Is the Era of “More Chips” Ending? Samer Choucair Reads the Major Shift in the AI Race
Investment leader Samer Choucair said the artificial-intelligence infrastructure boom has entered a decisive phase after Nvidia’s quarterly revenue surpassed $96 billion, driven by data-center demand, while the company also raised system prices and expanded long-term memory-purchasing commitments to unprecedented levels.
Choucair said Chinese open models, led by DeepSeek and Qwen, have demonstrated that smarter software can materially reduce memory and processor requirements. He noted that DeepSeek showed the potential to cut memory requirements by roughly 96% in certain workloads, arguing that this has begun to reprice the value of each unit of compute and shift part of the competitive advantage away from chip vendors toward software developers capable of extracting greater efficiency from existing hardware.
From Buying Silicon to Maximizing Compute Efficiency
Samer Choucair said Silicon Valley spent years compensating for software inefficiencies by buying more graphics-processing units, supported by enormous capital budgets. At the same time, the rapid evolution of Nvidia’s architectures made rewriting millions of lines of code costly in both time and engineering resources.
Choucair added that accelerator-server price increases of more than 15%, pressure on high-bandwidth memory supply, and Nvidia’s long-term commitments running into hundreds of billions of dollars have exposed the limits of simply absorbing computational waste.
Investors, he said, had largely priced the AI boom as a linear function of capital expenditure by Microsoft, Amazon, Alphabet, and Meta. That relationship alone is no longer sufficient.
Ignoring inference efficiency, Choucair warned, could obscure a decline in the marginal return generated by every additional chip deployed.
Nvidia: Strong Demand Meets Pressure From China
Samer Choucair said Nvidia’s data-center revenue reached approximately $89 billion in its latest quarter, more than doubling year over year, while the company projected revenue of roughly $108 billion for the following quarter, excluding China data-center sales because of regulatory uncertainty.
Choucair noted that Amazon has arranged to deploy roughly two million additional Nvidia chips, while market estimates put combined 2026 data-center spending by the largest U.S. cloud operators at approximately $800 billion.
He said sales of Nvidia’s H200 chips into China represented less than 1% of data-center revenue, while the company wrote down approximately $400 million of excess H200 inventory after actual demand came in below the level implied by U.S. export licenses.
At the same time, Nvidia’s historical market share in China, once above 90%, has declined as Huawei, Cambricon, and other domestic suppliers have expanded their positions.
China Is Rewriting the Production Equation
Samer Choucair said U.S. restrictions have not removed China from the AI race. Instead, they have pushed Chinese companies to compress models, optimize attention mechanisms, reduce memory consumption, and rely more heavily on lower-cost domestic hardware.
The competition, Choucair said, is increasingly developing between a U.S.-led ecosystem that dominates frontier models, advanced chips, and CUDA, and a Chinese ecosystem focused on open-weight models, distillation, and lower-cost inference.
He pointed to the growing global presence of Qwen, DeepSeek, and GLM during 2026, as well as Nvidia’s support for Chinese open models such as DeepSeek V4 Flash and Qwen, even as the company continues to warn that additional U.S. restrictions could affect its ability to support open-source Chinese applications.
Where Capital Is Moving
Choucair said institutional capital is increasingly being distributed across several parts of the AI value chain rather than concentrating exclusively on leading-edge chips.
One major area remains semiconductors, high-bandwidth memory, and advanced packaging. Another is software designed to optimize compute scheduling and cluster efficiency, an area underscored by Nvidia’s reported multibillion-dollar spending on strategic software capabilities and licensing.
Capital is also flowing toward Chinese semiconductor platforms and lower-cost inference technologies, as well as toward the physical infrastructure required to support AI at scale, including electricity generation, power cables, cooling systems, and industrial real estate for data centers.
Samer Choucair said Gulf funds do not need to replicate Silicon Valley’s investment model. Their advantage, he argued, lies in combining energy availability, digital infrastructure, and long-duration capital.
Saudi Vision 2030 and the Public Investment Fund create opportunities to invest across data centers, cloud infrastructure, and AI applications in sectors including energy, logistics, and financial services.
The Gulf Opportunity and the Risks
Choucair said artificial intelligence consumes electricity, water, land, and long-term financing, creating an opportunity for Saudi Arabia and the wider Gulf to become important nodes in the global AI supply chain even without manufacturing the world’s most advanced chips.
He cautioned, however, that investors face several significant risks. These include rising memory costs, tighter U.S. restrictions on China, rapid advances in low-cost inference, increasing dependence on debt financing for data-center development and the resulting sensitivity to interest rates, and the possibility that the technology industry continues fragmenting into two geopolitical ecosystems.
Choucair expects U.S. infrastructure demand to remain supportive over the next 12 to 18 months because the economics of training frontier models have not fundamentally broken down.
Over time, however, he expects a portion of value creation to migrate from pure chip intensity toward software intensity.
For institutional investors, he said the discipline is increasingly about maintaining selective exposure to the AI supply chain while closely monitoring memory margins and server pricing, tracking the adoption of open models outside China, and directing Gulf capital toward energy, data centers, cloud computing, and applications that reinforce broader economic-diversification strategies.
Samer Choucair concluded that Nvidia has begun to reveal the limits of an investment equation built primarily around spending more on hardware, while China has forced the market to become more serious about writing smarter code.
“In 2026, a capital allocator will not be judged by how close they are to the newest chip,” Choucair said. “They will be judged by their ability to distinguish between compute that is simply purchased and governance that compounds value.”