TechSparks 2026 Day 2: From Krishna Byre Gowda’s Bengaluru blueprint to Stripe’s global playbook, and more

Day 2 examined how India can build stronger cities, companies, and technologies. The main themes were governance, Bengaluru’s infrastructure, global-first startups, AI economics, and deeptech. Founders, investors, policymakers, and technologists discussed both domestic needs and international opportunities.

Krishna Byre Gowda focused on using technology to improve public services and reduce corruption. He also outlined Bengaluru’s metro expansion plans and wider-area planning. Manish Maheshwari discussed how AI lowers the cost and time needed to build companies for global markets. Stripe’s planned OpenRouter acquisition added a focus on controlling AI spending.

Deeptech investors linked India’s talent advantage with the geopolitical need for self-reliance in critical technologies. They said founders need patience, nerve, and an eight-to-10-year opportunity window. The next day was set to cover engineering education, AI sovereignty, climate innovation, and India’s AI opportunity.

Bengaluru’s metro plan has two milestones. The state government wants to expand the network from its current 96 km to 170 km over the next 18 months. Its longer-term ambition is much larger: reaching 400 km by 2037, the year Bengaluru turns 500.

The plan is part of a wider effort to manage the city’s growth. Planning is already underway for a 2,000 sq km area around Bengaluru. That is substantially larger than the 750 sq km covered by the Greater Bengaluru Authority. Gowda also said the government hoped to show visible progress toward a cleaner city within six months.

The targets show that transport is being treated as part of broader city planning. Expanding the metro could reshape how people move across Bengaluru, while the larger planning area reflects the city’s expanding footprint. The article does not provide project-by-project timelines for reaching 400 km.

OpenRouter is a system that routes AI queries to the most suitable model. In practice, companies can use it as a decision layer between their applications and different AI models. The goal is to match each request with an appropriate model instead of sending every query through one option.

The article describes Stripe’s agreement to acquire OpenRouter as a bet on helping companies control rapidly rising AI costs. This matters because AI usage is priced around the volume of model processing. Choosing a suitable model for each request can help companies manage that usage more carefully. The article does not specify the acquisition’s financial terms.

The deal reflects a broader change in startup economics. Maheshwari said some new-age companies spend more on tokens than salaries. If AI use continues expanding, tools that help select models and manage processing could become important parts of company infrastructure. The article presents OpenRouter’s routing capability as central to that opportunity.

The acquisition is aimed at a growing cost problem for AI companies. Using AI models requires processing requests, and the expense can rise quickly as products gain users or run more automated tasks. Stripe’s Manish Maheshwari said some new-age companies already spend more on tokens than salaries.

OpenRouter’s mechanism is query routing. It sends an AI request to the model considered most suitable for that query. This gives companies a way to manage model use rather than treating every request identically. Matching requests with models is presented as a way to help control the costs created by rapidly increasing AI activity.

The agreement also shows Stripe moving beyond ordinary payment infrastructure into a key challenge for AI businesses. The article does not provide the acquisition price or a detailed product roadmap. It does show why the issue matters: AI can make global companies faster and cheaper to build, while its usage costs can become a major expense.

Technology can improve government by making services easier to deliver and processes easier to monitor. Gowda said greater technology adoption could improve service delivery and lower corruption. Digital systems may also help narrow the trust gap between citizens and government by making administrative actions more visible and consistent.

He connected this idea to Bengaluru’s property records and taxes. The city has an estimated 35 lakh to 50 lakh properties, but only 23 lakh have registered khatas and just 18 lakh pay property taxes. These figures show that technology must support better records, compliance, and public administration, not merely add another digital interface.

Gowda cautioned that technology alone cannot solve systemic problems. Civic behaviour must also change. People and institutions still need to register properties, pay taxes, and follow rules. The government’s broader plans include metro expansion, regional planning, and cleaner-city results, showing that digital tools must work alongside policy and civic participation.

Deeptech matters because it can help India develop and control important technologies instead of depending entirely on external sources. The investors at TechSparks linked this opportunity to India’s talent advantage and a geopolitical push for self-reliance in critical technologies. That makes deeptech relevant to both business creation and national technological capability.

The panel included investors from Speciale Invest, Seafund, Avatar Venture Partners, and Kae Capital. Mayuresh Raut said there is still an eight-to-10-year window of opportunity. Vishesh Rajaram said “nerve and patience” distinguish founders who endure. These comments point to the time and persistence required for difficult technology ventures.

The article does not say that every deeptech company takes eight to ten years to build. It presents that period as an opportunity window. The implication is that founders and investors must be prepared for long development and commercialisation journeys, rather than expecting quick results. Patient capital therefore becomes especially important.

In AI, tokens are small pieces into which text or other input is divided for processing. A model reads input as tokens and produces output as tokens. Companies are commonly charged according to how many tokens their systems process, so token use grows when applications handle more requests, longer prompts, or longer answers.

The article gives the business consequence rather than a technical definition. Manish Maheshwari said that in some new-age companies, spending on tokens is higher than spending on salaries. A company serving many users can send large numbers of requests to AI models. Each request may involve both incoming and generated tokens, creating a substantial cumulative bill.

The comparison shows how AI changes a startup’s cost structure. Salaries are usually tied to a workforce, while model-processing costs can rise with product usage. Stripe’s OpenRouter acquisition is presented as a response to this challenge. The article does not provide token prices, usage volumes, or a specific savings estimate.