The AI Capex Pivot: Meta Sells Power, Palantir Complains, and Zhipu Becomes Silicon Valley’s New Obsession

The era of "blind" investment in Artificial Intelligence (AI) is drawing to a close. For nearly three years, markets fueled Big Tech with billions of dollars, with the sole requirement being the purchase of as many Nvidia H100 processors as possible. Today, in the summer of 2026, the narrative of Capital Expenditure (Capex) is undergoing a violent transformation. It is no longer enough to own computing power; you must know how to sell it, how to rationalize it, and, above all, how to survive the emerging competition from the East.

Meta and the "Compute-as-a-Service" Strategy

Mark Zuckerberg, once criticized for his obsession with the Metaverse, is now emerging as the most unexpected player in the computing power market. Meta is no longer just a social media company. With the massive accumulation of GPUs it began in 2024, the company has transformed into a de facto infrastructure provider. Meta's recent move to "rent" or offer computing power to strategic partners for training Llama models marks a paradigm shift.

Instead of being limited to internal use, Meta is leveraging its surplus power to create an ecosystem where Llama is the operating system of AI. However, this is causing friction. Meta's "open-source" strategy, while appearing philanthropic, is actually an attempt to turn computing power into the new commodity, undermining the closed models of OpenAI and Google. When computing power becomes a product for sale, value shifts from the model itself to the ability to execute it at scale.

Palantir’s Discontent and the "Application Gap"

On the other side of the Atlantic, Alex Karp’s Palantir is expressing a sharp discontent that resonates throughout Wall Street. Palantir's criticism is not about the technology itself, but about the disconnect between massive Capex spending and real business value. Karp has repeatedly emphasized that companies are "burning" capital on computing power without a clear strategy for how this power will translate into operational efficiency.

For Palantir, AI is a decision-making tool for critical infrastructure and the battlefield, not a text-generation toy. The company's "grumbling" reflects a broader investor fatigue: AI Capex has reached levels reminiscent of the late 1990s telecom bubble. If companies continue to buy compute power they cannot productively utilize, the correction will be painful. Palantir demands a shift toward "Applied AI," where value is measured by outcomes rather than model parameters.

Zhipu AI: The New Star Silicon Valley Fears

While the West bickers over spending, a Chinese company, Zhipu AI, has achieved the unthinkable: becoming the "talk of the town" in Silicon Valley. Zhipu, born out of the laboratories of Tsinghua University, is not just another competitor. It is proof that China can produce world-class models (such as ChatGLM) despite US chip export restrictions.

The rise of Zhipu AI changes the geopolitical narrative. Silicon Valley investors are watching with awe at Zhipu's ability to achieve high performance with fewer computational resources—a necessity born out of sanctions. This creates a new competition: the West has the power (Compute), but the East is developing the efficiency (Efficiency). Zhipu has become the "beacon" for how AI can be developed in a resource-constrained environment, challenging American companies to reconsider their Capex waste.

The Need for a New Narrative

The conclusion is clear: the AI Capex story urgently needs a new script. The old script of "build it and they will come" no longer convinces anyone. The market is entering a maturity phase where computing power will be treated like electricity or oil—a resource that must be managed wisely.

• Meta is leading the way in commoditizing infrastructure.
• Palantir serves as a reminder of the need for practical utility.
• Zhipu AI proves that innovation knows no borders, even under sanctions.

From now on, the winners will not be those with the largest data centers, but those who manage to bridge the gap between computing power and human need, while keeping their balance sheets in order. The geopolitical chessboard of AI is becoming more complex, and Silicon Valley must learn from its rivals if it wishes to maintain its hegemony.