3 weeks ago
Enterprise AI Spend Gets Reality Check as Token Costs Collapse
Using artificial intelligence is getting cheaper every year - the price of processing a million tokens has dropped about 10 times each year.
But big companies are not just buying more AI because it is cheap.
Instead, leaders are asking where the AI is being used and whether it is worth the money.
Some companies now make teams get permission before they can use an AI model.
Every team also gets a monthly budget, and every use of AI is tracked.
In one example, a company cut a planned AI cost from $2,000 to $400 just by checking the plan first.
Tech experts say cheaper models have already lowered software development costs by up to 30% on some projects.
Many Indian companies are testing open-weight AI models for simple tasks like customer support.
These models can reduce costs by 30-70%, but companies worry about safety rules and reliability, so they still use the most powerful models for complex work.
The cost of processing a million tokens is falling by roughly 10X every year, with DeepSeek V4 Flash at $0.14, OpenAI's GPT-5.6 Luna at $0.20 and Meta's Muse Spark 1.2 at $1.25 per million input tokens.
Enterprise leaders are shifting the debate from how much AI companies can afford to where it should be used and whether returns justify the spend.
AI usage is getting budget guardrails: at Eightfold AI, no feature can use an AI model until approved, and one projected cost was cut from $2,000 to $400 before the experiment began.
ShepHertz CTO Anuraag Kochhar says falling token prices have cut software development costs by up to 30% on some projects, and 25-40% of Indian enterprises are piloting open-weight models.
Fewer than two in three organisations actively monitor AI usage with clear budget limits, and AI consumption is increasingly treated like a budget request that must earn its place.
- Who
- Enterprise technology leaders, including Ed Huang, CTO and cofounder of TiDB, and Anuraag Kochhar, CTO of ShepHertz, along with engineering teams at companies like Eightfold AI.
- What
- Companies are adding budget approvals, usage caps and value tracking to AI spending even as token prices collapse, moving the debate to governance, budgets and balance sheets.
- Where
- Global enterprises, with a particular focus on Indian enterprises and the Indian AI market.
- When
- No specific date is given; the analysis reflects current AI market conditions in the newsletter edition.
- Why
- Because token prices have fallen so far, the question is no longer affordability but proving that AI usage produces business value and justifies its place on budgets and balance sheets.
Cheaper AI Should Mean More AI
Value-First AI Governance
What falling token prices should mean
Cheaper AI Should Mean More AI
Cheaper technology is naturally treated as a licence to use more of it, so collapsing token prices should drive wider, open-ended AI usage across enterprises.
Value-First AI Governance
Leaders are deliberately avoiding token-hoarding; the goal is to prevent tokens from being spent on work that does not matter, with approvals, monthly budgets and proof that every token produces value.
Open-weight vs frontier models
Cheaper AI Should Mean More AI
Open-weight models from China and elsewhere should be used selectively for high-volume, less sensitive workloads; 25-40% of Indian enterprises are piloting them, and inference costs can drop 30-70%.
Value-First AI Governance
Premium frontier models remain necessary for complex reasoning and sensitive work; compliance, data governance and reliability barriers limit wider adoption of open-weight models.
Key facts
- DeepSeek V4 Flash price
- $0.14 per million input tokens
- OpenAI GPT-5.6 Luna price
- $0.20 per million input tokens
- Meta Muse Spark 1.2 price
- $1.25 per million input tokens
- Annual token price decline
- Roughly 10X per year
- Example cost cut (Eightfold AI)
- Projected cost reduced from $2,000 to $400 (80% cut)
- Software development cost savings
- Up to 30% on some projects (per ShepHertz CTO)
- Indian enterprises piloting open-weight models
- 25-40%
- Inference cost reduction from open-weight models
- 30-70%, depending on the use case
Quotes
Thiyagaraj T
Director of engineering at AI talent platform Eightfold AI
“a team that wants to build something with AI must explain what it is making, which model it will use and what it will cost per month.”
inc42.com
“Do we know where our tokens are going, and are they producing enough value to justify their usage?”
inc42.com
Sources
Related news
Uber CTO Says 'Tokenmaxxing' Era Ending as AI Costs Fall
Enterprise AI’s Next Phase Focuses on Managing Model Evolution
Anthropic Launches Claude Fable and Mythos 5.1
CEO Says AI Agent's $1,000 Weekend Bill Isn't Biggest Problem
Palantir CEO: AI Firms Trying to 'Drug Addict' Businesses
Social Media Companies Intensify Efforts to Identify Child Users
Yorkville America Expands Beyond Trump-Linked ETFs With AI Strategy
Meta’s $18-Billion Child Safety Settlement Faces Praise and Criticism
School Awards Honour Courage, Commitment and Educational Excellence
AI Expansion Puts Jamnagar and Thane Under Water Stress
