Google Cloud has made a competitor’s web search a native option inside Gemini. The decision reveals more about the money moving through the agentic web than any funding round could.
On 16 July 2026, Parallel, the search startup founded by former Twitter chief Parag Agrawal, said its web search is now built natively into Google Cloud’s Gemini Enterprise Agent Platform. Developers can call it through the Gemini Enterprise API, select it as a grounding source when building agents, and pay for it on the same Google Cloud invoice they already receive. The significance is not the deal itself. It is that Google, the largest search company in history, is giving shelf space to a search rival inside its own enterprise stack. That tells you where value is shifting as software agents, rather than people, become the main users of the web. The people to watch are Agrawal, Google Cloud president Matt Renner, and the enterprise buyers now choosing whose search to trust. The open question is who keeps the tolls once the traffic is real.
Read quickly, the Parallel Google Cloud partnership is a routine notice. Read properly, it is stranger than that. Google has spent two years building its own search grounding into Gemini. Companies do not usually invite a competitor to set up a stall in their own shop. When they do, it is worth asking why.
The customer that does not click
The answer starts with a claim Agrawal has made since Parallel came out of stealth. His bet is that agents will soon use the web far more than people do, and that search then has to be rebuilt, because the thing doing the searching no longer behaves like a person.
Human search is organised around a page of blue links and the adverts beside them. A person scans, clicks, reads, and Google earns on the way through. An agent wants none of that. It wants clean, structured, machine-readable answers it can act on, at speed and at volume, with citations it can trust. Parallel built a separate web index tuned for that job rather than adapting a product designed for human eyeballs.
This is where Google’s choice makes sense. Its advertising business depends on humans clicking. Its cloud business depends on companies building on Google’s rails and spending more each quarter. Those two businesses do not always want the same thing. Google Cloud is in a hard fight with Amazon and Microsoft to become the default place enterprises build AI agents, and in that fight, openness is a weapon. A buyer picking a platform wants the best tools inside it, not a walled garden. Offering a respected outside search option, and metering it through the Google bill, signals neutral ground. Google keeps the customer, the consumption and the invoice. Parallel supplies one part. On the maths of a cloud business, that is a sound trade.
What Agrawal actually gains from the Google Cloud partnership
For Parallel, the prize is not validation. It is distribution, and the removal of a problem that kills most enterprise startups long before quality does.
Selling infrastructure to large companies is slow because procurement is slow. A new vendor means security review, a fresh contract, a new line item, a finance team asking who signed off. By landing on the Google Cloud Marketplace with usage charged to an account the customer already holds, Parallel skips most of that. The buyer does not onboard a supplier. They tick a box. For a company of roughly 80 people, borrowing Google’s enterprise reach and billing machinery is worth more than any benchmark score.
The benchmark scores, to be fair, are respectable rather than commanding. An independent test by AIMultiple this year placed Parallel’s premium tier in the top group of search APIs, statistically level with Brave, Firecrawl and Exa, though its cheaper tier fell below that band and its speed lagged faster rivals. Parallel is a serious contender. On the public evidence it is not in a class of its own. So the Google deal is not a reward for being the obvious best. It is a strategic placement, and that cuts both ways.
A market that keeps rebuilding itself
This category has been rewired twice in under a year. Microsoft retired its Bing search API and, in doing so, handed the field its opening. Tavily, one of the early names, was bought by Nebius in February. At the same time, OpenAI and Anthropic folded web search straight into their own tool-use loops, so a developer building on those models often needs no outside search vendor at all.
That last shift is the quiet threat to every independent, Parallel included. Native search is convenient, but it ties you to one model provider. Parallel’s real position is the counter-argument. Serious enterprises do not want to bet everything on a single model. They want to stay portable, swapping models as prices and capability move, while keeping one dependable search layer underneath. Sitting inside Gemini as an option, rather than being absorbed by it, is how you sell that independence. It also leaves Parallel exposed to the platform it now leans on. Google can change terms, promote its own grounding, or lose interest. Distribution through a giant is never free.
There is a second front to the bet, and it is the more original one. In May, Parallel launched Index, a system to pay publishers and data providers when agents use their work, with The Atlantic, Fortune, PitchBook and others signed on. It is an attempt to build the business model of the agentic web, not just the pipes. The logic is sharp. If agents strip-mine the web without paying, the web those agents depend on stops being worth maintaining. Solve the supply side and you become harder to remove from the stack. Whether publishers earn enough through it to matter is unproven, and it will take real revenue, not launch partners, to settle.
The founder underneath
It is easy to forget how Agrawal reached this point. He spent eleven years at Twitter, rising from engineer to chief technology officer to chief executive, and was pushed out within months when Elon Musk completed his takeover in 2022. He is an IIT Bombay and Stanford product, part of a familiar line of Indian-origin technologists building core infrastructure in Silicon Valley rather than consumer brands. What he did next is the tell. He did not build another social network or chase another public role. He went to the least glamorous layer of the stack and started laying pipe for a shift that had not fully arrived.
That is the calmer, more interesting story inside the announcement. The valuation, around two billion dollars on roughly 230 million raised across three rounds, is a price on a thesis, not on today’s revenue. Parallel is being backed for a future in which machines are the main users of the internet and someone has to build the roads they travel. Early adopters such as the legal AI firm Harvey and a production deal with Genpact show the thesis has paying customers, not just slides.
What happens next
The Parallel Google Cloud partnership is the clearest sign yet that the largest players expect some version of the agent-first web to arrive. They are willing to let a rival stand inside their own house to be ready for it. The question Agrawal has not yet answered is whether the company that builds the roads gets to keep the tolls, or whether the giants hosting him today decide, once the traffic is real, that they would rather run the roads themselves.
FAQ
What is Parallel and who founded it?
Parallel, formally Parallel Web Systems, is a PaloAlto company founded in 2023 by Parag Agrawal, the former chief executive and chief technology officer of Twitter. It builds web search infrastructure designed for AI agents rather than human users, delivering structured, citation-backed results from a proprietary index tuned for machine retrieval.
What does the Google Cloud deal actually give Parallel?
It makes Parallel a native option across Google Cloud’s Gemini Enterprise Agent Platform. Developers can call it through the Gemini Enterprise API, choose it as a grounding source in Studio, and buy it on the Google Cloud Marketplace, with usage billed to their existing Google Cloud account. The real gain is distribution and a shortcut through enterprise procurement.
Why would Google host a search competitor?
Google’s advertising business and its cloud business have different incentives. Google Cloud is competing with Amazon and Microsoft to be the default platform for building AI agents, and offering the best third-party tools makes that platform more attractive. Google keeps the customer relationship, the cloud consumption and the billing, while Parallel supplies one component.
How does Parallel compare with Perplexity, Exa and Tavily?
An independent 2026 benchmark by AIMultiple put Parallel’s premium tier in the top group alongside Brave, Firecrawl and Exa, with the leaders statistically close together. Its cheaper tier scored lower and its latency trailed faster rivals. Parallel is competitive at the top end rather than a clear leader, which is why the Google placement matters as a distribution advantage.
How much has Parallel raised and what is it worth?
Parallel has raised roughly 230 million dollars across three rounds from investors including Sequoia Capital, Khosla Ventures, Index Ventures, Kleiner Perkins and First Round Capital. Its most recent round, a 100 million dollar Series B in April 2026 led by Sequoia, valued the company at about two billion dollars. It employed roughly 80 people as of May 2026.
