An artificial intelligence company does not need to build its own data centres in a country to make it central to its ambitions. It needs customers, computing capacity and partnerships that can help businesses move from experimenting with AI to using it in everyday operations.
That is the bet Anthropic is making on India.
On August 13, the American AI company appointed Rajat Pandit, an AI infrastructure executive, as its Head of Applied AI for India. He was its third senior India hire in seven months, following managing director Irina Ghose in January and head of policy and external affairs Amlan Mohanty in April. The appointments followed the opening of its Bengaluru office, its second in Asia after Tokyo.
Anthropic is now looking to turn interest in its Claude models into enterprise adoption. India is already an important market: the company described it as the second-largest market for Claude.ai when it announced its Bengaluru expansion in February 2026. It also reported that its India revenue run-rate had doubled since its October 2025 market expansion announcement.
The opportunity extends beyond individual users. Anthropic wants Indian businesses to incorporate Claude into software development, internal workflows and other complex operations. But the company is pursuing this ambition at a time when the economics, geopolitics and security of artificial intelligence are becoming increasingly complicated.
Anthropic's local presence does not mean it is building and operating its own Indian data centres. Instead, it relies on cloud infrastructure supplied by Amazon Web Services (AWS).
Its models are available through Amazon Bedrock, AWS's platform for accessing foundation models. Anthropic develops Claude; AWS provides the computing infrastructure required to run it.
The arrangement became more significant on September 29, when AWS announced that Claude Opus 5, Sonnet 5 and Haiku 4.5 would be available on Amazon Bedrock with India-region inference through its Mumbai and Hyderabad regions.
For businesses handling sensitive information, local inference can help address concerns about where data is processed. It can also make it easier to meet certain internal and regulatory requirements. However, local inference does not mean that model development, training or every aspect of the service takes place in India.
The arrangement also reflects the economics of the AI industry. Developing frontier models requires enormous computing capacity, specialised chips and capital. Partnering with established cloud providers allows model developers to access infrastructure without independently building it in every market.
Anthropic's own funding needs underline the scale of the business. In May 2026, it raised $65 billion at a post-money valuation of $965 billion, saying the funds would support computing capacity, safety research and growing demand for Claude.
Its India strategy, therefore, is not simply about finding more users. It is about establishing relationships with businesses that could eventually make AI a part of their core operations.
The biggest commercial challenge for Anthropic is that powerful AI is becoming cheaper to access.
Chinese companies such as DeepSeek have demonstrated that capable reasoning and coding models can be offered through open-weight systems at comparatively low costs. Unlike closed proprietary models, open-weight models allow developers to access model parameters and deploy or customise them, subject to licensing conditions.
For enterprises, this changes the calculation. If a cheaper model can perform a particular task sufficiently well, there may be little reason to pay a premium for a proprietary alternative.
Anthropic must therefore demonstrate that its models offer enough additional value in performance, reliability, security and enterprise support to justify their cost.
This is particularly relevant in India, where businesses are likely to weigh AI investments against the productivity gains they can deliver. The competition is no longer simply between American and Chinese companies; it is also between different business models for delivering AI.
For Anthropic, building deeper enterprise relationships in India could help establish its position before businesses settle on which models offer the best combination of performance and price.
The commercial challenge is only part of Anthropic's problem. Its relationship with the US government has also become increasingly complicated.
Founded in 2021 by former OpenAI researchers, Anthropic initially built its identity around AI safety and the responsible development of powerful models. But disagreements over military and surveillance applications have put that positioning to the test.
In March 2026, the Pentagon designated Anthropic a supply-chain risk to national security following a dispute over the conditions under which its models could be used by the US military. The disagreement centred on restrictions the company wanted to maintain around applications such as mass domestic surveillance and autonomous weapons. Anthropic challenged the designation in court, and the legal dispute has continued.
The conflict raises a difficult question: who decides how powerful AI models can be used once they become part of critical government operations?
The disagreement escalated in June, when the US government issued an export-control directive restricting foreign-national access to Anthropic's Fable 5 and Mythos 5 models. Anthropic said it had not been given specific details of the national-security concern. Unable to enforce the restriction selectively, it suspended access to both models globally. The restrictions were subsequently lifted, with access to Mythos remaining limited to selected organisations.
The episode exposed the vulnerability of a global AI business to government restrictions that can disrupt access to its technology across borders.
There are also questions about what these increasingly capable models can do. On July 30, Anthropic disclosed three incidents during security evaluations in which its models accessed the internet despite being expected to operate in isolated environments. The models subsequently gained unauthorised access to the production infrastructure of three organisations. Anthropic attributed the incidents to a misconfiguration in the testing environment.
The incidents do not establish that publicly available Claude models are routinely breaking into systems. But they highlight the difficulty of safely testing technology that can identify software vulnerabilities and potentially exploit them.
The same capabilities that help organisations strengthen cybersecurity can also be misused. For businesses considering large-scale AI deployment, reliability and safeguards are becoming as important as performance.
India's policy environment is another factor in Anthropic's expansion.
The Union Budget 2026 proposed a tax holiday until 2047 for eligible foreign companies providing global cloud services using Indian data-centre infrastructure, subject to specified conditions. The incentive is part of a broader effort to attract investment in the country's cloud and data-centre ecosystem.
India's data protection framework also matters. The Digital Personal Data Protection Act permits cross-border transfers of personal data, subject to restrictions the government may notify. Sector-specific rules can be stricter. The Reserve Bank of India, for instance, requires payment-system data to be stored in India.
Local inference can help address some concerns for regulated businesses, but it does not automatically guarantee compliance. Companies must still consider the nature of the data, applicable sectoral rules and how AI systems are integrated into their operations.
There is also a physical cost to the AI infrastructure boom. Data centres require substantial electricity, cooling capacity, land and water. As India attracts more investment, these requirements will have to be balanced against local resource needs and community concerns.
Anthropic's India expansion also has a cybersecurity dimension through Project Glasswing, a restricted programme that gives selected organisations access to Claude Mythos Preview to identify software vulnerabilities.
The initiative is not a general commercial release of Mythos. Access is limited to selected organisations working on cybersecurity and critical software infrastructure.
In June 2026, Anthropic announced that the programme was being extended to approximately 150 additional organisations across more than 15 countries. India was among the countries included in this international expansion. Reports indicated that a small number of Indian government agencies and private companies had received access, with discussions underway to extend it to cybersecurity and AI teams at some of the country's largest firms.
For India, participation could help security teams identify weaknesses in critical digital systems, including those supporting banking, telecommunications and public services. But the programme also raises questions about who gets access to such powerful technology and how dependent institutions will become on foreign AI providers.
Anthropic's India ambitions bring together commercial opportunity, infrastructure partnerships and national-security concerns. India offers a growing digital economy, a large developer community and an expanding data-centre ecosystem. For the company, it is an increasingly important market for enterprise AI.
But becoming a major consumer of AI is not the same as becoming a country that shapes its development. Access to foreign models does not automatically translate into ownership of the underlying technology, computing infrastructure or expertise.
Anthropic is betting that India will become central to the next phase of AI adoption. The bigger question is whether India can turn that access into technological capabilities of its own.