The AI boom has a dirty secret: it is creating a new class system in technology. On one side sit Google, Microsoft, OpenAI, and Nvidia — companies with the money, machines, and minds to build the future. On the other side sits everyone else, including most enterprises in India, watching their negotiating power shrink by the quarter.
This is not just a story about who builds the smartest chatbot. It is a story about supplier power, and what happens when critical business infrastructure ends up in very few hands.
The Three Bottlenecks That Matter
The AI industry runs on three scarce resources: specialised chips, elite research talent, and massive capital. Nvidia controls roughly 80 percent of the market for GPUs — the processors that train and run AI models. Microsoft has poured over $13 billion into OpenAI and locked up its technology for Azure. Google is racing to catch up with its own chips and models while sitting on one of the world’s largest talent pools in AI research.
For enterprises, this concentration creates a chain of dependencies. You cannot easily train a custom model without Nvidia hardware. You cannot access GPT-4 outside Microsoft’s ecosystem. You cannot match Google’s research output without Google’s budget. Each link in this chain gives suppliers more leverage over your costs and your roadmap.
The talent picture is equally stark. Top AI researchers command salaries that exceed what most Indian IT services firms pay their CEOs. These researchers cluster at a handful of labs — OpenAI, Google DeepMind, Anthropic, Meta AI — creating a brain drain that leaves everyone else scrambling for second-tier expertise.
What This Means for Procurement
If your enterprise uses AI through Azure OpenAI Service or Google Cloud’s Vertex AI, you have already made a bet on a single supplier. That bet comes with switching costs that will only grow as you build more workflows around proprietary APIs and model behaviours.
Vendor lock-in in AI is stickier than traditional software licensing. Your prompts, fine-tuning data, and integration code are all optimised for a specific model’s quirks. Moving to a competitor means rewriting, retesting, and often retraining — a project that can take months and deliver unpredictable results.
Pricing power follows concentration. OpenAI has already raised API prices for its most capable models. Cloud providers bundle AI services with compute and storage, making it harder to compare true costs. For Indian enterprises managing tight IT budgets, these bundled deals can obscure how much you are actually paying for intelligence versus infrastructure.
The Talent Squeeze Hits Home
India produces more engineering graduates than any country except China. Yet the AI talent that matters — researchers who can build and improve foundation models — remains scarce globally and almost non-existent locally at the cutting edge.
This creates two problems for Indian enterprises. First, hiring decisions get harder. Do you pay global rates to attract AI talent, or do you accept that your team will work with models rather than build them? Second, your strategic options narrow. Without deep in-house expertise, you become a consumer of AI rather than a creator, dependent on whatever your vendors choose to release.
Some Indian IT services giants are investing heavily in AI training programmes. TCS, Infosys, and Wipro have each announced plans to upskill tens of thousands of employees. Whether these programmes produce genuine AI capability or just familiarity with existing tools remains to be seen.
Hedging Your Bets
Smart procurement in a concentrated market means building optionality. Open-source models from Meta’s Llama family and Mistral offer genuine alternatives for many use cases, though they require more in-house expertise to deploy and maintain.
Multi-cloud strategies help, but only if you architect for portability from the start. Abstraction layers and model-agnostic frameworks add complexity but reduce the cost of switching later. Indian cloud providers like Jio and Yotta are building AI infrastructure that could offer local alternatives, though their model ecosystems remain limited.
Partnerships with smaller AI vendors and startups can also diversify risk. Companies like Cohere and AI21 Labs offer commercial models outside the big three ecosystems. Indian startups like Sarvam AI are building models optimised for local languages and contexts.
What This Means for You
The AI vendor landscape will not become more competitive anytime soon. Capital requirements are rising, not falling. The companies with today’s advantages are reinvesting their profits to widen the gap.
For CIOs and CTOs in India, the practical response is defensive. Audit your current AI dependencies and map the switching costs honestly. Build procurement frameworks that account for concentration risk, not just feature comparisons. Invest in talent that understands multiple platforms, not just one vendor’s certification programme.
The AI gold rush is real. But the gold is flowing upward, to the companies that own the mines. Your job is to make sure your enterprise does not end up paying whatever the mine owners decide to charge.