DeepSeek V4 and the New Economics of AI Tokens
The new AI pricing battle is a shift in how enterprises evaluate AI models, as buyers move from chasing maximum capability to managing the real cost of every processed token across large-scale workloads. This change centers on AI model token costs and is pushing vendors to rethink enterprise AI subscriptions and how they charge for usage. DeepSeek V4 pricing has become the reference point in many boardrooms because its lower token rates are drawing enterprise workloads away from OpenAI and Anthropic. Axios reporting, cited in industry coverage, says Microsoft is weighing a self-hosted DeepSeek V4 for its Copilot Cowork service as it moves to a metered, token-based architecture, signaling that cost now rivals performance in strategic importance for major buyers.
Microsoft’s Pivot Signals Enterprise Defection to Cheaper Models
Microsoft’s reported move to consider DeepSeek V4 for Copilot Cowork highlights how lower-cost models can disrupt long-standing supplier relationships. Copilot Cowork combines enterprise productivity features with advanced AI, and shifting its engine to a self-hosted DeepSeek V4 would reduce dependence on OpenAI and Anthropic for expensive workloads. Enterprises are especially sensitive to runaway usage when agentic workflows and long coding prompts inflate token counts and bills. The move toward a metered model, where customers pay by total tokens used instead of a flat rate, makes DeepSeek V4 pricing even more attractive. According to Wccftech’s summary of the Axios scoop, soaring token costs are already pushing customers away from OpenAI and Anthropic. For large buyers, this is not about testing a niche model; it is about re-platforming everyday AI work to protect budgets.
OpenAI and Anthropic Weigh Price Cuts as Tokenmaxxing Backfires
As DeepSeek undercuts the market, OpenAI and Anthropic are being pushed toward OpenAI price cuts and broader revisions to enterprise AI subscriptions. The Wall Street Journal, cited by Mashable, reports that OpenAI is debating product-wide reductions in subscription usage costs, including potential cuts for high-demand tokens, to hang on to price-sensitive customers. This is a reversal from the recent era of “tokenmaxxing,” where many companies encouraged heavy AI use only to see budgets spiral. One quoted example: Uber burned through its entire AI budget for 2026 in four months after incentivizing staff to increase usage. Executives now complain that high AI model token costs are unsustainable, and investors are cooling on the sector’s economics. OpenAI’s exploration of an IPO while considering cuts underscores how pricing pressure is colliding with the need to show strong margins.
How Cost-First Procurement Is Rewriting Vendor Negotiations
For many enterprise buyers, the performance gap between top-tier models is narrowing, making total cost of ownership a decisive factor. DeepSeek V4 pricing aligns with a demand for predictable, lower AI model token costs, forcing procurement teams to challenge the premiums charged by incumbents. As Microsoft moves Copilot Cowork to a metered token model, customers will scrutinize not only per-token rates but also caps, throttling, and fine print that limit real usage. Negotiations are shifting from feature checklists to questions like: which workloads can be safely moved to a cheaper model, and how much budget does that free? This is reshaping multi-vendor strategies: many enterprises now plan to keep a premium model for peak tasks while routing routine or coding-heavy work to more economical engines such as DeepSeek V4.
Intensifying Global Price Competition and the Road Ahead
The rise of DeepSeek V4 intensifies price competition and challenges the market dominance of established Western AI leaders. DeepSeek’s latest funding round of USD 7.4 billion (approx. RM34.1 billion) at a USD 50 billion (approx. RM230.5 billion) valuation shows that investors see a large opportunity in undercutting premium AI vendors on cost. At the same time, regulatory tensions and security concerns over advanced models create political risk around cross-border AI adoption, adding another layer to enterprise decision-making. Anthropic’s recent public listing plans alongside OpenAI’s IPO filing point to an industry racing toward public markets while facing a margin squeeze from cheaper challengers. For enterprises, the likely outcome is a more competitive landscape where AI model token costs fall across the board, and price transparency becomes as important as raw model performance.
