Premium Capacity Is Not a Strategy

Buying a bigger pool of compute does not create a number the room will use.

Capacity is headroom. Strategy is the model, the owner, and the meeting that actually runs on it. Mid-market companies mix those up at renewal. Then they pay for a quieter spinner and the same five margins.

Capacity is infrastructure. It is not the program.

A CEO hears “Premium” or “capacity” and thinks the company bought a reporting system. A CFO sees a line item and hopes the next close will be faster. Neither thing is what the SKU does.

Capacity buys scheduled refresh, query throughput, and isolation from shared tenants. Those are real. They are also empty if the work is five private models, a pack that still pastes, and no steward who can freeze a measure.

This is not the unused-seat problem. Seats are who can open the door. Unused licenses are a corridor nobody walks. Capacity is the size of the corridor. If nobody trusts the destination, a wider hallway is not a strategy.

The vendor catalog will keep renaming SKUs. Do not memorize a ladder. Do not invent a price in the board deck. Ask what load you actually run, and whether that load is the work you meant to fund.

A Fabric vs Power BI conversation is a platform-and-ownership call. Capacity shows up inside it. It is not a substitute for deciding who owns the semantic product. The model is the product. The SKU is the factory floor. Floors do not invent the goods.

What buying headroom without a model costs

1. You paid for RAM and still argue the number. Refresh finishes. The tile is green. Finance and sales still disagree on revenue. Compute did not reconcile grain. It hosted the fight at higher clock speed.

2. Slow pages get a bigger SKU instead of a smaller model. Bloated visuals, bi-directional relationships, and import of every column will choke any pool. Paying up the ladder to avoid a model review is not architecture. It is a habit. Optimization is a different job from strategy.

3. Capacity becomes a cover for sprawl. Every extra dataset is another refresh job competing at 6 a.m. A larger pool lets the sprawl survive another quarter. The tax is still there: dashboard sprawl is not self-service. Headroom hides the inventory problem. It does not retire it.

4. The platform conversation postpones ownership. “We will size capacity after we land the lake” is how a year disappears. Meanwhile close still waits on a file. Delivery still needs a named owner and a cut: analytics programs fail in delivery. A SKU cannot say no to a VP.

5. Shared pain gets treated as a status purchase. Dedicated capacity can be the right isolation. It can also be a prestige buy while three teams still publish competing “official” models. Isolation without a steward is a private mess. Every team building their own model is not fixed by a bigger node.

6. Renewal treats spend as proof of strategy. “We already invested” is not a decision metric. It is sunk cost. The next order form repeats the last one. IT is measured on uptime of a pool. The CFO is still measured on a pack that arrives late.

Look at your refresh window, your concurrent meetings, and your one certified app. If those are undefined, you are shopping. You are not sizing.

What not to do

Do not start with a SKU comparison. Start with the workload: which model must refresh before the stand-up, which app the ELT opens, who signs the measure.

Do not mix license waste with capacity. Turning off quiet seats does not shrink a bloated dataset. Growing a pool does not create viewers who trust the page.

Do not wait for a perfect lakehouse before you fund one semantic product. Capacity should serve that product, not a future architecture slide.

Do not let “the reports are slow” be the only ticket. Slow is a symptom. Grain, visuals, and gateway design are often the cause. Fix those first. Then size.

How to treat capacity as a tool, not a strategy

1. Name the work before you name the SKU. One official semantic model. One app leadership will open. One steward who can freeze grain. If you cannot name those three, you are not ready to size anything. You are ready to assign an owner.

2. Inventory load against the actual calendar. What must be fresh at 7 a.m. on close week. How many concurrent viewers hit the pack in a Monday meeting. Which datasets are ghosts. Size for the jobs that matter. Do not size for the long tail you should retire.

3. Fix the model before you buy headroom. Reduce columns that nobody uses. Kill visuals that query the whole fact table to paint a card. Split a monster import if the grain is really two grains. Then measure. If the page is still late with a sane model, capacity may be the constraint. Until then, assume the work is the problem.

4. Buy isolation for concurrency you can point to. If shared capacity is actually colliding with other tenants, or your refresh window cannot finish before the stand-up, that is a real infrastructure case. Document it. Fund it. Do not fund it as a substitute for certified measures and a steward.

5. Put refresh on the same board as the P&L pack. A pool that fails at 6 a.m. is close risk, not a capacity curiosity. Operations ownership sits next to the steward. Uptime of the wrong model is not a win.

6. Review at renewal like a factory, not a catalog. What decisions did the app serve. Which datasets died. Which refresh jobs still exist because nobody said no. Strategy alignment is the roadmap, not the order form. Ongoing model ownership is Managed Data & AI Advisory.

What a sane capacity picture looks like

A short certified set. Refresh that finishes before the meeting that uses it. Authors who publish against the official model, not a private twin. Viewers on one app.

The CFO does not need a SKU map. The CFO needs Monday’s number to match the close, on time, with a name behind it. Capacity is how that job gets compute. It is not the job.

[Judgment:] if the strategy slide is a capacity diagram, you do not have a strategy. You have a shopping list.

Frequently asked questions

Do we need dedicated capacity to get value from Power BI? Only if shared capacity is the actual constraint—refresh window, isolation, or concurrency you can show. Most stalled programs are ownership and model, not the pool.

Isn’t a larger SKU cheaper than rewriting the model? Rewriting is not the first move. Trimming grain, visuals, and duplicate datasets usually is. Paying for a bigger pool to keep five revenues is the expensive path. It just looks like a clean invoice.

How is this different from unused licenses? Licenses are people. Capacity is compute. You can have full seats and an oversized pool and still have no trusted number. Fix the product. Then match seats and compute to the people and jobs that use it.

Get started

You do not need a bigger pool to look strategic. You need one model the room will sign, sized for the work it actually does.

Need a 30-minute look at capacity versus the model you actually run? Contact Alluvium. We’ll map refresh load, the certified set, and whether the next spend is headroom—or ownership.

Book a 30-minute consult.