Field report S-9 / L2 · sheet /deals

M&A & DealsFiled from the territory

Bajaj Finance, Pennant Invest $4 Million in Soket AI

Bajaj Finance and Pennant invested $4 million in Soket AI, funding sovereign models, on-premises deployments and banking agents for KYC, collections, compliance and fraud detection.

Grid
S-9 / L2
Surveyed
November 19, 2026
Reading
4 min read
Correspondent
Sophie Lindqvist
Bajaj Finance, Pennant Invest $4 Million in Soket AI
Fig. S-9 / L2Bajaj Finance, Pennant Invest $4 Million in Soket AI — nattanan23 / pixabay

Field notes

  1. Bajaj Finance and Pennant invested about $4 million in Soket AI on 18 November 2026.
  2. Bajaj Finance committed about $2 million through Finserv Intelligence; Pennant added $2 million.
  3. Soket AI’s EKA-1 has 26 billion parameters and focuses on mathematics, coding and reasoning.
  4. The roadmap targets more than 100 languages, including Indian and Global South languages.
  5. Pennant plans banking-specific agents for KYC, onboarding, collections, compliance and fraud detection.

By Sophie Lindqvist

Bajaj Finance Ltd and Pennant Technologies invested about $4 million in Soket AI on 18 November 2026. Bajaj Finance committed about $2 million through Finserv Intelligence, Bajaj Finserv’s applied research initiative, while Pennant, a lending-technology firm already backed by Bajaj Finance, added $2 million in the same round.

Soket AI is a Bengaluru-based Indian research company founded in 2019 and led by founder and CEO Abhishek Upperwal. It develops sovereign large language models (LLMs, text-generating AI systems trained on broad datasets).

Why is a lender funding sovereign AI?

The round links one of India’s largest non-bank lenders directly to homegrown research into foundation models, or general-purpose AI systems trained on broad data. Funding will support indigenous AI for uses where data security, reliability and technological sovereignty are critical.

The partnership is also expected to open applied AI research, enterprise applications and capability building using Bajaj Finance’s industry experience and Soket AI’s research. It sits within Bajaj Finance’s wider technology programme, including its core lending transformation with Pennant, and extends the group’s AI push in lending during H1 FY27, the first half of fiscal year 2027.

Soket AI also receives support from the Government of India’s IndiaAI Mission. The announcement discloses no valuation, ownership split, funding allocation or delivery milestones, so the $4 million establishes strategic intent rather than commercial traction.

What can EKA-1 do?

  • EKA-1: Soket AI’s flagship Project EKA recently produced a 26-billion-parameter foundation model focused on mathematics, coding and reasoning.
  • Roadmap: Soket AI plans substantially larger models and support for more than 100 languages, including Indian and Global South languages.
  • Deployment: The company says its models can run in secure, controlled settings, including on-premises infrastructure operated within an organisation’s own systems.
  • Tooling: Soket AI has released Loop, an open-source agentic harness, or software framework for orchestrating AI agents on long-running tasks.

The announcement provides no independent performance benchmarks, evaluation results or confirmation that all 100 languages are already available. It also offers no operational evidence from live deployments, leaving the roadmap’s scale and reliability claims unverified.

How will Pennant use the models?

Pennant said the two companies will develop and evaluate Soket’s models, Loop and inference runtime, the software layer that executes a model. The systems will target banking agents, document processing and collections intelligence.

They also intend to jointly build banking-specific language models and AI agents for know-your-customer (KYC) checks, onboarding, collections, compliance and fraud detection. The plan makes collections an early test of agentic AI, or AI systems that execute multi-step tasks, in banking, financial services and insurance (BFSI).

However, the announcement gives no pilot schedule, transaction volumes or unit-cost evidence, so the commercial case remains unproven. The proposed use cases also extend beyond collections into regulated functions, where errors could carry financial and compliance consequences.

Why does data control matter?

Sanjiv Bajaj, Chairman and Managing Director of Bajaj Finserv Ltd, said: “the margin for error must be minimal” when AI becomes integral to critical information infrastructure. He argued that such systems must be “reliable, secure, auditable” and able to operate inside trusted environments.

Rajeev Jain, Vice Chairman and Managing Director of Bajaj Finance Ltd, said Soket AI’s reasoning-led models “can make advanced AI economically viable for financial services.” Jain also said on-premises deployment gives Bajaj Finance full control over data, while Soket AI’s Indian-language depth could help customers use a language they are comfortable with.

Neither assertion comes with a cost benchmark or deployment result, and the release does not explain how control affects model performance. The economic claim therefore needs validation against live banking workloads before lenders can treat viability as established.

As banks and India’s non-bank financial companies (NBFCs) move from pilots to production, AI is becoming the operating layer for financial institutions. Model-security concerns have also driven deals such as Protectt.ai’s stake in AIShield, intended to secure AI in banking applications.

What does this change for lenders?

For lenders, the practical change is a funded route from sovereign-model research to controlled banking deployments. The announcement does not establish changes in credit decisions, customer journeys or regulatory exposure because it supplies no production date, benchmark, unit-cost comparison or evidence that the proposed agents outperform existing systems.

The decisive evidence will be measured performance in KYC, collections, compliance and fraud detection, not the funding headline alone. Until then, the round marks an institutional test of sovereign AI, not a proven change in banking economics.

via bfsi.eletsonline.com (Source)

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

Correspondent covering money, funding and market shifts around corporate strategy and chief-executive leadership at CEO Frontier.

479 articles

Filed under

Adjacent sheets

« Previous articleNext article »

End of sheetS-9 / L2 · November 19, 2026