The Art of Culture

Culture is designed.

This article is part of The Art of Leadership series, where I cover twelve principles of leadership that my research found underpin superior company performance. Read the full series here.

Sixty people report directly to Nvidia CEO Jensen Huang. Sales, engineering, software, and finance leaders, senior vice presidents among them, all report to a single individual at the head of a company valued in the trillions of dollars.

There are no one-on-one meetings and no written performance reviews. No piece of information reaches Huang’s senior staff without reaching everyone else at the same time. Speaking at the Stanford Institute for Economic Policy Research, he noted that there is nothing he tells his executives that he does not also tell the rest of the company.

Nearly every management convention of the past forty years suggests this arrangement should fail. The recommended span of control is roughly seven to eight direct reports; one-on-ones are widely treated as a foundation of people management; and written reviews are assumed to be the mechanism through which development is structured. Removing all three should, in theory, produce an overloaded leader, a disconnected senior team, and an organisation unable to decide anything.

Instead, the company is worth around $4.8 trillion.

The structure creates the culture

Huang has described his reasoning. In an interview with Lex Fridman, he explained that individual one-on-ones are effectively impossible at that scale; problems are instead placed in front of the entire group, and the group works on them together.

In a conventional organisation, one of the most valuable assets an executive holds is information: what the CEO is thinking, sensitive data about performance, what was discussed at the board. That information functions as a form of currency, and the hierarchy exists in part to control how it is distributed. Remove the scarcity, and much of that apparatus has little left to do.

Huang has framed the intent explicitly. The company, he has said, was designed for speed, so that information moves as quickly as possible and people are “empowered by what they are able to do, not what they know.” When no one can accumulate status through privileged access, the remaining currency is contribution.

The design also changes the nature of disagreement. Because Huang reasons through problems in front of the group rather than announcing conclusions, colleagues can challenge the logic rather than the verdict.

The larger point is that the structure was chosen to produce a particular culture. Once private information is removed, status has to be earned elsewhere. In the absence of written reviews, feedback becomes continuous and public or it does not happen at all. And a leadership team of sixty cannot sustain an inner circle, because there is no outer circle to distinguish it from. The organisational design generates the cultural outcome, rather than the reverse.

What follows is a three-part diagnostic for identifying what a culture actually is, as opposed to what leaders hope it is, along with a method for changing it.

Culture is not what you value

An organisation’s values are what it says it cares about. Its culture is how its people actually behave. These are distinct, and when they conflict, behaviour prevails, because people trust observed actions over stated intentions.

Consider how a new employee comes to understand a culture. It is rarely learned in onboarding. It is learned in the first month by observation: who was promoted and what they were like, who was moved aside, how leadership responded when a project failed, and how it responded when someone delivered inconvenient news. Within weeks, the employee has constructed a working model of what is safe and what is rewarded, and that model tends to govern behaviour for years.

Culture, in this sense, is less a sentiment than a response to leadership behaviour. People continuously observe how leaders are likely to react and calibrate accordingly, and each of those judgements is assembled from evidence leaders supply, often without realising they are supplying it.

The three signals

A culture can be diagnosed along three dimensions.

The first is what gets rewarded: what produces promotion, budget, visibility, and status. If the people who advance are consistently those who hit their targets irrespective of method, that is the operative culture, whatever the values statement asserts.

The second is what gets tolerated: the behaviours that are noticed and allowed to pass. When someone is attacked in a meeting and no one intervenes, tolerance is not neutral; it is a decision, and the organisation reads it as permission.

The third is what gets punished, whether formally through process or informally through exclusion.

Applied to Nvidia, the pattern is coherent. Contribution is rewarded rather than access, because access has been rendered largely worthless. Being wrong in public is routinely tolerated, since reasoning happens in the open and no one is correct every time. And information hoarding is constrained structurally rather than personally, because little remains to hoard. Read together, these three signals make the flat structure look less eccentric than inevitable, the only configuration that could deliver what Huang intended. The same diagnostic can be applied to most organisations, and it does not take long to run.

Why leaders lose control of the culture they build

Few leaders set out to create a culture they would be ashamed of, yet many do. Three mechanisms are usually responsible.

The first is that a leader’s reactions are read by the entire organisation. When someone raises an inconvenient truth and the leader responds by defending the decision, or by asking who was responsible before asking what happened, the lesson that challenge is risky propagates well beyond the individual conversation. No formal punishment is required. Visible discomfort, once, in front of the right audience, is sufficient to shape behaviour.

The second is that culture is effectively set one or two layers below the top. A senior leader may be genuinely open, but the more consequential question is whether the manager two levels down is, because that manager’s reactions are what most of the organisation actually experiences. This is the cascade problem that undermines accountability in large organisations, and it is the problem Huang addressed by removing layers rather than attempting to reform them.

The third is that measurement overwhelms rhetoric. When an operational review interrogates delivery dates in forensic detail while treating quality in general terms, the organisation infers the real priorities regardless of what is said aloud. People direct their attention towards whatever they expect to be asked about, which is why what leaders attend to communicates more than what they declare.

How to change what a culture rewards

Culture responds to changed consequences far more than to communication. Four interventions are useful, roughly in sequence.

Audit recent promotions. Examine the last several people to advance and ask what behaviour was in fact being rewarded. If every promotion went to reliable on-time delivery and none went to someone who surfaced a difficult problem early, the culture has been identified, along with its cost, which typically includes the departure of the strongest people, who read the pattern sooner than anyone.

Remove a piece of privileged information. This is a scaled-down version of the Nvidia approach. Take something currently shared only with the senior team and distribute it broadly, then observe how the behaviour of the former information-holders changes. The exercise reveals how much of the hierarchy performs a genuine function and how much simply guards a channel.

Recognise the right behaviour publicly and specifically. When an individual surfaces a problem that averts a costly decision, the recognition should be explicit: naming the person, describing what they raised, and connecting it to the outcome. Generic praise for “speaking up” accomplishes little; a concrete, publicly credited example gives everyone else a specific behaviour to emulate, and tends to outperform any formal values programme.

Make the tolerated visible. Identify one behaviour that has been allowed to pass, state plainly that it needs to stop, and then act the first time it recurs. The enforcement is the entire point. Announcing a standard and failing to uphold it is worse than silence, because it teaches the organisation that stated standards carry no consequences.

What this is not

None of this describes a soft or uniformly comfortable environment. By most accounts Nvidia is a demanding place to work; ten current and former employees told Bloomberg of seven-day weeks, workdays ending at one or two in the morning, and meetings that escalated into open conflict. Huang has said he would rather “torture you into greatness” than dismiss someone, framing the intensity as belief in the individual. Asked on 60 Minutes whether he is demanding and difficult to work for, he agreed, adding that extraordinary results should not come easily.

And yet Nvidia’s voluntary turnover in 2023 was 2.7%, against a semiconductor-industry average of 17.7%. One of the sector’s most demanding employers loses its people at roughly a sixth of the prevailing rate.

This is the element most culture programmes misread. A strong culture is defined by clarity about expectations. Transparency is not the same as leniency, and openness does not imply diminished standards. What Nvidia appears to have is a culture with unusually legible signals: employees know what is rewarded, what is tolerated, and what the standard is. People will absorb considerable difficulty when they know precisely where they stand; what they tend not to tolerate is ambiguity.

Nvidia’s particular intensity is not the point, and it is a reasonable thing to reject. What no organisation can reject is the fact that its culture is already transmitting these three signals, whether or not anyone designed them deliberately.

Implications for leaders

Huang built a structure suited to a company of engineers solving shared technical problems at speed, and it would not transfer cleanly to a retailer or a bank.

What transfers is the method. Huang treated culture as a design problem. Rather than issuing a values statement about transparency, he removed the conditions under which privileged information could exist, and allowed the culture to follow from the structure.

The practical starting point is an honest audit. What does the organisation in fact reward? What does it allow to pass? And what, formally or otherwise, does it cost an individual to tell senior leadership that it is wrong? The answers to those questions describe the culture that exists, which is the only one that matters.