Will Bryk doesn’t want your typical Silicon Valley engineer. He wants the ones who got suspended from school for hacking the grading system. The ones who built something dangerous in their garage at 15. The ones other companies are afraid to hire.
That philosophy — recruit rebellious technical talent and point them at one of the hardest problems in artificial intelligence — has turned Bryk’s startup Exa into one of the most closely watched companies in the AI search space. And now, flush with fresh capital and expanding to Singapore, Exa is betting that its contrarian hiring strategy and API-first approach to search will let it outrun both entrenched incumbents and well-funded rivals.
Exa, founded in 2021 and based in San Francisco, builds AI-powered search technology designed not for consumers typing queries into a browser, but for AI systems themselves. Its core product is an API that lets large language models and AI agents retrieve real-time, high-quality information from the web — a capability that has become increasingly critical as companies race to build AI applications that need to access current data rather than relying solely on static training sets. The company’s search engine uses neural networks and embeddings to understand meaning rather than just matching keywords, returning results that are contextually relevant in ways traditional search engines often aren’t.
According to Business Insider, Exa recently closed a $400 million funding round that valued the company at $2.4 billion. Nvidia participated in the round, a signal that the GPU giant sees Exa’s approach to search as a key infrastructure layer for the next generation of AI applications. The funding will support Exa’s international expansion, starting with a new office in Singapore, and an aggressive hiring push targeting engineers with unconventional backgrounds.
Bryk, Exa’s CEO, told Business Insider that the company specifically seeks out engineers who have demonstrated a rebellious streak — people who’ve built things outside the system, challenged authority, or shown the kind of obsessive technical curiosity that doesn’t always fit neatly into a corporate résumé. “We want people who have a history of doing things they weren’t supposed to,” Bryk said, describing the company’s hiring ethos. It’s a deliberate cultural bet: that the kind of mind willing to break rules is also the kind capable of building something genuinely new.
This isn’t just founder mythology. Exa’s hiring process reportedly includes screening for candidates who’ve demonstrated independent, sometimes transgressive, technical initiative — side projects that pushed boundaries, contributions to open-source tools that solved problems no one asked them to solve, or even youthful hacking exploits that showed raw capability. The company believes this approach gives it an edge in a talent market where the most capable AI engineers are being courted by every major tech company on the planet.
The Singapore expansion is strategic on multiple levels. Southeast Asia has become a magnet for AI investment, with sovereign wealth funds, government incentives, and a growing pool of technical talent making the region increasingly attractive for startups looking beyond the Bay Area. Singapore in particular has positioned itself as a hub for AI development, offering regulatory clarity, strong infrastructure, and proximity to major Asian markets. For Exa, the move provides access to a new talent pool and positions the company closer to Asian enterprise customers who are building AI applications at scale.
But the expansion also reflects a broader reality about the AI search market: it’s getting crowded, and fast.
Perplexity AI, which has raised over $500 million and reached a valuation north of $9 billion, is building a consumer-facing AI search product that directly competes with Google. Tavily, another API-focused search startup, targets developers building AI agents. Google itself has been integrating AI-generated answers directly into its search results through its AI Overviews feature. And OpenAI has launched SearchGPT, signaling that the company behind ChatGPT sees search as a natural extension of its large language model business.
Exa’s differentiation lies in its focus on the machine-to-machine layer. Rather than competing for consumer eyeballs, the company is building infrastructure that other AI systems use to access the web. Think of it as search-as-a-service for artificial intelligence — a picks-and-shovels play in a market where everyone else is trying to build the mine. When an AI agent needs to look something up, verify a fact, or pull in fresh data, Exa’s API handles the retrieval. This makes the company less visible to end users but potentially more embedded in the AI stack.
The Nvidia investment underscores this positioning. Nvidia has been systematically investing in companies that form the infrastructure layer for AI — from cloud providers to data platforms to now search APIs. Jensen Huang’s company understands that the value of its GPUs depends in part on the quality of the software and data infrastructure built on top of them. A search API that makes AI applications more capable and accurate increases the overall value proposition of the AI hardware stack.
Exa’s technical approach is worth examining. Traditional search engines like Google rely heavily on link-based ranking algorithms descended from PageRank, combined with keyword matching and, increasingly, machine learning signals. Exa takes a different tack. Its search engine converts queries and web content into high-dimensional vector representations — embeddings — and finds results based on semantic similarity. This means a query phrased as a natural language statement or a complex multi-part question can return results that are conceptually relevant, even if they don’t contain the exact words used in the query.
For AI agents, this is a significant advantage. LLMs don’t formulate queries the way humans do. They generate natural language descriptions of what they need, and a search system optimized for that kind of input can deliver far better results than one designed for the terse keyword queries humans have been trained to type into Google over the past 25 years.
The $2.4 billion valuation places Exa in rarefied territory for a company that’s still relatively early in its commercial trajectory. But it also reflects the enormous expectations investors have for the AI infrastructure market. According to estimates from Goldman Sachs and other analysts, spending on AI infrastructure — including compute, data, and software tools — is expected to exceed $200 billion annually within the next few years. Search APIs that help AI systems access and process web data sit squarely in that spending stream.
Not everyone is convinced the bet will pay off. Critics point out that the major AI labs — OpenAI, Google DeepMind, Anthropic — all have the resources and incentive to build their own search capabilities in-house. If retrieval-augmented generation becomes standard practice for LLMs, the argument goes, the biggest players will simply internalize that function rather than relying on a third-party API. Exa’s counter is that search is hard — really hard — and that building a world-class search engine requires specialized expertise and infrastructure that even well-funded AI labs would rather outsource.
There’s precedent for this argument. Stripe didn’t get killed by banks building their own payment APIs. Twilio didn’t get crushed by telecom companies internalizing communications infrastructure. In enterprise software, specialized infrastructure providers often thrive precisely because their customers — even very large, very capable ones — decide that building the capability themselves isn’t worth the opportunity cost.
Whether Exa can follow that playbook depends on execution. And execution, in Bryk’s view, comes down to people.
The rebellious-engineer hiring strategy is more than a branding exercise. In a market where top AI talent commands compensation packages exceeding $1 million annually, startups need a compelling reason for engineers to choose them over Google, Meta, or OpenAI. Exa’s pitch is cultural: come work with other obsessively talented, rule-breaking builders on a problem that matters. It’s a bet that a certain type of engineer — the kind who chafes at bureaucracy and craves technical freedom — will find that offer irresistible, even if the cash compensation doesn’t match what a FAANG company would pay.
The Singapore office will test whether that culture can translate across geographies. Silicon Valley startups have a mixed record when it comes to international expansion, particularly when their competitive advantage is rooted in a specific cultural identity. Bryk has said he believes the rebellious-engineer ethos is universal — that there are brilliant, unconventional builders everywhere, not just in San Francisco — and that Singapore’s technical community includes many such people.
So far, the market seems to agree with Exa’s thesis. The company’s API is already being used by a range of AI companies and enterprise customers, though Exa has been relatively quiet about specific client names and usage metrics. What is known is that demand for real-time web retrieval capabilities has surged as AI agents become more sophisticated and more widely deployed. Every AI assistant that needs to answer a question about today’s news, check a company’s current stock price, or verify a recent scientific finding needs some form of web search — and Exa wants to be the default provider of that capability.
The timing of the Singapore expansion also coincides with growing interest from Asian governments and corporations in building sovereign AI capabilities. Countries like Singapore, Japan, South Korea, and India are investing heavily in domestic AI infrastructure, partly out of strategic concern about dependence on American and Chinese technology providers. A search API company with a local presence could be well-positioned to serve these markets, particularly if it can demonstrate that its technology works well with non-English content and local data sources.
Exa’s story is, in many ways, a microcosm of the broader AI infrastructure buildout happening right now. Hundreds of billions of dollars are flowing into the companies and technologies that make AI systems work — not just the headline-grabbing foundation models, but the unglamorous plumbing that connects those models to the real world. Search APIs, vector databases, data pipelines, evaluation frameworks, inference optimization tools. These are the companies that will determine whether the AI industry’s enormous capital expenditures actually translate into useful, reliable products.
And if Bryk is right, the people who build that plumbing will be the ones who never quite fit in anywhere else. The ones who got kicked out of class for asking too many questions. The ones who took things apart just to see how they worked.
The rebels.
WebProNews is an iEntry Publication