Everyone at Cypher 2025 knew what Uber was. Rather fewer walked in knowing that Uber sells enterprise AI services.
Closing that gap was the point of the exercise. Uber AI Solutions took a gold sponsorship, a keynote slot, and Booth 45 at KTPO, with a stand carrying a line that worked as a thesis statement — the best of Uber, now available to every enterprise, everywhere — and beneath it, an explanation of what that means in practice: from accuracy down to a pixel, to consistently reliable large language models.
Decoded, that is data collection, annotation, testing and localisation across multiple modalities — capability Uber built internally over a decade and is now selling outward. It is one of the least visible layers of the AI economy and one of the most necessary. As Anant Bhat, who works in global sales enablement at Uber, put it after day one: building great AI takes more than models and computing. It takes data.
Five people from the team wrote publicly about the three days. Read together, their posts amount to something more useful than a sponsor recap — a record of what an exhibitor actually learns by standing still in one place while the market walks past.
The keynote came first
Megha Yethadka, Global Head of Uber AI Solutions, delivered the session the rest of the presence was built around: Building what’s next for the world: Uber’s latest AI and data growth bet.
By the team’s own account the room was full. According to Mohammad Shaheer of Uber AI Solutions, Yethadka set out how the company thinks at infrastructure scale — across languages, regions and modalities — and argued that human-in-the-loop validation, robust QA frameworks and evaluation pipelines are becoming core to building AI responsibly rather than optional extras bolted on at the end.
Anchal Jaiswal of Uber AI Solutions noted that attendees came by the booth afterwards specifically to compliment the company on the benchmark it was setting. That is the mechanism a keynote is supposed to produce and frequently does not: a talk that sends qualified people to your stand with a reason to start the conversation.
The sequencing matters. A stand alone gets footfall. A stand plus a keynote gets footfall that already knows what you do and has decided you are worth talking to.
Three days of listening produced three problems
The most valuable thing the team carried home was not a lead list. It was a diagnosis.
Shaheer wrote that across three days — from keynote stages to booth conversations — the same themes kept surfacing. He named three.
The pain of eval drift. The hidden cost of annotation chaos. And the difficulty of collecting regionally relevant, contextually rich data to fuel global machine learning models.
Consider what that list represents. Not a survey, not a vendor-sponsored report, but the aggregated, unprompted concerns of several hundred practitioners who stopped to talk because they had a problem on their mind. Every one of the three is a maintenance problem rather than a build problem — which is itself the signal. A market worried about eval drift and annotation consistency is a market that has already deployed and is now living with what it deployed.
Shruti Tomar, who works in growth at Uber, drew the same conclusion in one line: Cypher 2025 gave a clear signal that organisations are no longer just exploring AI, they are actively looking for ways to scale it responsibly. Conversations at the booth, she wrote, reinforced a single theme — the future of AI depends on high-quality data pipelines and real-world deployment.
Exploration budgets and scaling budgets are different budgets, held by different people, released against different evidence. Knowing which one your buyer is now operating from changes everything about how you sell to them. That is a strategic input, and three days on a floor produced it.
Who actually walked up
Shaheer also did something most exhibitors do not: he published, by name, the leaders who stopped by Booth 45 to — in his phrasing — throw darts at real bottlenecks.
A Group Chief AI Officer from Narayana Health. An Associate Director from PepsiCo. The Chief AI and Technology Officer of ANAROCK. An AI/ML and cloud architect from AWS. A director of solutions engineering from Iron Mountain. A chief product officer from Sodales Solutions. Two leaders from Deutsche Telekom Global Business. The technical solutions group at Burns & McDonnell. A data science leader from Alstom. A VP of technology from Kitchen365.
Healthcare, consumer goods, real estate, cloud infrastructure, information management, industrial engineering, rail, telecoms and food technology — from one stand, over three days, and that is only the group one person remembered to name. Shaheer apologised in the post for anyone he had missed.
This is the answer to the question every prospective exhibitor asks and few get a credible response to: who is actually in the room? Two of those names — Burns & McDonnell and Alstom — were themselves exhibiting at Cypher 2025, which is a useful reminder that the floor is a marketplace among exhibitors as much as between exhibitors and attendees.
Two exhibitors, two playbooks
We wrote recently about SmartQ at the same event — an exhibitor that built a photo frame, handed out props and instant polaroids, and generated a dozen public posts including four from visitors with no connection to the company. Put the two side by side and they are almost opposite approaches. Both worked.
SmartQ ran a reach play. The booth manufactured a reason to take a photograph, and photographs become posts. The output was visibility across a dozen separate professional networks — what you want when the market does not yet know your name.
Uber AI Solutions ran a signal play. A keynote drew a qualified crowd, demos and a prize wheel converted passers-by into conversations, and the conversations produced a diagnosis of the market plus a roster of live buyers. That is what you want when you have a product ready and need to know how to position it.
The mistake is running neither play deliberately. A stand with no hook produces no conversations. A stand with a hook but nobody who can hold a technical conversation produces conversations that go nowhere. Decide which of the two you are doing before you design the space, and staff it accordingly.
What the brand did and did not buy
Uber arrived with something most exhibitors lack: a name every attendee already recognises. That is worth real money in footfall — nobody needs persuading to walk over.
It also creates a specific problem. Recognition is not relevance, and a visitor who knows you as a consumer app has to be actively re-educated about why you belong in an enterprise AI conversation. That is the job the signage was doing, and the job the keynote was doing at larger scale. As Anant Bhat described it, Yethadka’s session gave a front-row view of how the company is making its tools, talent and technology available to customers, bringing a decade of internal innovation to the broader ecosystem.
For any company entering an adjacent market, that is the transferable lesson. Whatever recognition you arrive with buys the first ten seconds. What happens after depends on whether the new proposition is legible in one line, and whether the person standing there can demonstrate it.
The 2026 floor
Cypher 2026 runs 7–9 October at KTPO Whitefield, Bengaluru — the tenth edition, with 5,000+ attendees, 150+ speakers and 100+ exhibitors expected. The speaker list and schedule are published.
If you are weighing a stand, the two cases now on record give the honest range of what one produces: reach if you build for it, market intelligence if you staff for it, and in the better version of both, a keynote that sends the right people to your booth already knowing why they are walking over. Book a booth, review the sponsorship options, or get in touch.
And if you are attending rather than exhibiting, note what the exhibitors are getting out of this. The people on those stands spend three days hearing what the entire market is struggling with. Ask them what they have learned — passes are here.
