Shyam Sankar on Defense Procurement, Software and Building Talent

Shyam Sankar tells Harry Stebbings that his work ethic came from his father, whose siblings helped send him to college. Sankar says the family suffered armed violence in Nigeria, where his father had gone to work, and resettled in the United States when Sankar was a young child. He recalls a family emphasis on gratitude, staying together and making the most of opportunity.

Sankar believes achievement is usually a team effort, more like a band than a solo act. On luck versus hard work, he says both probably matter, but he prefers to concentrate on the work he can control. Stebbings adds that people may also find it difficult to acknowledge their own hard work publicly.

After Cornell, Sankar went to Stanford to reach Silicon Valley and joined money-transfer company Xoom as its fifth employee. He later joined Palantir when it was roughly 12 people, drawn to national-security problems. He says the September 11 attacks and the 2006 Mumbai train bombings, in which his uncle was a victim, shaped his interest.

At Palantir, engineers needed someone to work closely with customers on difficult problems, even though the early team lacked access to classified environments and experience selling to government. Sankar says they built a sales and product approach from first principles, with considerable difficulty, rather than importing the conventional model.

He describes conventional government contracting as often paying for labor and costs plus an allowed profit. Sankar says this can be understandable when a government is the dominant buyer of a specialized capability, but argues it can weaken incentives for privately financed innovation. Stebbings notes that allied countries can also be buyers, complicating the single-buyer analogy.

Palantir’s alternative, Sankar says, was to finance product development and sell software on a license or subscription basis rather than accepting service revenue by the hour. He acknowledges that convincing government buyers to adopt that model was difficult. As Palantir’s CTO, he has a direct commercial interest in this procurement approach.

In government, the buyer and the user may be different people. Sankar says Palantir spends time with users to identify needs and make the case for a product that buyers must acquire through a complicated process. He summarizes a deal as needing a requirement, funding and a contract, though the path varies by opportunity.

Asked about land and expand, he says the government version is driven by product capabilities, not simply selling more seats. Teams need to work with users in operational settings to understand adjacent problems. He argues that investment in infrastructure to operate software in controlled, air-gapped environments is also necessary for scale and margin.

Sankar argues procurement is built around a hardware lifecycle: fund research, buy a finished capability and then sustain it. Software, he says, keeps evolving and requires continuing development, so separating procurement from sustainment can fit poorly. This is his characterization of government practice, not a description of every contract.

He traces the US defense industrial challenge to post-Cold War consolidation. Using World War II mobilization as a contrast, he praises the role of industrial leaders and founders in scaling production. He says the 1990s budget contraction encouraged consolidation and shifted attention toward financial transactions rather than new venture-backed innovation; this is his historical interpretation.

Sankar sees a divide between government’s mission and the perception that suppliers are merely seeking profit. He argues the deeper issue is a market that rewards conformity and hourly work over entrepreneurial risk. Stebbings asks whether wartime urgency gives private suppliers leverage; Sankar instead emphasizes leaders’ willingness to break normal processes to meet an urgent need.

He cites Lockheed’s Kelly Johnson and Skunk Works as an example of an engineer given room to innovate quickly. In Sankar’s view, excessive process and risk aversion today reduce catastrophic failures but also prevent exceptional successes. His historical comparisons are used to support an argument, not a controlled comparison of acquisition programs.

Stebbings points out that tender and insurance requirements can exclude smaller suppliers. Sankar agrees and argues that rules intended to demonstrate fairness can produce neither fair competition nor speed. He considers European procurement slower than the US process and says risk of protests and blame discourages decisions; these are his assessments, not measured cross-country findings.

Sankar says the United States can act especially quickly when political will and a clear problem align, but often fails to do so in routine conditions. He wants the Department of Defense to pick multiple winners and award enough business for successful companies to scale, arguing that spreading small sums among too many firms can leave them unable to grow.

He cites roughly $100 billion of recent defense-tech venture investment and describes hypothetical venture return dynamics to make his case for concentrating awards. Neither the investment total nor the portfolio return examples are substantiated in the interview. His proposal could benefit suppliers such as Palantir, and he frames it as a way to sustain a broader ecosystem.

On domestic preference in defense buying, Sankar distinguishes avoiding dependence on adversaries from using government purchasing to support domestic employment. He argues that small-business innovation grants may serve a jobs objective but, in his view, should not be assumed to create scaled defense champions. He leaves the policy choice to policymakers; the research claims he references are not detailed.

Stebbings notes that conflict can expand the addressable market for suppliers. Sankar compares US defense spending as a share of GDP with Cold War levels, explicitly saying his figures should be checked. He favors a level between the contemporary share and the historical peak, while saying that does not necessarily mean doubling the budget; these are his policy views, not verified budget figures.

Speaking after the Israel–Hamas war began and the invasion of Ukraine, Sankar says his defense customers were already serious about conflict, but battlefield learning is accelerating. He points to low-cost, single-use drones as an example of how tactics, supply chains and procurement assumptions can change. The conversation was recorded on December 7, 2023, and published in January 2024.

If he could change one thing at the US Department of Defense, Sankar would run multiple competing programs for a capability. He argues that parallel approaches can test technical uncertainty, motivate program managers and speed iteration. This is a proposal and expected effect, not proof that competition always lowers costs or improves outcomes.

He sees a disconnect between combat commanders who use equipment and those who procure it. His proposal would give operational leaders greater influence over how much of competing variants is purchased, with allocation adjusted as they learn. He also favors revisiting requirements after users experience the product, rather than treating an old list as permanently correct.

Asked about elections, Sankar says leadership changes matter, especially in the UK where he sees more cabinet involvement in contracts. In the US he considers congressional budget delays more consequential: under a continuing resolution, he says, new starts can be blocked. These observations describe his experience and the period of the interview.

He says venture capital and entrepreneurs have increasingly entered defense technology and asks investors to explain their economics to policymakers. Sankar contrasts Washington’s concern with conflicts of interest and Silicon Valley’s preference for having financial commitment behind a view, while acknowledging that conflicts can be harmful and that a middle ground is needed.

Turning to company building, Sankar says substance and creativity matter more than process. Process can help scale an idea, but following a previous procedure without the underlying insight will not recreate success. Stebbings challenges how that works as teams grow; Sankar agrees both matter but gives substance priority.

He calls talented individuals asymmetric and recommends shaping roles around their strengths rather than forcing everyone into a fixed organizational slot. His “artist colony” analogy allows for periods when a person’s work is less successful, with managers making a longer-term commitment rather than judging every project in isolation.

Sankar says managers should give talented people autonomy, resources, room to learn and protection from some distractions. If work continues to disappoint, retaining the person remains a judgment call, and he acknowledges the original assessment can be wrong. He credits Palantir colleagues Alex Karp and Joe Lonsdale for spotting early talent.

Stebbings asks what happens when budgets are cut and layoffs are required. Sankar accepts that external realities can force difficult choices. He argues constraints can drive creativity, but too many budget constraints can undermine it; a manager’s job is to balance financial limits with the environment in which people work.

His biggest hiring mistakes, he says, came from importing a supposedly proven profile rather than starting with the people and problem at hand. He now asks whether he would want to work alongside someone for a decade and whether that person can grow into new responsibilities. Prior experience still needs translation into a different company’s context.

Reinvention is difficult not only because of the work, Sankar says, but because individuals must surrender some earned status and learn again. A new employee may know more about the new problem than someone who has been there for years. He has come to see helping people through that discomfort as part of the work rather than simply trying to eliminate it.

On AI, Sankar predicts value will accrue to infrastructure and application owners while model providers face pressure, with many models becoming commoditized. He argues that useful applications combine models with conventional code, workflows and human judgment. Stebbings pushes back that even commoditized infrastructure can support enormous businesses; Sankar agrees infrastructure can capture value.

He advises a model company to consider moving into applications and distinguishes a thin chatbot wrapper from deeper integration with tools, state and workflows. He says some existing model providers may remain winners while more can emerge. These are views from December 2023; Palantir sells enterprise software and has a commercial interest in application-layer adoption.

Sankar worries about confusing a striking proof of concept with a product that reliably creates value. He cites the long gap between an early autonomous-car demonstration and commercial deployment as an analogy, and urges technologists to focus on usable outcomes. He says his own thinking has shifted toward doing incremental, hands-on work to make good ideas scale.

Looking ahead five years from the interview, he predicts more geopolitical change and institutional strain than many expect. He also sees AI as an opportunity to improve institutions by partnering with the people inside them rather than tearing them down. The forecast reflects his perspective at the time, not an established outcome.