In the competition to seize control of the forthcoming technological landscape, artificial intelligence firms are investing vast sums, resulting in what some perceive as boundless expenditure—often misinterpreted as extravagance. For keen observers monitoring the advancement of technology and its potential for investment, this "financial loss" signifies not wastefulness but a calculated struggle for dominance over the most essential asset: computational capability. Underneath each report of escalating expenses lies a critical contest, where the victor claims not only market position but also the trajectory of AI advancement itself.

Apart from the excitement surrounding AI processors, the true financial strain stems from High-Bandwidth Memory (HBM). In leading AI processors such as NVIDIA's B200, HBM constitutes nearly fifty percent of the overall expense, significantly surpassing the cost of the GPU core itself, as demand significantly exceeds the supply from the three primary producers, driving prices to unprecedented levels.

Advanced Packaging: An Expensive Requirement

State-of-the-art AI processors necessitate intricate CoWoS packaging for the integration of HBM and GPU cores, a method that contributes an additional third to the overall cost of the chip due to the challenging manufacturing difficulties and low production yields. As processors enhance in power, the intricacy of this packaging—and the associated costs—only escalates, compelling firms to maintain high spending to remain competitive.

Token Black Holes: Concealed Operational Expenses

AI applications like OpenClaw generate obscure "token black holes"—every action, even straightforward tasks, requires substantial computational resources, with some users indicating that daily costs surpass $100 for personal usage, and are much higher for corporate applications. This ongoing token expenditure transforms operational expenses into an unending drain.

To diminish dependence on costly cloud computing power, firms are investing significantly in model slimming techniques—distillation, quantization, and pruning—to condense large models without sacrificing effectiveness. Although this research and development-heavy approach is essential, it contributes to the financial burn rate as teams strive to refine efficient, deployable models.

Hardware Obsolescence: The Rate of Value Depreciation

AI hardware experiences "rapid depreciation"—a one billion dollar chip cluster can become obsolete within just two years as new processors are released, forcing companies to perpetually replace their infrastructure to meet the demands of their models, a cycle that consumes capital but is crucial for maintaining a competitive edge.

This ongoing process of replacing hardware imposes a significant financial strain on both companies and affluent technology investors. Additionally, the swift pace of obsolescence results in substantial electronic waste and poor resource distribution, transforming sophisticated computing frameworks into temporary assets instead of enduring long-term investments, while simultaneously increasing the urgency to stay abreast of continuous AI advancements.

Security Investments: Protecting Against Concealed Risks

With the increasing power of AI tools, security vulnerabilities become expensive liabilities. Firms are compelled to invest heavily in addressing these weaknesses—more than sixty percent of OpenClaw instances possess exploitable weaknesses—and safeguarding user information, which is a vital cost to prevent reputational harm and fines from regulatory bodies.

For affluent audiences, this spree of spending represents not disorder but a strategy for future worth. The AI firms incurring expenses today are engaged in a battle for control of the computational resources that will shape innovations in the future, transforming their strategic expenditures into more than just exorbitant spending—they represent investments in a technology-driven tomorrow.

（Writer：Tommy）