From the Bank of NVIDIA to AI in Your Pocket: Who Will Control Supply Chain Intelligence?
NVIDIA is helping mobilize hundreds of billions of dollars to construct AI factories. Governments are investing in sovereign computing infrastructure to reduce their dependence on foreign technology providers. Meanwhile, increasingly capable artificial intelligence is moving onto the smartphone in your pocket.
We are building the world’s largest centralized computing infrastructure at precisely the moment that intelligence itself is becoming increasingly distributed. The departure of Jensen Huang from NVIDIA brings this apparent contradiction into sharper focus.
On November 12, 2026, NVIDIA announced the resignation of its founding CEO and the appointment of longtime Chief Financial Officer Colette Kress as interim chief executive. Three days later, U.S. Treasury Secretary Scott Bessent publicly advocated that NVIDIA recruit its next CEO from outside the company, declaring that it needed a “dramatic shakeup.”
Questioning whether NVIDIA had circumvented semiconductor export restrictions to increase revenues and earnings, Bessent added, “I think we need answers in the next 10 days from the board and from the outside counsel.”
These developments raise a question that extends considerably beyond NVIDIA’s leadership: Who will control artificial intelligence, and what degree of independence will enterprises and individuals retain as intelligence becomes embedded throughout the global economy?
For supply chain executives, the answer may ultimately determine the architecture of their operations, the resilience of their organizations, and their ability to compete.
The Bank of NVIDIA
Under Huang’s leadership, NVIDIA evolved from a semiconductor manufacturer into an architect of the AI economy itself. Its processors, networking technologies, CUDA software ecosystem, and increasingly its financial relationships have become foundational elements of an industry dependent on accelerated computing.
In February 2026, OpenAI announced a funding round that included $30 billion from NVIDIA. In September 2025, CoreWeave disclosed a $6.3 billion computing capacity arrangement under which NVIDIA agreed, subject to contractual conditions, to purchase residual capacity that CoreWeave could not otherwise sell.
In August 2026, NVIDIA announced partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs, and KKR intended to mobilize more than $500 billion in third-party capital for AI infrastructure. NVIDIA also disclosed conditional guarantees capped at $105 billion supporting approximately 4.25 gigawatts of computing capacity at SB Energy’s PORTS Technology Campus in Pike County, Ohio, on behalf of an OpenAI affiliate.
These figures represent different financial instruments, not cash already invested or amounts that should be combined into a single measure of exposure. Nevertheless, the emerging structure is remarkable: NVIDIA is helping finance the infrastructure that will consume NVIDIA products.
The metaphorical Bank of NVIDIA describes an ecosystem in which semiconductor sales, customer financing, infrastructure investment, and anticipated computing demand reinforce one another. Supply chain professionals will recognize the concentration risk inherent in these relationships. Apparently independent organizations may share substantial exposure to the same underlying assumptions about future demand.
But what happens if a growing proportion of AI inference moves away from centralized infrastructure? Improvements in computational efficiency may expand overall AI demand rather than reduce it. However, the relationship between AI adoption and centralized computing investment becomes considerably more complex.
Huang’s Entrepreneurial Paradox
One of the most interesting aspects of Huang’s strategy is that he was building both sides of this emerging architecture. While NVIDIA was financing enormous AI factories, it was also developing technologies that could place increasingly sophisticated intelligence in the hands of individuals and independently managed enterprises.
In March 2026, NVIDIA launched its Nemotron Coalition, bringing together developers including Mistral AI, Perplexity, LangChain, and Thinking Machines Lab to advance open frontier models. The initiative envisioned models developed using centralized computing resources that enterprises could subsequently adapt to their own industries and applications.
Earlier that year, at CES, Huang demonstrated a personal AI agent running locally on NVIDIA’s DGX Spark desktop computer. The company building some of the world’s most powerful centralized computing infrastructure was simultaneously demonstrating how intelligent systems could operate outside that infrastructure.
There is considerable commercial logic behind this strategy. A company confident in its ability to innovate can expand its market by enabling others to innovate. Open models, personal AI systems, industrial edge computing, and independently developed applications can increase demand for accelerated computing without requiring NVIDIA to control every application.
Huang’s departure raises an important question about the continuity of that entrepreneurial vision. Will NVIDIA’s next leadership team maintain the same emphasis on open innovation while managing substantial infrastructure commitments? There is no established evidence that its strategy will change, and NVIDIA has indicated that its product roadmap and customer commitments will continue. Nevertheless, the balance between entrepreneurial innovation, institutional governance, and capital allocation deserves attention.
Sovereign AI: Independence for Whom?
Governments increasingly regard artificial intelligence as a strategic resource comparable to energy, telecommunications, and semiconductor manufacturing. The European Union’s July 2026 initiative to establish up to seven AI Gigafactories, intended to mobilize more than €30 billion in investment, illustrates the growing importance of domestic computing capabilities.
National security and economic competitiveness provide legitimate reasons for governments to pursue technological independence. However, national sovereignty over artificial intelligence does not necessarily translate into individual or enterprise sovereignty over artificial intelligence.
A government can reduce its dependence on foreign technology companies while simultaneously increasing the dependence of domestic enterprises and citizens on nationally controlled infrastructure. The result could be a world in which AI becomes decentralized internationally while remaining highly centralized within individual jurisdictions.
Bessent’s remarks following Huang’s departure illustrate another dimension of this issue. Discussing semiconductor export controls, the Treasury secretary stated, “I don’t think we need the legislation. We have all the rules we need. We just need them enforced.”
The underlying investigations remain ongoing, and allegations of export-control violations should not be treated as established findings. Nevertheless, the fact that a Treasury secretary is publicly advocating changes to the leadership of one of the world’s most consequential technology companies demonstrates how closely AI infrastructure has become intertwined with national industrial policy.
For supply chain executives, this introduces another dimension of concentration risk. When access to intelligence depends upon relatively few semiconductor manufacturers, cloud providers, and model developers, government decisions affecting those organizations can propagate throughout the global industrial ecosystem.
Centralized infrastructure creates identifiable points through which access to intelligence can be regulated. Distributed intelligence offers a complementary architecture, potentially reducing dependence on individual providers and jurisdictions without eliminating legitimate regulatory obligations.
The Intelligence in Your Pocket
While governments and corporations construct enormous AI factories, another transformation is occurring. Apple’s Foundation Models framework enables applications to perform AI workloads on compatible devices without requiring internet connectivity. Advances in model quantization, specialized processors, and software optimization continue to expand the capabilities of smartphones, personal computers, and industrial edge devices.
The distinction is between where intelligence is developed and where it is applied. Training frontier models may require extraordinary centralized computing resources, but applying intelligence to many operational problems does not require continuous access to those resources.
We have seen this architectural evolution before. Mainframes concentrated computing capability within large institutions. Personal computers distributed that capability, and the internet subsequently connected independently managed computing environments. Distributed AI offers a similar opportunity to expand access to intelligence, improve privacy, and strengthen operational independence.
Local AI does not eliminate dependence on semiconductor manufacturers, operating systems, or software providers. It does, however, change the relationship between the individual and the infrastructure provider. For logistics, an industry characterized by mobile assets, geographically dispersed operations, and independently managed enterprises, this distinction has immediate practical significance.
Supply Chains Are Systems of Systems
A global supply chain consists of suppliers, manufacturers, warehouses, transportation providers, retailers, and customers, each possessing information, expertise, operational constraints, and decision-making authority that other participants do not. Its intelligence architecture should reflect that reality.
Consider an independent truck driver arriving at a distribution center with damaged freight. A locally deployed AI application could examine photographs, interpret shipping documents, retrieve cached handling procedures, and prepare an exception report. When connectivity becomes available, it could communicate relevant information to the carrier’s transportation management system and the consignee’s warehouse management system.
An inventory agent could evaluate downstream product availability, a transportation agent could assess recovery options, and an enterprise planning system could determine the broader financial and customer-service implications. Each participant contributes intelligence within its own domain of responsibility without necessarily surrendering all its proprietary information to a centralized platform.
Multiply this architecture across thousands of suppliers, manufacturing facilities, distribution centers, and transportation providers. The result is not one enormous AI application, but a network of interoperable intelligent systems. A small manufacturer or independent owner-operator could gain access to analytical capabilities historically associated with substantial enterprise software investments.
Manufacturing has understood this architectural principle for decades. Distributed control systems allow local controllers to manage individual industrial processes while supervisory systems coordinate broader operating objectives. A manufacturing facility does not transmit every sensor measurement to corporate headquarters and await instructions before adjusting a motor.
Supply chain AI should follow a similar systems engineering philosophy, while recognizing that probabilistic AI recommendations are not substitutes for validated industrial control and safety systems.
Centralized Governance, Distributed Execution
In my ARC Advisory Group white paper, AI in the Supply Chain: Architecting the Future of Logistics with A2A, MCP, and Graph-Enhanced Reasoning, I explored several technologies that can support this architecture.
Agent-to-Agent communication enables coordination among independently developed agents. The Model Context Protocol provides standardized access to information and operational tools, while Graph RAG supports reasoning grounded in relationships among suppliers, products, facilities, and transportation networks. Collectively, these technologies help separate operational intelligence from individual models, applications, and computing platforms.
They do not eliminate the need for data governance, mathematical optimization, secure identities, authorization, or human accountability. Enterprise planning systems must still resolve competing network-wide objectives, while local agents operate within defined financial, safety, and operational constraints.
Supply chain executives should evaluate their AI investments accordingly. Can critical operations continue when an external AI service becomes unavailable? Can models be replaced without rebuilding enterprise workflows? Can agents collaborate across organizations without surrendering proprietary information? Which decisions require centralized optimization, and which should be made closer to the point of execution?
These questions should become part of enterprise architecture, technology procurement, and business continuity planning. An enterprise that outsources its intelligence without preserving control over its data, interfaces, business rules, and decision-making authority may discover that it has outsourced a significant portion of its operational independence.
Who Will Control Intelligence?
The Bank of NVIDIA, sovereign AI, and the intelligence in your pocket represent different but potentially complementary visions of the future. Centralized infrastructure will remain essential, governments will continue pursuing technological independence, and increasingly capable intelligence will move closer to the people and machines that create economic value.
Huang’s legacy is particularly relevant because he helped advance both centralized computing and the technologies that make distributed intelligence possible. His departure, and the subsequent involvement of government officials in the debate over NVIDIA’s future, bring into focus the relationship between technological architecture and the distribution of economic power.
The strategic challenge for supply chain leaders is to capture the capabilities of centralized AI without creating unnecessary dependencies that compromise operational sovereignty.
Supply chain management has long demonstrated that coordination does not require common ownership and that effective governance does not require centralized execution. Artificial intelligence should extend those principles into the intelligent economy rather than undermine them.
The future of supply chain intelligence may belong not to those who concentrate the most computing power, but to those who most effectively connect intelligence distributed throughout the network.
Featured Logistics Research
Independent ARC Advisory Group research and Logistics Viewpoints analysis.














