- What Did NVIDIA’s Q2 Actually Reveal About Concentration?
- Why Is 92 Percent of Revenue Not the Same as 92 Percent of Resources?
- What Does 80/20 Squared Actually Tell You to Concentrate?
- Why Did NVIDIA Line Up $500 Billion to Build a Long Tail?
- What Happens When Operators Get the Split Backwards?
- How Do You Run the Concentration Split on Your Own P&L?
- What Does This Look Like Inside a Manufacturing Business?
- What Should You Actually Cut This Quarter?
- Frequently Asked Questions
- About the Stagnation Assassin
Every time I teach portfolio concentration to an operating team, the same hand goes up. Someone in finance says the quiet part out loud: if we cut everything except the top slice, are we not building a business that dies the day one customer leaves?
It is a fair question. It is also the wrong question, and on August 26, 2026, the most concentrated large company on earth published the numbers that prove it.
What Did NVIDIA’s Q2 Actually Reveal About Concentration?
NVIDIA reported second quarter fiscal 2027 revenue of $96.2 billion, up 106 percent from a year earlier, with Data Center revenue of $89.0 billion, up 117 percent. Data Center accounted for more than 92 percent of total quarterly revenue. Inside that segment, hyperscale customers delivered $48.7 billion against $40.3 billion from everyone else.
Read those two layers again, because the second one is where the lesson lives. One segment carries better than nine of every ten revenue dollars. Inside that segment, a small group of buyers carries slightly more than half. That is a Pareto distribution stacked on top of another Pareto distribution, disclosed in a public filing, at the largest scale in the history of commerce.
It gets sharper when you look back one year. In the same quarter of fiscal 2026, NVIDIA’s own 10-Q disclosed that a single direct customer represented 23 percent of total revenue and a second direct customer represented 16 percent. Two buyers, 39 percent of the company. If concentration were fatal, that filing was the obituary.
Instead, revenue doubled. Guidance for the following quarter came in at $108 billion, plus or minus 2 percent. Supply commitments reached $279 billion. The most concentrated revenue base in the market produced the fastest growth in the market, and it did so in the same quarter that the company was actively spending to make itself less concentrated.
That combination is the whole article.
Why Is 92 Percent of Revenue Not the Same as 92 Percent of Resources?
Revenue concentration measures who pays you. Activity concentration measures what you spend resource on. They are different portfolios with opposite management rules. Revenue concentration is exposure you monitor and buy down. Activity concentration is discipline you build on purpose. Operators who confuse the two either refuse to cut or cut the wrong thing.
Almost every objection I hear to portfolio work is a category error. Someone reads “concentrate on the top 4 percent” and hears “bet the company on four customers.” Those are not the same sentence.
The activity portfolio is everything your organization spends hours, capital, engineering capacity, floor space, and management attention on. Every SKU you tool for. Every quote you configure. Every custom variant, every legacy platform, every reporting package nobody reads. Concentration here is pure upside. The fewer things you do, the better you do them, and the cheaper each one gets.
The revenue portfolio is who signs the checks. Concentration here is a risk position. It is not automatically bad, and refusing it costs you more than accepting it usually does, but it is a number you carry consciously rather than a number you engineer upward.
NVIDIA is a masterclass in running both correctly at once. Its activity portfolio is brutally narrow: one architecture family, one software ecosystem, an annual platform cadence, and a Data Center segment carrying more than 92 percent of revenue. Its revenue portfolio is concentrated too, but that side is being deliberately widened, and the company is spending real money to widen it.
Across the five major transformations I have led, Strategic Hostage exits have consistently produced 40 to 60 percent of the first-wave profit improvement in the first 90 days of transformation. That is what activity concentration pays. Revenue concentration pays nothing. It only costs, and the bill arrives all at once.
What Does 80/20 Squared Actually Tell You to Concentrate?
80/20 Squared is a recursive analysis of the activity portfolio, not the customer list. The top 20 percent of the top 20 percent, roughly 4 percent of all customer-product combinations, generates about 64 percent of total profit. The unit of analysis is the combination, not the buyer.
Run the recursion a third time and the picture gets more extreme still: 0.8 percent of combinations carries 51 percent of profit. This distinction is the one most people miss on first contact with the framework, and it is why the fragility objection keeps coming back. The 80/20 Matrix of Profitability does not rank customers. It ranks customer-product intersections, which is a completely different object.
A single large account can sit in your best quadrant on three products and your worst quadrant on eleven others. Cutting the eleven does not cut the customer. It cuts eleven activities that were consuming tooling, changeover time, engineering support, and working capital while destroying margin. The account often gets more profitable and more loyal, not less.
That is why the framework’s own headline number is a combination number. Four percent of combinations, 64 percent of profit. In the businesses I have run, the top quadrant regularly generates 140 to 200 percent of total company profit, which is only arithmetically possible because the rest of the portfolio is actively burning the difference. Full definitions for each level live in the Stagnation Assassin glossary.
Subtracting activities is also the only version of growth that funds itself. The classic argument for pruning before expanding is laid out well in Harvard Business Review’s work on growth outside the core, and the sequencing holds: you cannot buy your way into focus, you can only cut your way there.
Why Did NVIDIA Line Up $500 Billion to Build a Long Tail?
On August 10, 2026, NVIDIA signed memorandums of understanding with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR to mobilize more than $500 billion in third-party capital for its customers. The structure lets buyers outside the hyperscaler group finance infrastructure without using their own balance sheets, which manufactures demand from a wider base.
Read that as a portfolio move rather than a finance headline. The constraint on selling to smaller buyers was never appetite. It was creditworthiness. So the company went and built the credit market instead of waiting for one to appear.
The results were already visible in the quarter. The non-hyperscaler bucket, covering AI clouds, industrial, and enterprise, grew 138.5 percent year over year against 101.2 percent for hyperscalers, and 25 percent sequentially against 13 percent. The smaller side of the revenue portfolio is growing faster than the bigger side. That is deliberate de-risking, executed while activity concentration stayed exactly where it was.
One honest caveat, because the number is being quoted carelessly everywhere: these are memorandums of understanding, not definitive agreements. Individual commitments and collateral terms have not been disclosed. The intent is unambiguous. The execution is not yet proven.
What Happens When Operators Get the Split Backwards?
There are two standard failure modes. The first is refusing to prune activities because concentration feels dangerous, which produces a business that is busy, complex, and unprofitable. The second is firing customers when the real disease was activity bloat, which shrinks revenue while leaving the cost structure fully intact.
The first failure is the more common one, and it always wears the same costume. The phrase is “we might need it someday.” The variant is “that customer is strategic.” I have never once found a strategic account that was strategic. I have found a great many that were simply old, large, and unexamined, which is a different thing entirely. The segment-level version of this analysis is the Right-to-Win Matrix, and its Red Cells tell you exactly which of those relationships have no defensible future.
The second failure is rarer but more expensive, because it feels decisive. A team runs a customer ranking, sees the tail, and starts terminating accounts. Revenue leaves immediately. The tooling, the SKU count, the changeover burden, the engineering support tickets, and the overhead structure that made the tail unprofitable all stay exactly where they were. Now the same cost base is spread across less volume. The business gets worse, and the operator concludes the framework does not work.
The framework worked fine. It was pointed at the wrong axis.
How Do You Run the Concentration Split on Your Own P&L?
Build two rankings, not one. Rank every customer-product combination by profit per unit of resource consumed, using activity-based costing rather than allocated overhead. Then separately rank customers by share of gross profit. Treat the first list as a cut list and the second as a risk register. They are never managed the same way.
The sequence matters more than the arithmetic.
- Rank combinations, not customers. Customer times product. If your system cannot produce that view, that limitation is itself the first finding.
- Cost with activity-based costing. Allocated overhead hides the tail by spreading it evenly. The tail is not evenly expensive. It is catastrophically expensive in specific places.
- Recurse the top slice. Run 80/20 inside your top 20 percent. The 4 percent that emerges is where your protection budget goes.
- Build the cut list from the bottom of the activity ranking. Not the bottom of the customer ranking. These lists overlap far less than anyone expects.
- Score revenue concentration separately. Largest customer as a percent of gross profit. Top three. Top five. Write those three numbers down. That is your risk register, and it gets a ceiling, not a target.
- Fund the widening with the cutting. The margin released by activity subtraction is what pays to develop the accounts that reduce your revenue concentration. This is the NVIDIA sequence in miniature.
Step six is the one teams skip, and skipping it is why portfolio work so often stalls after the first wave. Subtraction that funds nothing feels like austerity. Subtraction that funds a deliberate widening of the buyer base feels like strategy, because it is.
What Does This Look Like Inside a Manufacturing Business?
In manufacturing the revenue concentration risk is rarely the customer name. It is the platform program, the single qualified plant, or the one OEM specification your top products are designed around. Those dependencies sit underneath multiple customers at once, which makes them invisible on a customer ranking and lethal when they move.
I have watched this play out at close range. In one transformation documented in my research on the 80/20 Matrix, systematic portfolio rationalization moved a business from a negative 19 percent operating margin to breakeven over 36 months, worth $176 million in annual profit improvement. Same market, same sales team, same factories. The only thing that changed was which combinations we were willing to keep.
We took a business from negative 19 percent operating margin to breakeven in 36 months and released $176 million in annual profit. We did not add a single customer to do it. Every dollar came from subtracting activities that were destroying value while wearing the costume of revenue.
I also learned the vendor-side version of this the hard way. I bought, ran, and eventually sold a small industrial tank liner manufacturer, and the buyer was the company’s largest vendor. When one relationship is simultaneously your biggest supply dependency and your most credible acquirer, that is a concentration position on the input side of the business that no customer ranking would ever have surfaced.
Run the same two-axis test on suppliers, qualifications, and certifications. Concentration risk does not care which side of the P&L it lives on.
What Should You Actually Cut This Quarter?
Cut activities, not accounts. Pull the bottom quartile of your combination ranking by profit per resource unit, confirm the cost actually exits with the combination, and kill them in one wave rather than negotiating them one at a time. Then measure whether your revenue concentration got better or worse as a result.
The measurement at the end is the discipline most operators never install. If you cut correctly, activity concentration goes up and revenue concentration stays flat or improves, because the accounts you kept got healthier and the released capacity went into winning new ones. If revenue concentration climbs while activity concentration barely moves, you cut customers instead of work, and you should stop and reverse before the next wave.
NVIDIA published both halves of that scorecard in a single release: 92 percent of revenue in one segment, and the smaller half of that segment growing faster than the larger half. Narrow what you do. Widen who pays you. Those instructions only sound contradictory until you notice they describe two different portfolios.
Frequently Asked Questions
Does 80/20 Squared mean I should fire my smallest customers?
No. 80/20 Squared ranks customer-product combinations, not customers. A large account can occupy your best quadrant on some products and your worst on others. Cutting the unprofitable combinations usually keeps the account and improves its margin. Customer termination is a separate decision with separate criteria.
Is high customer concentration always a risk?
It is always an exposure, which is not the same as always being a mistake. NVIDIA reported one direct customer at 23 percent of total revenue and a second at 16 percent in the second quarter of fiscal 2026, then doubled revenue the following year. The discipline is measuring the number and buying it down deliberately, not avoiding it.
What is the difference between activity concentration and revenue concentration?
Activity concentration measures how few things you spend resource on, and more is better. Revenue concentration measures how few buyers fund you, and it is a risk position you cap rather than maximize. The 80/20 Squared analysis governs the first. Account development and channel strategy govern the second.
About the Stagnation Assassin
Todd Hagopian is a Fortune 500 transformation executive who has generated $3B+ in shareholder value across Berkshire Hathaway, Illinois Tool Works, Whirlpool, and JBT Marel, where he serves as VP of Global Product Strategy. Known as The Stagnation Assassin, he is the author of two published books: The Unfair Advantage: Weaponizing the Hypomanic Toolbox and Stagnation Assassin: The Anti-Consultant Manifesto. His blog is published in 15+ languages and read by operators worldwide. Bring him to your stage via his speaking page or connect with him on LinkedIn.
Most executives can name their largest customer. Almost none can name the 4 percent of customer-product combinations carrying 64 percent of their profit, or the bottom quartile quietly burning it. Book a 15-minute 80/20 Squared Concentration Audit and we will find both numbers in your business. Start the audit here.