for the same case volumes, and almost every dollar of improvement having to come from inside plants that were already running.

I worked the hot end through a chunk of that freeze. It reshapes how you think about the job. When nobody is building new capacity, the only lever left on the table is what you do with the capacity you've got. That's a different job to the one plant managers had in the 1990s, when a bad line just got mothballed and replaced. There's no replace button anymore. There's only optimise, or don't.

Why did North America stop building furnaces?

The short answer is beer cans and demand shift. Container glass in the US has spent two decades losing packaging share to aluminium cans and PET, particularly in beer and soft drinks, while wine, spirits, and premium food packaging held glass share reasonably well. That left US demand roughly flat to slightly declining for long stretches, and nobody sinks $150-250M USD into a new furnace campaign against flat demand.

O-I, Ardagh, and Anchor Hocking (plus the newer Arglass Yamamura entrant) have instead spent that capital on furnace rebuilds, efficiency upgrades, and selective rationalisation. Ardagh closed capacity in the mid-2010s. O-I has run repeated network reviews trimming underperforming lines. The result is a smaller, more concentrated fleet than existed in 2005, but one running harder per remaining furnace. Fewer plants, less redundancy, and much less tolerance for a bad changeover, because there's no spare line two states over to shift the order to.

What that freeze does to changeover economics

Here's the bit that doesn't show up in the market reports. When you can't add capacity, every hour of downtime is a bigger percentage of your annual output than it used to be. A plant running at 92% of theoretical capacity in 2006, with three sister plants absorbing overflow, could eat a rough job change without anyone at corporate blinking. That same plant in 2023, running as one of only four or five furnaces left in the network, cannot.

I audited a mid-sized US plant in 2019, three furnaces, mixed narrow-neck press-and-blow and NNPB lines, where cross-shift job change time on an identical SKU swung from 38 minutes on the best crew to 94 minutes on the worst. Same mould set. Same recipe card. The gap wasn't equipment, it was the 0600 handover missing the previous shift's swabbing notes and mould preheat data, which happens on most lines I've audited, not just that one. The night crew re-derived settings from memory instead of a locked recipe, and the first four gob cycles ran out of spec until someone caught it on visual.

That's not a furnace problem. A handover problem. And it's the exact failure mode a systemised Job Change Tool exists to close, because a locked SKU Library with versioned mould and recipe specs means the night-shift guy isn't reconstructing anything from a notebook, he's following the same checklist the day crew used.

USMCA and the Mexico side of the ledger

Mexico doesn't get talked about enough in North American glass coverage, and that's a mistake. Vitro and other Mexican producers run substantial capacity feeding both domestic demand and cross-border US supply under USMCA, and the labour cost differential means Mexican plants absorb a meaningful share of the commodity end of the market, tequila and beer bottles particularly, freeing US plants to lean into higher-margin wine, spirits, and food segments. That division of labour is now baked into how US operators plan network capacity, not an afterthought.

The practical effect on the hot end is that US plants increasingly run higher-mix, higher-value SKU books with more frequent job changes, because premium segments carry more distinct bottle geometries than commodity beer bottles do. More job changes per month means changeover variance compounds faster. A plant doing eight job changes a month can absorb one bad one. A plant doing twenty-two can't.

Nobody budgets for a new furnace on the strength of a good changeover. But nobody survives without one either.

What incumbents are actually optimising for now

Talk to a hot-end superintendent at any of the big three US operators and the conversation isn't about capacity anymore, it's about defect rate and changeover speed on the fleet they've already got. First-ware quality after a job change, time-to-stable-pack, and section-level gob weight variance are the KPIs that move the P&L when you can't build your way out of a problem.

Gob weight CV target of 0.4% or better sounds like a rounding error until you're chasing stones and cords through a shift because the swing on section three drifted past spec and nobody caught it at first ware. I've seen plants where the operator owns the checklist step but the hot-end superintendent owns recipe lock, and when that line of authority gets blurry, that's exactly when someone changes a set point without sign-off and you spend the next six hours explaining a reject spike to the plant manager.

• Furnace campaigns get longer, not shorter, because rebuild capital is precious
• Job change frequency rises as SKU mix shifts toward premium, lower-volume runs
• Labour turnover erodes the tribal knowledge that used to cover for the lack of a written system
• Board-level reporting increasingly demands defensible KPI data, not shift-supervisor anecdotes

Where this leaves plant managers

If you're running a plant in North America container glass right now, you're not waiting for a competitor's furnace to shut down and hand you volume. You're competing on how tightly you run what you've already got. That's a harder job than the one your predecessor had, and it rewards different skills. Discipline over improvisation. Written specs over memory. Trended KPIs over gut feel at the Monday meeting.

I built the Job Change Tool because I got tired of watching good plants leave 30-60% changeover variance on the table simply because nobody had systemised the nine-stage lifecycle from plan through to post-mortem. It's not a rebadged SMED workshop and it's not tied to any OEM's forming equipment. It's vendor-neutral by design, because a plant running Emhart TC on one line and a newer Heye machine on another doesn't need two different changeover philosophies, it needs one that works on both.

Zaid Hassoneh built this from the floor up, starting at O-I Brisbane in 2005 and later running the $220M USD Arglass Yamamura greenfield build in Georgia, and that operator background is why the methodology reads like a shift schedule instead of a slide deck. If you want the fuller picture on North America container glass plant dynamics and where the controllable wins actually sit, that's worth a look before your next capital planning cycle. And if you want to know how that operator background actually shaped the tool, the Arglass Yamamura experience page has the detail.

So the question worth asking at your next ops review isn't whether the market will hand you more volume. It's whether your changeover time on your top five SKUs would survive someone asking for the data in front of the board.