Oct 28, 2026  @Tanveer Padode

A first-hand account of Bentley’s 2026 awards, where AI, digital twins and engineering innovation are reshaping infrastructure delivery.

- By Tanveer Padode, CEO, IMPACCT: A first-hand account of Bentley’s 2026 awards, where AI, digital twins and engineering innovation are reshaping infrastructure delivery.

Two days in Singapore

Construction is won or lost in two places: how a project is designed and how it is delivered. Technology has promised to improve both for years, and the question I carried into Bentley's Year in Infrastructure, held at the Fairmont Singapore on 6 and 7 October, was what the world has actually built with it. The finalists gathered there had the answer, project by project: what happens when software, data and AI are used on real tunnels, bridges, grids and cities, and what it takes to get there.

This was the 22nd edition of the awards, often called the Oscars of Infrastructure, and the format was refreshed. Futurist Chris Luebkeman hosted as the first external emcee. Strategist Parag Khanna reminded the room that the world's population is expected to peak at 9.5 billion by 2045, mostly in cities, and that "we continue to be infrastructure maximalists." In Bentley Labs, all 36 finalist projects sat as interactive 3D models on a large LED wall, with a headset for anyone who wanted to step inside. A tunnel under Porto's old city is a very different thing as a place you can stand in than as a slide.

More than 300 projects from 53 countries were nominated this year. I spent two days among the people behind them, and among the analysts who watch the sector for a living. What follows is what stayed with me: the keynote, five conversations, and how Indian teams fared.

The keynote: 22 years from the Gherkin to AI agents

Bentley CEO Greg Bentley opened day two with some history. The awards began in 2004 as the BE Awards of Excellence, when computational design was the new thing. One early winner was London's 30 St Mary Axe, the Gherkin, whose team wrote that models which took days to draft by hand could now be generated in minutes. Algorithmic exploration, generative design, days compressed into minutes. Twenty-two years later the pitch hasn't changed. The scale has.

Nearly 70 per cent of this year's finalists, and more than a quarter of all submissions, used AI somewhere in the project. Bentley was careful to say this will not turn into an "AI in Infrastructure" award, and that it will keep celebrating good use of software with or without AI. But from planning to operations, the pattern was hard to miss.

Two projects made the case better than any slide. On the Porto Metro expansion, AYESA used AI agents to check PLAXIS simulation results against design criteria. Tunnelling under a historic city usually means evaluating a handful of scenarios and stopping at the first one that meets the safety standard. AYESA ran more than 100, cut design costs by 30 per cent and saved 4,500 cubic metres of concrete. It went on to win the Bridges and Tunnels category.

POWERCHINA's team on a 400 MW offshore wind farm in Fujian built its own optimisation platform on SACS for 28 jacket foundations that have to survive 18-metre waves and typhoons. Optimisation cycles ran ten times faster and steel costs fell 3 to 5 per cent. For 20 foundations, that is about 1,900 tonnes of high-grade steel. It was a Founders' Honoree in Design Optimisation.

Both teams had to write code to get there, and most firms can't. Bentley's answer is to roll out Model Context Protocol (MCP) servers across its products so engineers can instruct software in plain language. In a lab demo, frontier AI models used MicroStation to rebuild the Gherkin's shell from the original mathematical formula, then model finer details from reference photos, then generate the structure in STAAD. The line I wrote down: AI supplies the high-level reasoning, and the engineering software supplies deterministic rigour.

Two guardrails, and why I agree with both

The part of the keynote I valued most was the warning. When anyone can ask software for a design in plain language, the risk is what Bentley called "AI sloppiness", and in infrastructure the margin for error is zero.

The first guardrail is human judgment. Optimisation belongs to the machine, and navigating trade-offs belongs to the engineer. How do you balance building fast against fifty-year resilience? An algorithm can offer options, but, as the keynote put it, an algorithm does not sign and stamp the drawings. I came away more convinced than before that AI makes good engineers more valuable, since the scarce skill is now knowing which of a hundred good answers is right. I have a small version of this lesson myself. I only gave my own AI assistant access to my email after running it for about six months on a separate machine with no personal accounts. Trust was earned in steps. A bridge deserves at least that much.

The second guardrail is deep context, starting with clear intent. AI can't guess what you haven't defined, and vague requirements just produce hundreds of optimised answers to the wrong problem.

That led to the idea I expect to quote for months: the Durable Corpus. Bentley argues that to lead in infrastructure AI, an organisation has to capture three digital threads. The Design Journey records what was decided, how and why. The As-Built Reality records what was actually constructed. The Operational Truth feeds decades of field performance back into design. WSP, a Founders' Honoree, showed what it looks like in practice: an iTwin AI platform where any team member can ask in plain English for, say, all drainage structures between 5 kilometres and 15 kilometres and get structured data back in seconds, cutting data extraction time by 75 per cent.

Bentley will take its own product announcements to the Bentley Tech Summit in Toronto next spring. That leaves Year in Infrastructure to do what it does best, which is to put projects and the people behind them at the centre.

Five conversations between sessions

The stage sets the themes. The real texture came from the corridors, where I sat down with five people from market research, engineering software and the water sector. They disagreed on plenty, but the same few ideas kept coming back.

Jeff Hojlo, IDC: AI is a tool belt, and it needs one place to live. Hojlo's research on agentic AI in construction is blunt about where the industry stands. Construction is, in his words, "quite immature when it comes to AI use" next to healthcare or consumer goods, and slightly ahead of energy. The first real use cases are unglamorous. Task automation, mining quality and safety records for lessons, and giving partners one unified view of project data. Next come pre-construction, where an agent can weigh a subcontractor's past performance on time, quality and regional experience, and handover, where the old paper binder gives way to a digital package that carries its own agents. He thinks of AI as a collection of tools, machine learning to connect workflows, generative AI to gather data, agents to decide and natural language as the interface. One contractor told him, "I love to have this tool belt of technologies." Whatever the tools, every project needs a common data environment, and geospatial data is increasingly where projects begin. A US contractor put it to him this way: "We start our infrastructure construction jobs with a foundation of geospatial information."

Rita Liu, ARC Advisory: physical AI arrives slowly where mistakes are expensive. Liu comes from the elevator and building-technology world and watches China closely. Humanoid robots already work in shops and pharmacies there, but on heavy plants and construction sites she expects the next two to three years to be about humans and physical AI working side by side, "because any mistake will cause or lead to more severe damage." I pushed the other way, with the example of laying railway track in punishing heat, work people increasingly don't want to do. I also described what I saw last year at Bauma in Munich, where an autonomous excavator cleared an entire site on instructions from an AI. She noted that remotely operated diggers have existed for years, which is fair, and that the jobs most exposed are standardised, rule-based ones such as programming, accounting and analysis. Strategy and creativity stay human. Her wider point on Chinese contractors was practical: with the domestic market maturing, they are heading to Africa, the Middle East and Southeast Asia, where local standards, labour law and environmental rules will test them.

Kaushik Chakraborty, Bentley Systems: it's a choice, not a cost. Chakraborty spoke about India with refreshing directness. Technology is "still under 1 per cent of our infrastructure projects in India, as opposed to say a global average of 5 per cent or 4 per cent," because it is seen as a cost and not a strategic enabler. Asked what the 5D BIM work on the Nagpur and Pune metros changed, he said there was nothing special about it: "It just takes willingness." Working in silos, he argued, is a choice. Quality tends to arrive with private and institutional capital and long concessions, and India's progress so far has been "individual excellence rather than institutional excellence." I raised an idea that I'll repeat here: insurers could price premiums by how much technology and certainty a project can offer on risk, which would turn technology into a disguised cost reduction. He also made a point about openness worth noting. Bentley's platform federates data from Esri, Autodesk and Trimble and uses open file formats, because on a real project nobody works in one vendor's software alone.

Annemarie Briggs, Verdantix: the engineer as supervisor. Briggs expects AI to move engineers from drawing to defining. They set the problem, the rules and the constraints, then check what the AI proposes. Her blunt version: you become "almost like the supervisor" running the agents. Asked where an EPC should start among the six simulation types in Verdantix's June report, she picked process simulation, the most mature tool with the least risk, and the source of data for mechanical, piping and electrical work downstream. On handovers, she said the cost shows up as rework. The answer is a data agreement on day one, interoperable platforms, and closing the design loop so operating data flows back into design. I told her about our own bid-analysis and project-tracking work, and she said Verdantix wants more depth in Asia and the Middle East.

Chloe Meyer, Bluefield Research: follow the capital, then the water. Meyer walked me through the UK's AMP8 cycle, roughly £100 billion of water spending to 2030. Sewer overflows, leakage and large supply schemes are shaping it, and contractors are being asked for modular, pre-assembled construction, BIM and the ability to work on live sites. Beyond Europe, she sees growth in the Middle East, with a trend toward more, smaller desalination plants joined by interconnected pipelines, "instead of having those huge 10 plants, maybe have 25 or 40 smaller plants." In India she points to the government's water missions shifting toward wastewater, stricter discharge rules and flood protection. We spent a good while on reuse. Her view is that reuse projects succeed when they match users to volumes and find a business model that gives them a reason to switch. I argued that industrial users should help finance reuse in return for sustainability credits, and that India's pumped-storage projects already sell industrial water that way. Her fair point was that the fine print differs when assets are publicly owned. For anyone tracking money before it is announced, her advice was to follow four streams: institutional funding programmes, utility and corporate filings, master plans, and public-private partnership databases.

What the five had in common

AI is useful already and still early. The human decision-maker isn't going anywhere. Data in one trusted place beats data scattered across tools. And the biggest obstacles are rarely technical. They are contracts, incentives and willingness. Kaushik's line could have come from any of the five.

India at the awards

An independent panel picked 36 finalists across 12 categories, and Bentley's founders and executives separately named 16 Founders' Honors projects. To see how India did, I used one filter: where the project actually sits, not where the firm is registered. On that test, India had three of the 36 finalists and two of the 16 Founders' Honors, and one of those three took a category trophy.

Pinnacle Future Build won Cities and Facilities with the Pinnacle Infotech digital twin of its Madurai campus, a live model of buildings and facilities built for smart operations. It beat two strong finalists, SUDOP PRAHA's digital twin of Prague Main Railway Station and Baosteel's industrial rust-belt rebirth in China. It is a win I like for a particular reason. The campus is in Madurai, not a metro, and it is run through its twin, not just designed with one.

NMA (Nikhil Mahashur and Associates) was a finalist in Rail and Transit for the Hajimalang Gad funicular rail, which I take to be the Haji Malang funicular near Kalyan in Maharashtra. It replaces a climb of about 2,600 steps, nearly two hours, with a ride of about 10 minutes over 1.2 km. It was proposed in 2004, construction began in 2012 and it opened in January 2026. The engineering deserves the recognition. The timeline is the part I would put on a slide. The category went to SUDOP PRAHA's Central Bohemia high-speed rail.

Haskoning India Consulting was a finalist in Water and Wastewater for the Delhi stormwater drainage masterplan, aimed at a flood-free national capital. Anyone who has watched Delhi, or Mumbai, struggle through a monsoon will understand why this matters. Evides N.V.'s operational water-distribution digital twins in the Netherlands won.

The Founders' Honors added two more Indian projects:

• WSP, in Connected Data, for the iTwin AI Project Lifecycle Intelligence Platform, delivered in Bangalore. This is the platform from the keynote. Beyond the 75 per cent cut in data extraction time, it produced an embodied carbon report of 16.6 tCO2e and an estimate of US$ 8.7 million in under 15 seconds, flagged structural fill as 92.2 per cent of the emissions, and caught a 26.3 mm settlement event from live sensor alerts before it escalated. It has also been deployed on the Tel-Aviv M2 Metro and Mountain View Corridor projects.
• Almondz Global Infra Consultant, in Digital Twin at Scale, for NAKSHA, the national programme to survey urban land. Almondz built city-scale 3D digital twins of Nadiad, Navsari and Anand in Gujarat, about 318 sq km, from more than 100 TB of drone, LiDAR and imagery data. 3D mesh reconstruction time fell 40 per cent, rework fell by more than 35 per cent, and the twins reach better than 10 cm precision. The early estimate is a 10 to 15 per cent rise in municipal revenue through better property tax assessment.

Looking across the five Indian projects, I notice that most are about data: twins, reality capture, land records and platforms. India's recognition came less from heavy construction than from knowing, measuring and managing what we have. That is where the gap between our best and our average is also widest.

The 2026 category winners
The 12 winners came from 11 countries, with Australia winning twice. Something else stood out as I went through the list: none of the 12 winners was led by a US or UK organisation, though both had several finalists. The winning teams came from Portugal, India, Ireland, China, the Philippines, Poland, the Czech Republic, Australia, Indonesia, Brazil and the Netherlands. That is worth remembering the next time we assume the frontier is somewhere else.

The Founders' Honors added a second layer. They included the Parliamentary Complex of Nusantara in Indonesia, where AI optimised heavy-equipment routes using regularly captured site data, and Ampirical's work powering the Western Hemisphere's largest data centre.

What I'm taking home
I've been to plenty of conferences where the technology is impressive and the room is quiet. This one felt different, and part of it was how Bentley treated its finalists: a dedicated lounge, engraved name badges, gifts made by local artisans. It sounds small, but it tells you who the event is for. Finalists, vendors, analysts and engineers shared the same rooms and corridors for two days, and my five conversations came out of exactly that mix. I left with a long list of things to try, which is the best outcome a conference can offer.

For India, I come away with four thoughts.

Twenty-two years ago these awards celebrated pioneers who used computational design to deliver landmark buildings. This year's finalists, from a campus in Madurai to a pumped-storage plant in Chongqing, suggest the pioneers are now everywhere and the tools are within reach of ordinary projects. What happens next depends less on the software than on our willingness to use it well. At IMPACCT, we have tracked more than 150,000 Indian projects and 35 lakh tenders, and what that data shows me is how much project intelligence never makes it past the tender. I left Singapore a little more hopeful that this is starting to change.

Sources