Writer: Aidan Morgan Chan
Editor: Emilia Marshall
In early 2025, Kazakhstan’s National Investment Corporation (NIC) announced plans to allocate capital into artificial intelligence infrastructure funds, marking the Central Asian nation’s entry into a sector that has consumed billions in venture capital and state investment globally.
The NIC, which manages $3.4 billion of the country’s $60 billion National Fund, intends to commit approximately $50 million each to one or two AI-focused venture capital funds. This represents a departure from the fund’s historical reluctance toward venture capital. In an interview at the NIC’s headquarters in Astana, Chief Executive Officer Serikzhan Rysbekov stated that the corporation declined venture capital investments in 2021 due to inflated valuations, but now views AI development as making the asset class attractive again.
The timing merits examination. Kazakhstan’s decision arrives after significant corrections in technology valuations and amid persistent uncertainty about returns on AI investments. For a sovereign wealth fund tasked with preserving intergenerational wealth derived primarily from hydrocarbon extraction, the calculus differs substantially from that of a Silicon Valley venture firm. A venture fund operating within Silicon Valley’s ecosystem is geared toward technology commercialization due to the concentrated talent, established networks, and existing infrastructure in the region. By contrast, Kazakhstan will be investing finite national wealth from depleting reserves into an environment lacking these foundations; the country currently has minimal research capacity, a dependence on imported technology, and an economy where core sectors remain undigitized.
This analysis evaluates the economic rationale underlying Kazakhstan’s AI infrastructure investment strategy. The research question centers on whether such investments constitute sound portfolio diversification for a commodity-dependent economy or misallocate sovereign resources into speculative assets. The inquiry extends beyond simple risk-return metrics to encompass broader questions of economic development, fiscal sustainability, and institutional capacity.
The contribution to existing literature operates on two levels. First, it contributes to the limited empirical literature on sovereign wealth fund allocation to venture capital and emerging technologies, examining how funds with dual mandates–wealth preservation and fiscal stabilization–evaluate illiquid alternative investments. Second, it analyzes the portfolio dilemma facing resource-dependent economies: how to diversify financial assets away from commodity exposure while the underlying economy remains structurally dependent on those same commodities.
Methodologically, the analysis examines AI infrastructure investment through multiple lenses. It assesses the risk-return profile relative to alternative allocations available to NIC, incorporating the dual mandate constraints that distinguish sovereign wealth fund decision-making from conventional portfolio theory. It evaluates sector-specific digitization requirements across oil and gas, agriculture, and mining to determine whether demand exists for AI capabilities that infrastructure investment would enable. Institutional capacity is examined through a comparative analysis of sovereign wealth funds managing alternative investments, with particular attention to staffing constraints and manager selection challenges.
Kazakhstan offers a particularly valuable case study for several reasons. Unlike massive sovereign wealth funds where individual allocation decisions disappear into trillion-dollar portfolios, the NIC’s $3.4 billion scale means AI investments will materially affect portfolio performance and institutional focus. The $50 million commitments represent roughly 1.5% of the fund’s assets, substantial enough that success or failure will generate tangible institutional consequences without threatening overall fund stability. Furthermore, Kazakhstan’s economic structure—commodity-dependent and lacking a robust technology ecosystem—will provide a case study on whether AI infrastructure investment can catalyze technological development in contexts lacking the prerequisites that allow such investments to succeed in established innovation hubs. The country sits at an analytical sweet spot: developed enough to have the institutional capacity for sophisticated investments, yet disconnected enough from global technology networks that outcomes will reveal whether AI infrastructure creates its own demand or requires pre-existing digital maturity.
Graphic By: Sharvani Andurlekar
II. Macroeconomic Context & Investment Environment
Fiscal Architecture
Kazakhstan’s National Fund operates as a stabilization and savings mechanism, established in 2000 to manage revenues from the country’s hydrocarbon sector. The fund held approximately $60 billion as of 2025, representing roughly 20% of the nation’s GDP in 2024. Its structure follows a dual mandate: stabilization transfers to the state budget during revenue shortfalls, and long-term savings to preserve wealth for future generations.
The National Investment Corporation functions as the fund’s alternative investment arm. Managing $3.4 billion, it receives capital from two sources in roughly equal proportion—direct allocations from the National Fund’s savings portfolio and a portion of the central bank’s foreign exchange reserves. This arrangement creates a unique mandate profile. Unlike pure stabilization funds that prioritize liquidity, or pure savings funds that can pursue longer time horizons, the NIC must balance both objectives.
Current allocation constraints limit the NIC to 3% of the National Fund’s savings portfolio, though the cap permits expansion to 5%. The savings portfolio itself contains $54 billion of the fund’s total assets, meaning the NIC could theoretically manage up to $2.7 billion from this source alone. Rysbekov has indicated that the organization seeks an increase in this ceiling, reflecting broader frustration with return constraints under the existing framework.
Revenue volatility presents the central challenge. The Organization of the Petroleum Exporting Countries reported that Kazakhstan produces roughly 1.8 million barrels of oil per day, making it Central Asia’s largest producer. Brent crude prices ranged from $69 to $91 per barrel during 2024, leading to revenue swings that the National Fund must absorb. The fiscal sustainability calculation depends heavily on oil price assumptions. At $70 per barrel, the National Fund faces gradual depletion under current transfer rules. Above $85, it accumulates reserves. This creates pressure to enhance returns on existing assets rather than simply preserve nominal value.
B. Market Timing & Valuation Environment
The NIC’s previous rejection of venture capital investments in 2021 provides essential context for the current timing of this allocation decision. That year marked the peak of venture capital activity globally, with deal values tripling from $50 billion in Q2 2020 to $170 billion by Q3 2021. Software companies commanded unprecedented revenue multiples as zero interest rates and pandemic-fueled digital transformation drove investor appetite to historic highs.
By 2024, the environment had shifted considerably. Venture capital deal values had plunged 60% back to $50 billion by Q2 2023, with rising interest rates compressing multiples across growth-oriented sectors. Down rounds became common.
This correction makes the NIC’s 2025 entry analytically interesting, though the timing presents complex tradeoffs. Lower valuations reduce absolute capital at risk compared to 2021 peaks, yet the market correction itself reflects genuine uncertainty about AI commercialization timelines and monetization paths. The deployment of capital by major technology firms into AI infrastructure—data centers, chip production, model training—has reached unprecedented levels. Microsoft, Google, Amazon, and Meta collectively spent over $200 billion on capital expenditures in 2024, much of it AI-related. Yet monetization remains largely prospective, with even leading technology companies struggling to demonstrate clear paths from AI investment to profitable revenue streams.
For Kazakhstan, lower entry prices address neither the structural challenges of manager selection nor the fundamental question of whether AI infrastructure investment aligns with the country’s economic development needs. The NIC’s ability to access top-tier fund managers remains constrained regardless of valuation levels. More critically, the mismatch between AI infrastructure capabilities and Kazakhstan’s undigitized core economic sectors—oil extraction, agriculture, and mining—persists regardless of venture capital pricing dynamics.
The venture capital cycle typically operates on a 7-10 year time horizon. Funds raised in 2025 will deploy capital over 2-4 years, then hold positions for several more years before exits materialize. This means the NIC is effectively betting on the state of AI markets in 2030-2035, not 2025.
Infrastructure funds represent a different timing dynamic. These vehicles invest in physical assets—data centers, fiber networks, and power infrastructure supporting AI facilities. Returns come from cash flows rather than exit multiples. A data center generates revenue from leasing space and power to clients, producing relatively predictable returns if occupancy remains high. This shifts the risk profile away from pure valuation risk toward operational and demand risk.
III. Strategic Allocation Analysis
The proposed AI investment must be evaluated within the context of the NIC’s portfolio mandate and existing constraints. The corporation manages $3.4 billion with a dual mandate: generating returns on National Fund savings while maintaining sufficient liquidity for fiscal stabilization transfers during oil revenue shortfalls.
This dual mandate creates tension with illiquid alternative investments. Infrastructure fund commitments totaling $200-250 million would lock up roughly 6-7% of NIC assets for a decade or more, limiting the fund’s capacity to respond to fiscal stress.
The diversification rationale hinges on whether AI returns correlate with oil prices. If correlation is low or negative, the allocation could provide genuine portfolio benefits. Empirical evidence remains limited, but technology sector returns have historically shown a weak correlation with crude oil prices. However, AI infrastructure may differ: data centers consume substantial electricity, creating sensitivity to energy costs that could produce a positive correlation with oil prices, reducing rather than enhancing diversification.
Manager selection also presents a structural challenge. Returns in venture capital exhibit extreme dispersion—top-quartile funds return 20-30% IRR while bottom-quartile funds often fail to return committed capital. This dispersion far exceeds public markets, making it critical to access quality managers. Kazakhstan’s NIC likely lacks competitive positioning with leading firms. A $50 million commitment offers limited strategic value to top-tier managers who have ample capital sources and no particular interest in Central Asian market access.
The opportunity costs are also significant. Alternative allocations to private debt or real estate in major global cities offer proven risk-return profiles. For AI venture capital to justify its allocation on a risk-adjusted basis, it must generate returns sufficient to compensate for both higher volatility and extended illiquidity—a substantial hurdle given Kazakhstan’s disadvantages in accessing top-tier managers who generate outsized returns.
IV. Economic Development Implications
A. Domestic Infrastructure Spillovers
Beyond portfolio returns, AI infrastructure investments carry potential economic development effects for Kazakhstan’s domestic economy. The most direct channel runs through the partnership between state-owned Kazakhtelecom and Nvidia, which aims to build local AI infrastructure capacity.
Data center construction has high potential to generate economic activity. Yet from the perspective of a country like Kazakhstan, the critical question is not aggregate global investment, but rather the composition of expenditure and its domestic capture. Whether data center investment translates into substantial domestic economic benefit depends on the share of expenditure that circulates within the local economy versus leaking out through imports and foreign contractors. The multiplier effects depend on local content: if most equipment is important and construction contracts go to foreign firms, the domestic economic impact remains limited.
Power infrastructure presents both opportunities and constraints. Data centers consume enormous amounts of electricity. A facility processing AI workloads might require 200 megawatts of continuous power. Kazakhstan generates electricity primarily from coal and has surplus capacity in some regions. However, power grid reliability varies, and AI workloads cannot tolerate frequent outages.
Connectivity represents another bottleneck. AI data centers need high-bandwidth, low-latency connections to the global internet backbone infrastructure. Kazakhstan’s geographic position between Europe and Asia offers theoretical routing advantages, but connectivity infrastructure remains underdeveloped for frontier AI applications. In 2023, Kazakhstan provided approximately 10 Tbit/s in internet transit volume across its routes, a figure that falls well short of the hundreds of terabits to petabits of bandwidth that major AI training facilities require.
In February 2025, Kazakhstan’s Ministry of Digital Development signed a memorandum of cooperation with Freedom Telecom Holding to construct a fiber-optic highway and data centers for international internet traffic transit. The West-East national highway aims to expand data transmission capacity and establish an alternative route for internet traffic between Europe and East Asia, with completion scheduled for 2026. The project, financed through private investment, is expected to attract IT firms and telecommunications companies seeking data transit services.
While such infrastructure development signals intent to address connectivity constraints, execution risk remains substantial. The project’s reliance on private financing from a domestic broadband provider—Freedom Telecom, which currently operates local internet access services—raises questions about whether sufficient capital and technical expertise exist to deliver infrastructure meeting frontier AI data center requirements.
B. Productivity & Growth Channels
The productivity argument for AI infrastructure investment rests on technology spillovers. Proximity to advanced technology facilities theoretically accelerates adoption by local firms and develops technical human capital. The empirical evidence for such spillovers varies considerably across contexts.
The development of Silicon Valley is a classic example of cluster effects. Concentrations of technology firms, venture capital, and technical talent generated network effects that accelerated innovation and commercialization.
Replicating this in Kazakhstan faces obvious challenges. The country lacks an existing base of technology firms that would benefit from proximity to AI infrastructure. Its venture capital ecosystem is nascent. While this does not mean spillovers are impossible, they would likely require deliberate policy coordination beyond simply building infrastructure. Tax incentives for technology firms, streamlined business registration, protection of intellectual property, and visa policies that attract technical talent are all institutional factors that matter as much as investment in physical infrastructure.
The human capital dimension deserves particular attention. AI infrastructure operations require specialized skills: machine learning engineers, data scientists, and systems administrators familiar with high-performance computing environments. Kazakhstan’s technical education system would require adaptation to produce these skills at scale. In the short term, facilities would likely staff themselves with expatriate talent, limiting domestic knowledge transfer.
That being said, government investment in AI infrastructure could catalyze private sector investment if structured appropriately, that is, by reducing coordination problems. Individual firms might hesitate to invest in Kazakhstan due to concerns about power reliability, connectivity, or talent availability. If government investment addresses these gaps in infrastructure and human capital, private investment could become more viable.
V. Risk Assessment & Stress Analysis
A. Technology-Specific Risks
The sustainability of current AI investment levels remains uncertain. Large language models cost tens to hundreds of millions of dollars to train and require substantial ongoing compute resources to operate. Whether revenue streams can cover the massive infrastructure investments already deployed is unclear, creating risk that a contraction in AI investment could leave early movers exposed.
For Kazakhstan’s investment strategy, an AI investment slowdown would have severe consequences. Venture capital funds deployed into AI startups would see portfolio companies fail to achieve viable business models. Infrastructure funds invested in AI data centers would face declining occupancy as demand for computational capacity contracted.
Stranded asset risk would be particularly high for physical infrastructure. Data centers built specifically for AI workloads cannot easily be repurposed if demand disappears, as their cooling requirements, power density, and network architecture differ from conventional data centers. Computing hardware typically depreciates over 3-5 year periods, accelerating potential losses.
C. Opportunity Cost Analysis
Every dollar allocated to AI venture capital or infrastructure funds is a dollar not allocated to alternative investments. In attempts to diversify the NIC’s existing portfolio, different items offer varying risk-return profiles and degrees of proven performance.
Private debt, for example, typically targets 8-12% returns through lending to middle-market companies. Returns come primarily from interest payments rather than equity appreciation, providing more downside protection than venture capital.
To illustrate the risk-return tradeoff, consider a simplified Monte Carlo simulation comparing these allocations. Assume a portfolio allocates 10% to AI venture capital with an expected return of 15% and standard deviation of 40%, versus an alternative allocation to private debt with an expected return of 10% and standard deviation of 15%.
The simulation works by randomly generating 10,000 potential return sequences for each strategy over a 10-year period, drawing from normal distributions with these parameters. In each iteration, annual returns fluctuate around the expected value with the specified volatility. The venture allocation’s higher volatility (40% vs. 15%) means returns swing more dramatically, generating 50%+ gains in some years while producing 30%+ losses in others. By contrast, the lower volatility of private debt produces steadier but more modest outcomes.
Aggregating across 10,000 Monte Carlo simulations reveals that the venture allocation outperforms private debt in roughly 55% of scenarios—its higher expected return translates to better outcomes when volatility works in its favor. However, the venture allocation underperforms by larger magnitudes in the remaining 45% of cases. This asymmetry matters: when the venture allocation loses, these losses tend to be of greater magnitude due to high volatility, while private debt’s downside remains bounded. For a sovereign wealth fund prioritizing capital preservation, the distribution of outcomes may matter more than the expected value alone.
The breakeven analysis depends heavily on assumptions about the distribution of returns. Venture capital returns follow a power law distribution—a small number of investments generate most of the returns, while the majority fail or return minimal capital. If the NIC’s venture allocations fail to capture exposure to the few breakout winners, returns will likely disappoint relative to alternatives.
VI. Policy Economics & Welfare Considerations
A. Intergenerational Equity
Sovereign wealth funds derived from exhaustible resources face a fundamental tension. Current generations extract and sell natural resources that future generations will not have access to. The fund represents an attempt to transform finite resource stocks into financial assets that can generate perpetual returns.
The permanent income hypothesis suggests that optimal consumption of an exhaustible resource should remain constant across generations. This requires converting resource wealth into financial assets that yield sustainable returns. A fund that takes on excessive risk threatens this intergenerational compact. Current projections suggest Kazakhstan’s proven oil reserves will last approximately 40-50 years at current extraction rates.
The question becomes whether AI infrastructure investments align with intergenerational equity principles. If these investments generate superior risk-adjusted returns, they enhance the fund’s ability to provide for future generations. If they destroy capital or simply underperform compared to safer alternatives, they violate the intergenerational compact.
The answer depends partly on the time horizon. Venture capital and infrastructure funds lock up capital for 10-15 years. For a fund with a multi-generational mandate, this represents a relatively short horizon. If investments fail, the fund has decades to recover through alternative allocations.
However, the size of the allocation matters. For example, committing 6-7% of assets to a single speculative theme could concentrate risk in ways that may not be appropriate even with long-term horizons. AI represents a coherent risk factor—if the technology fails to commercialize successfully, both venture investments and infrastructure investments would likely suffer.
B. Market Failure Rationale
Government intervention in markets is typically justified by identifying a market failure that private actors cannot adequately address. Several potential market failures could provide a rationale for sovereign investment in AI infrastructure.
Coordination externalities represent one candidate. Building AI infrastructure requires simultaneous investment in multiple complementary assets: data centers, fiber networks, power generation, technical education, and regulatory frameworks. Private actors may underinvest because each component depends on others being in place.
This logic is grounded in historical precedent. Government investment in highway systems, telecommunications infrastructure, and internet backbone networks all addressed coordination problems that private markets struggled to solve.
Whether this logic applies to AI infrastructure in Kazakhstan is debatable. The bottleneck may not be a coordination failure but rather reflects fundamental economic challenges. If AI data centers are not economically viable in Central Asia due to distance from major markets, unreliable power, or lack of technical talent, government investment will not make them viable.
Public goods characteristics provide another potential justification. If AI infrastructure generates knowledge spillovers that private investors cannot fully capture, private investment will be suboptimal.
C. Comparative Institutional Analysis
Resource-dependent economies have taken various approaches to managing commodity wealth. Norway’s Government Pension Fund Global represents the conservative end of the spectrum—highly transparent, strictly rule-bound, and focused on liquid public markets. The UAE’s investment vehicles operate with less transparency but have demonstrated sophistication in alternative investments.
The empirical relationship between resource wealth management and economic outcomes has been extensively studied. Countries with transparent, rule-based sovereign wealth funds tend to experience better economic outcomes than those where funds lack clear governance.
Kazakhstan’s track record has been mixed. The National Fund has provided important fiscal stabilization during downturns. It has maintained reasonable transparency about aggregate asset allocation, though specific investment decisions often lack public disclosure. Governance quality concerns apply particularly to alternative investments. Private equity and venture capital operate with limited transparency, even in well-governed funds. Adding another layer of opacity through sovereign wealth fund structures creates opportunities for corruption or self-dealing that would be difficult to detect.
International comparisons suggest that smaller sovereign wealth funds often struggle with alternative investments. The Canadian Pension Plan Investment Board and Singapore’s GIC successfully manage large alternative portfolios, but they manage hundreds of billions of dollars in total and employ thousands of professionals. Funds managing less typically lack the scale to build specialized expertise across multiple alternative asset classes.
This suggests Kazakhstan might achieve better results by concentrating expertise in fewer areas rather than diversifying across many alternative asset classes. If the NIC developed genuine competency in one or two domains—say, infrastructure and private debt—it could deploy capital more effectively than by attempting to cover venture capital, infrastructure, multiple private equity strategies, hedge funds, and other alternatives.
VII. Conclusion & Research Implications (In Progress)
Kazakhstan’s decision to allocate sovereign wealth into AI infrastructure investments represents a test case for how resource-dependent economies navigate technological disruption. The $50 million commitments to venture capital funds and similar allocations to infrastructure vehicles constitute a modest but meaningful shift in the National Investment Corporation’s strategy.
From a pure portfolio optimization perspective, the economic logic remains ambiguous. AI investments offer potential diversification benefits if returns prove to be uncorrelated with oil prices, but the empirical basis for this assumption is shaky. Expected returns depend critically on manager selection, whereas Kazakhstan likely lacks competitive advantages in accessing top-tier funds. The illiquidity premium required to compensate for the lock-up periods suggests AI venture capital must generate 13-16% returns to justify the allocation, while median venture performance historically clusters around 10-15%.
The opportunity cost dimension may prove decisive. Private debt, real estate, and other established alternative investments offer clearer risk-return profiles with less specialized expertise requirements. A sovereign wealth fund with approximately 40 investment professionals faces inherent constraints on the number of specialized domains it can effectively cover.
Economic development justifications carry more weight if AI infrastructure investments address genuine coordination failures or generate spillovers that private markets underprovide. The Kazakhtelecom-Nvidia partnership potentially fits this pattern, though success depends on execution quality and whether Kazakhstan can resolve infrastructure bottlenecks. The timeline for productivity spillovers can extend across decades rather than years.
Risk assessment identifies multiple channels through which investments could disappoint. Technology-specific risks include the possibility of an AI winter if commercialization timelines extend beyond investor patience. Execution risks center on information asymmetry and adverse selection when choosing a manager. Institutional capacity constraints raise questions about whether the NIC can effectively evaluate and monitor complex alternative investments.
The intergenerational equity framework also highlights fundamental tensions present when making alternative investments. A fund tasked with preserving resource wealth for future generations should maintain conservative risk tolerances. Speculative allocations to emerging technologies may violate this mandate even if they offer higher expected returns. The concentration of risk in a single technological theme—where venture and infrastructure investments both depend on AI commercialization success—compounds this concern.
Policy recommendations emerge from this analysis. Resource-dependent economies considering AI infrastructure investments should first assess whether genuine market failures (such as coordination problems) justify public investment, or whether private markets are already allocating capital efficiently based on economic conditions. They should also evaluate their institutional capacity honestly—a fund with limited professional staff and nascent expertise in the technology sector may destroy more value than it creates by attempting sophisticated alternative investments.
Future research should track outcomes as Kazakhstan’s AI investments mature. The 10-15 year horizon for these investments means definitive results will not emerge until the mid-2030s. However, intermediate metrics can provide earlier signals, such as the quality of managers selected, the fee structures negotiated, the occupancy rates and pricing for any data center investments, and the emergence or absence of technology sector spillovers in Kazakhstan’s economy.
Comparative research across resource-dependent economies pursuing similar strategies could help to shed light on any common patterns that emerge. Saudi Arabia’s more aggressive AI investments through the Public Investment Fund provide a natural case study for comparison, as do the UAE’s various technology initiatives.
This analysis does not take a stance regarding whether Kazakhstan’s AI investment strategy will ultimately succeed or fail. The outcome depends on factors that remain highly uncertain, including the trajectory of AI commercialization, the quality of execution by the NIC and its advisors, the ability of Kazakhstan to address infrastructure and institutional bottlenecks, and macroeconomic conditions over the next decade. What the analysis establishes is that the risk-return proposition is far from obviously favorable, that substantial downside scenarios exist, and that the institutional barriers to AI investment may exceed the financial challenges of selecting good investments.
Featured Image by Uladzislau Petrushkevich on Unsplash
