Every conversation about AI infrastructure eventually turns to power and computer chips. Fewer turn to fiber-optic cable, the glass wiring connecting thousands of AI chips to each other and linking data centers to the outside world.

That's a mistake. Fiber has quietly become one of the tightest constraints in the AI buildout, and most companies aren't managing it well.

See also: As AI adoption in manufacturing matures, data center controversies move to the fore

An AI-optimized data center needs far more fiber than a traditional one, since AI chips must be wired together more densely to work as a single system. Demand has grown sharply over the past two years, prices have exploded, and lead times have stretched out.

Some of the largest tech companies are now signing multibillion-dollar supply contracts directly with fiber manufacturers. This industry has never seen such engagement and with a supply chain built for steady demand, the sudden shock is exposing planning gaps that need to be confronted.

Structural mismatch, not a temporary squeeze

Fiber manufacturing is precise and capital-intensive. Building new capacity typically takes 18 to 24 months from start to finish, so money committed today won't meaningfully ease the market until 2027 at the earliest. The constraint isn't capital; it's time and manufacturing physics.

That timeline puts suppliers in a hard spot. Major forecasts, including Dell'Oro Group's projection of $1.7 trillion in global data center capital spending by 2030 and Goldman Sachs' estimate of roughly $5.3 trillion in hyperscaler capital spending through 2030, point to years of continued growth rather than a short spike.

See also: Industrial IoT starts with the physical layer: Rethinking cables, conductors and connectivity

But there's a real, live debate about overbuild risk. No one can say with confidence whether demand plateaus for years or partially reverses. Suppliers are being asked to commit hundreds of millions of dollars to an 18- to-24-month build cycle despite uncertain demand and limited planning transparency.

That uncertainty is not just a forecasting problem. It is compounded by several structural weaknesses in how customers and suppliers plan together.

Where the planning breaks down

A few issues make the underlying mismatch even harder to manage:

Inconsistent planning across hyperscalers. If you ask five hyperscalers for their 12-month build plan, you'll get five different answers, not just in the numbers, but in how much detail sits behind them, whether they specify location, timing, and technical requirements clearly enough for a supplier to make a rough-cut fiber acquisition and manufacturing capacity decision.

Some plans support that. Others don't come close. That inconsistency makes it hard for a supplier to build one coherent view of demand.

Project-based demand vs. standard product economics. Data center builds are large, one-off, schedule-driven projects, while fiber manufacturing runs on standard lines built for steady output.

Reconciling “this exact spec, this volume, by this date” with steady-state production creates friction and costly expediting. It cuts both ways: Lines also can sit idle waiting on a final order, then absorb all the demand at once, straining labor and cost.

Moving technical requirements and constrained allocation. The specific fiber types customers need are still evolving, so suppliers building capacity against a moving target risk having the wrong capability once it's built.

See also: What to do when ‘rip-and-replace’ is financially impossible

Meanwhile, every unit of fiber committed to a hyperscale AI contract is one not available to a broadband, telecom, or industrial customer, and that competition now reaches beyond civilian markets: Fiber-guided drones used in current military conflicts draw on the same bend-insensitive fiber grade, and the same upstream preform and drawing capacity, that data centers need, even though the finished products look nothing alike.

Manual, labor-intensive processes. Much of cable manufacturing and termination is still done by hand. Automating it would help absorb demand, but that shift is itself a significant undertaking layered on everything else straining the supply chain.

Individually, none of these is fatal. Together, however, they mean customers and suppliers are not working from the same view of real demand, which is exactly the gap collaborative planning is built to close.

The fix Is collaborative planning, not just more capacity

None of this is solved by investment alone. It gets solved by better decisions, made jointly, from a shared set of facts. Integrated business planning, linked to integrated tactical planning, is designed to close this gap.

IBP gives an organization a shared, forward-looking view of demand, supply, and financial trade-offs at the strategic level while ITP turns that view into daily and weekly execution decisions.

Internal discipline must come first. A company can't extend collaborative planning to its customers or suppliers if its own IBP and ITP aren't already working well. The strategic and execution layers need to be linked inside the company before they're worth linking across companies.

See also: Smart industrial innovation is commoditizing faster than you think

Once that foundation is in place, the same approach can extend outward. A data center customer and its fiber supplier can then work from the same demand signal, the same capacity constraints, and the same view of risk, rather than discovering the gap only after it has already become a problem.

Done well, this allows both sides to do things they currently struggle to do: remove duplicated demand from forecasts, make explicit trade-off decisions on allocation instead of ad hoc calls, time capacity investment against an honest view of how long demand will hold, align technical roadmaps to align new specs and new manufacturing capabilities, and give sales teams one accurate picture of what can actually be committed.

The takeaway

Compute and power get the headlines. Fiber is a quieter constraint and may be more exposed: New capacity takes longer to build than the industry’s confidence in how long today’s demand will last. Forecasters disagree, and suppliers can't wait for that to resolve before they build.

See also: Smart industrial innovation is commoditizing faster than you think

The answer isn't a bigger bet that the boom continues. It's building real planning discipline inside each company first, then linking it across the customer-supplier relationship, so both sides can make timely, shared decisions on service, cost, cash, and profitability as the picture changes.

Whichever way demand moves, the companies with that discipline in place will be the ones able to adjust without getting stranded.