Just as the industry gathered in Singapore for Token2049, its biggest event globally, AI frontier lab OpenAI published solutions to over 700 foundational math problems, spending just 3 hours of compute for each. Mathematicians were left in shock, and cryptographers raised their eyebrows.

Cryptography wasn't mentioned at all in the OpenAI release, an omission that Dragonfly VC Haseeb Qureshi called “weird.” One explanation could be export restrictions for this sensitive cryptographic research, which prohibit the labs from publishing certain findings. Or, as Qureshi speculated on stage, OpenAI could have found something alarming, such as the ability to break elliptic curve cryptography, and is right now working behind closed doors with the government on how to address it. 

The mechanism that protects crypto 

ECDSA, the elliptic curve digital signature algorithm in question, is a form of public key cryptography that generates and verifies signatures for secure transactions. It is used well beyond cryptocurrencies. Financial institutions rely on it, alongside other algorithms like RSA, to verify and secure mobile banking traffic. 

Across the internet, Transport Layer Security (TLS) and Secure Shell (SSH) rely on ECDSA to authenticate connections. As frontier AI keeps advancing, standards organizations and industry bodies are already racing to migrate these protocols to post-quantum algorithms. 

In the decentralized crypto space, developers are already on high alert, particularly after Google published an explosive paper on the progress of quantum science in March. Engineering teams are now exploring solutions that upgrade signature schemes, enabling crypto holders to move their funds to post-quantum wallets, while drafting contingency plans for when “Q-Day” inevitably arrives.

The hunt for Satoshi’s stash

Bitcoin, the most established and distributed network, is notoriously slow to execute protocol upgrades, raising concerns among investors whether it can adapt on time. There is also the question of what happens to the about 1.1 million of Satoshi’s coins that haven't moved since the bitcoin founder disappeared at the end of 2010.

Early coins are stored in so-called Pay-to-Public-Key addresses that expose the public keys directly on the blockchain. A quantum computer running Shor’s algorithm can reverse-engineer the private keys from the public keys, drain the funds, and probably crash the bitcoin price overnight. 

As few realistically expect Satoshi to return or for those funds to be moved to safety, the community has begun debating how to avoid a potential catastrophe. One option would be freezing the addresses which would require a protocol upgrade and soft fork that needs consensus of the node operators. In the worst case scenario, it could trigger a chain split. However, previous successful upgrades like Segwit and Taproot demonstrate that these changes can work out just fine. 

Freezing funds could also alienate bitcoin purists who view such an intervention as an attack on property rights and would likely oppose the soft fork. Satoshi’s funds would effectively become a massive bounty for quantum scientists racing to crack the algorithm first and cash the prize. It left some people wondering if the US government, leveraging its vast compute power, should try to get the coins first to prevent them from falling in the hands of a rogue state actor.

Shutting down the crypto doomers

Justin Drake, the Ethereum Foundation researcher who sparked the current debate, argues that ECDSA could fall well before Q-Day, “in the worst case in months not years.” He recommended entering “bunker mode,” urging holders to move funds to addresses that have never signed a transaction, thereby keeping their public keys hidden behind a cryptographic hash. Hash functions like SHA-256, a fundamental pillar of cryptocurrencies, remain virtually unbreakable, by AI and quantum computers alike. 

Qureshi dismissed this as cryptographic doomerism, comparing the rush to bunker mode to buying canned food in preparation for a nuclear winter. Still he thinks that the broader ecosystem, including developers, wallet providers, exchanges and end users, should mobilize and devise a strategy that it will adopt in a worst-case scenario. Users could for example generate a hash-based ZK-proof (for zero-knowledge proof) of their seed phrase to transition their assets to a new address.

A cryptographic hash acts as a one-way mathematical function that compresses data into a unique fixed-size output. A ZK-proof builds on this, proving that a statement is true without revealing the underlying data. Modern ZK systems use hashes as their primary building blocks.

Eli Ben-Sasson of StarkWare, the team behind the Ethereum layer 2 Starknet, has been a pioneer advocating for ZK-proofs in the blockchain world. As co-inventor of STARKs, he helped create the mathematical breakthroughs that made proof verification faster and therefore more practical. In the light of the current events, Ben-Sasson advocated in Singapore for greater “crypto agility,” urging the industry to move faster in replacing vulnerable algorithms, particularly by using hash-based STARK signatures. 

The recent progress of AI-accelerated math is another reason, along with quantum, why ECDSA might fall faster than expected, Ethereum founder Vitalik Buterin weighed in. He believes that even lattices, widely viewed as a leading quantum-resistant approach, could take serious hits from AI math in the coming years.

“For signatures and proofs, we already know how to go hash-only,” he wrote. For Buterin, the bigger challenge is the wider public-key encryption that goes far beyond blockchains. “Secure communication, anonymizing protocols, lots of things need public-key encryption.”

The event is further expected to showcase how the sector can support emerging AI use cases, and what came of the “pivot to AI” that experts suggested so enthusiastically. Agentic finance has become the latest craze. Programmable money will increasingly adopt and rely on blockchains and stablecoins. 

Payment infrastructure provider Stripe, which just wrapped up its own Asia tour in Tokyo, is already heavily relying on both. The company, now valued at $159 billion, catapulted itself to the top of the payment technology sector, with a flurry of acquisitions and new digital features. 

The emergence of robotic finance

In Singapore, the firm touted how users can build new businesses effortlessly from a simple prompt. Stripe created a machine payment protocol as an open standard, providing specifications for agents and services to execute payments programmatically. It enables microtransactions and recurring payments via stablecoins, fiat money, or cards. 

Currently, a purchase can be quite cumbersome, requiring an agent to create an account, navigate a pricing page, choose between subscription tiers, enter payment details, and set up billing, which still involves human intervention, the firm says. 

Businesses can connect their product listing to Stripe and subsequently select which AI agents they'd like to sell through. Stripe says it helps with discovery, checkout, payments, and fraud detection. It creates cards and fast global payouts to over 100 countries through stablecoins. 

The company claims that payments jumped for its users after they made the integration. Clients can store the currency they get paid in or convert it via API and hold it in treasury. In order to further bolster the service, Stripe in August spent $7.5 billion to acquire OpenRouter, an API marketplace with access to hundreds of AI models.

Why crypto rails are better 

Zach Abrams, the co-founder of stablecoin infrastructure platform Bridge, which Stripe acquired for $1.1 billion early last year, believes that digital tokens are the future of commerce. Speaking at Stripe's Singapore event, Abrams thinks that microtransactions will be the true catalyst for internet money, as they are immediate, integrated, borderless, and programmable.

In March this year, Stripe launched Tempo, a layer 1 blockchain tailored to its stablecoin needs, which it developed together with the established crypto firm Paradigm. Before that, in June of last year, it bought Privy, a digital wallet infrastructure that onboards users to the crypto world.

Abrams explains how these tools have become a fundamental part of its global payment infrastructure. Félix Pago, a Stripe client, quickly managed to grab more than 5% of the massive $60 billion cross-border payment corridor between the US and Mexico. 

Stablecoins reduce fees by more than 60%, he says. And payments are nearly instant, compared to a transaction via Swift, that can take over 20 hours. Stripe revealed that stablecoin volume on Tempo has crossed $1 billion in monthly transfer volume. 

A new stablecoin standard 

The company is not done yet. A week ago, Abrams announced that he will be leaving his role at Stripe to run the newly launched Open USD, an open stablecoin standard that counts firms like Coinbase, Visa, Mastercard, BlackRock, Shopify and DoorDash among its backers.

The new coin pledges transparency, especially when it comes to cost and governance. It makes supply and reserves publicly available on a website. 

Abrams believes that an open standard will drive further adoption. “At Bridge, we made stablecoins more accessible. But our impact was constrained by issuers' economics and incentives,” he wrote on X. “We hope to remove these constraints.”

Stripe has made OUSD the default on its platform, although it will still accept other coins as well, if clients prefer. The move challenges Circle, which has long led the charge in institutional stablecoin adoption. 

Towards a duopoly? 

The USDC issuer responded by launching its own chain last month, an open layer 1 blockchain named Arc, designed for stablecoin settlement and AI applications. Circle said that Arc kicked off with more than 100 institutional and ecosystem builders and more than 100 applications. Among the founding validators are BlackRock, ICE, Mastercard, Visa, MoneyGram, and banks like SBI Group, Standard Chartered and Sumitomo.

Like Stripe, Circle is aggressively building settlement systems for microtransactions and AI agents. The company also recently bought Singapore-based Tazapay for $400 million, becoming a merchant processor able to cover the last mile, the fiat-to-stablecoin conversion. 

While Stripe has concluded its Asia tour, Circle will be at Token2049 with a massive presence and a dedicated Circle house set to entice more institutions. And it's probably already plotting its next move in the battle of the stablecoins. 

Bali is expected to lure international capital and finance professionals, such as bankers, insurance brokers, and capital markets specialists. It could be servicing assets from treasuries to bullion and derivatives, catering to Islamic finance and family offices, as well as fintech firms.

In order to appeal to global investors, the Indonesian government is preparing a range of incentives, including golden visas, capital repatriation guarantees and the promise for tax relief. Estimates see possible inflows of capital between IDR 300 trillion (USD 17 billion) and IDR 500 trillion. 

The idea is not far-fetched. Bali stands out due to its international reputation, connectivity, and first-rate tourism industry. Its MICE sector is thriving, with high profile international events, such as the G20 meeting in 2022. The so-called Bali Fintech Agenda emerged in 2018 from the annual meeting of the International Monetary Fund and World Bank. 

The new Dubai? 

Coordinating Minister for Economic Affairs Airlangga Hartarto said at the end of August that the proposed site for Bali was still under evaluation. The targeted location would be Bali's Special Economic Zone, a spot named Kura Kura, situated on 500 hectares of largely reclaimed land on Serangan island, only minutes away from the Ngurah Rai International Airport. 

When this author visited it last week, the area—with vast stretches of mangrove forest—looked more like a conservation zone than a place that could house financial institutions. A shopping mall filled with plenty of luxury brands stood mostly empty. 

The COO of Danantara Dony Oskaria nevertheless compared the project already to the Dubai International Financial Center, which developed a hub with 50,000 professionals, offering corporate tax exemptions for up to 40 years. 

Indonesian media criticized the initiative. The legal immunity and zero-tax status for family offices in Bali would create a potentially systemic money-laundering loop. “Illicit funds could be laundered into state-sanctioned financial assets without prosecution risk, then funneled into family offices that offer asset protection through trusts and special purpose vehicles with no requirement to disclose ultimate beneficial owners,” wrote the Jakarta Post. 

Measures of this nature undermine the credibility of Indonesia's financial system and threaten its membership status in the Financial Action Task Force (FATF), said Mahfud MD, a former minister and legal academic. 

Financial worker’s paradise

For finance professionals, a career move to the island could be alluring. Bali so far has attracted primarily the nomad variety of fintech professionals. The number of hopefuls equipped with laptops and phones, running their small startups or acting as freelancers to businesses overseas, gained in popularity in the aftermath of the pandemic. 

The laid-back western lifestyle sometimes clashes with local Balinese culture and indigenous customs. When a group of Dutchmen started a running club that didn't admit an Indonesian woman, the ensuing online outrage alerted authorities. Director General of Immigration Hendarsam Marantoko said that this form of “creating a state within a state” was not tolerated and that the Dutchmen were not allowed to come back.

Officials have been becoming more diligent screening for overstaying tourists and reminded nomads entering on tourist visas that posting content online for revenue, payment, sponsorship or other commercial purposes is in breach of visa conditions, even if content is published after they have left the island. “It’s illegal to work, conduct research or volunteer if you’re on a tourist visa,” a website for expats warned.

Indonesian authorities have established a special task force to conduct patrols in areas with high concentrations of foreign nationals across Bali as part of efforts to tighten oversight into their activities. Between January and June 2026, the immigration directorate deported over 300 foreign nationals from 60 countries, for violations that included overstaying and misuse of residence permits.

According to Indonesia's Ministry of Immigration and Corrections, foreign nationals who plan to create content in Indonesia must at least obtain the C5A content creator visit visa that exists in a revised version since May 2025. There are only sparse official figures of approved visas for long-term stay, such as the E33G remote worker visa that requires a minimum salary of USD 60,000 per year. Estimates suggest that there are at least 3,000 of these registered long-term digital nomads in the country. 

Today, no major fintech companies call the island their home. Indonesia's digital finance firms and specialists remain nearly exclusively based in Jakarta. Bali’s new financial center will likely attract global VCs first, says Yoko Simon, a partner at Pillar.ai, which helps financial firms automate their sales operations. It could certainly incentivize overseas talent to return to Indonesia, he believes. 

Kimi, MiniMax and Alibaba’s Qwen have since grabbed global market share, shoving aside the more expensive frontier models from Anthropic and OpenAI. This went to a point that the US administration considered banning them, heeding the call of its home-grown AI powerhouses that have spent billions of dollars training their models. Now they will have to adapt or they will likely get taken out by Chinese competitors.

Local AI firms adopt open models

Sector experts and entrepreneurs at an annual event of the Singapore Chinese Chamber of Commerce last week seemed to have it figured out: front-end models like Claude are way too expensive. With open-source versions many times cheaper, it is now about value-maxxing rather than tokenmaxxing, they said. 

While cost seems to be top of mind for users, data security plays an almost equally important role. “Asians and most Americans don't want to trust Dario and Sam anymore,” said Charlie Hu, co-founder of enterprise AI firm OpenMax at one of the panels.

“Data sovereignty has become the first question in every sales discussion,” Christian Schneider of fileAI weighed in. Companies continue to trust and rely on their existing cloud providers, he added. This could play in favor of open weights that can run independently on local servers. 

Companies also don't want governments to regulate by banning models—they want choice. When it comes to regulation, conference participants lauded the agentic AI framework proposed by Singapore’s IMDA, which was first presented at the World Economic Forum in Davos earlier this year.

Safe AI: the next challenge 

The framework aims at creating a trusted environment, for “confident and safe” use, while at the same time “allowing cutting-edge innovation.” It provides organizations with guidance and use cases so they can deploy AI responsibly, by assessing risks upfront and placing limits on their agents.

In recent tests, several AI firms in both the US and China reported that deployed agents literally went rogue, escaping their sandbox and hacking their way into code repositories. Tech evangelists already celebrate the advent of AGI, artificial general intelligence, whereas the rest of the world watches in shock.

Singapore's framework requires every agent to have a human in the loop, ensuring accountability. It also expects the creators to define clear processes and set boundaries at which human approval is required for the agent to proceed. 

Companies have some work to do. A recent study by compliance firm Sumsub and the Singapore Fintech Association showed that fewer than a third of Singapore businesses—and only slightly more across Asia Pacific—can produce an audit trail to prove AI-driven decisions to regulators. According to the report, over 90 percent of Singapore businesses are today using or testing agentic AI.

Singapore wants to play too

Singapore's national AI strategy, named NAIS 2.0, aims to scale AI capabilities, dedicating more than S$1 billion ($780 million) to improve its current infrastructure, building research centers of excellence, and supporting businesses and local talent. 

Hyperscalers like AWS and Google Cloud have invested several billions of dollars in its Singaporean data centers to ready them for heavier AI workloads. OpenAI in May announced that it will set up its first AI lab outside the US in the city-state and hire 200 local staff.

Several Chinese AI firms meanwhile shifted their headquarters to Singapore, capitalizing on local expertise and funding. While this is meant to facilitate global access, every move is heavily scrutinized. They also can't fully escape the long arm of Beijing, as Meta’s failed $2 billion acquisition of the relocated Manus AI demonstrated in April. 

Beyond that, there's been recent controversy as Chinese open-weight models reportedly rented data centers in South East Asia to train their models on the latest Nvidia chips, bypassing the US chip ban. American competitors further accused the Chinese of destilling or harvesting capabilities of Anthropic’s and OpenAI’s front-end models.

The chip embargo doesn't seem to be hurting Kimi and Co. so far. Meanwhile, China’s homegrown technology is improving rapidly. Huawei, arguably China's top chip designer, and SMIC are increasing the output of their domestically produced AI chips. Chinese lithography tools are also delivering better results. While the US still leads China in data center capacity, the Middle Kingdom holds a clear advantage in installed electricity-generating capacity, perhaps the key battlefield in the race for global AI dominance.

Thousands turned up last weekend to the Moofest, Singapore’s largest retail investor event. They came to chase freebies and give-aways, but also to learn about the latest craze: options and futures that involve leverage. 

Chinese brokers such as Moomoo, organizer of the event, and Tiger Brokers have made inroads in this segment and gained market share, especially among the speculation-minded Asian populations. In the West, firms like Robinhood introduced frictionless stock trading on the phone and keep expanding at lightning speed with new features.

Korean retail investors lifted the country’s tech stock-heavy Kospi index, which earlier this year saw a massive speculative boom, mainly propelled by leveraged bets on firms that benefit from an elevated AI chip demand. They pushed the benchmark index to double before it plunged into a sharp bear market correction. 

The AI boom equally captured investment brokers. As people increasingly turn to their preferred AI models before consulting a doctor, lawyer, or financial adviser, trading apps have upgraded their tool boxes to cater to this new type of demand. They allow investors to upload a breakdown of their assets, preferences and life objectives to get an evaluation and financial plan in return. 

Robinhood's yearly product event earlier this month proposed new AI-driven "agentic trading" features enabling customers to plug their personal agents into their system. Depending on the type of query or prompt, users will be routed to the cheaper  conventional FAQ catalog, or if necessary, to an expensive and compute-heavy frontier model.

Blockchains ready for prime time

Robinhood also introduced its own blockchain-based digital exchange (DEX) in order to speed up execution and settlement of trades. The public chain, which is built with Arbitrum technology as a layer 2 on Ethereum, supports 24/7 trading of tokenized stocks and crypto. It allows decentralized finance (DeFi) instruments like lending where users can lock in their stablecoins to earn a return. 

The company is entering the terrain of native crypto exchanges, like Coinbase, who themselves have started to offer the tokenized variety of traditional equities. Blockchains that track and secure digital assets finally seem to gain in adoption and utility after years of infrastructure build-out. A decade back, the famous slogan ‘blockchains not bitcoins’ made the rounds in TradFi circles, where established firms explored the idea to use them due to their immutable and transparent ledgers, allowing fast transactions and settlement. 

In the first three weeks of its existence, Robinhood's chain reached $9 billion of cumulative DEX volume, overtaking the Singapore-based shooting star Hyperliquid, which popularized perpetual futures, “perps,” on indices and commodities, largely benefiting from an increase in market volatility after the start of the Iran war. 

In Robinhood's case, however, so-called meme coins still account for the lion's share of the volume. Tokenized real-world assets, nevertheless, are expected to become the main driver of activity in the near future.

Where's the leverage? 

RWA’s and perps have become the latest crypto growth story, following in the steps of stablecoins. Monthly RWA volumes across exchanges have grown from about $85 billion in January to about $470 billion in June, according to data from The Block. 

Tokenized equities benefited from pre-IPO trading of SpaceX stock, as well as from demand for futures of semiconductor names such as Micron or SK Hynix. Crypto exchanges on top offer leverage and round-the-clock trading.

Traditional venues are alarmed. Both Nasdaq and the New York Stock Exchange will switch to 24/5 trading by end of the year. Representatives speaking at the Moofest pointed to the much higher liquidity and trust that should help to fight off the digital newcomers. At least for the time being. 

At its own event, Robinhood mentioned that it received the green light from MAS in Singapore and started offering by-invitation-only service in the city-state. If and when it will offer perps or tokenized stocks remains unclear. Clients can buy crypto for now through the licensed partner exchange Bitstamp. In Indonesia, Robinhood acquired the local brokerage firm Buana Capital Sekuritas and digital asset trader Pedagang Aset Kripto.

Moomoo doesn't work with a third party in Singapore when it comes to crypto, however it uses Fireblocks as custodian. The company didn't mention and seems not to plan to have a blockchain at this point. 

Peter Noszek, co-founder of the SuperAI conference believes that Singapore might be in a great spot, occupying a strategic position in between the US and China, the dominant players in the space, currently locked in an arms race for hegemony. While the US with its potent frontier models has taken the lead in the enterprise world, the Chinese open source models are swiftly catching up, appealing to consumers that look for more affordable solutions.

Singapore could benefit both sides. It already brings together infrastructure builders, developers, policy makers and researchers on neutral ground, at the intersection of the East and West. It further boasts an innovation-friendly environment and a society with an overall favorable view on the industry, a skilled workforce and capital-rich investors.

Both GIC and Temasek have participated with significant amounts in the funding rounds of Anthropic and OpenAI. Singapore’s enthusiastic developer community, meanwhile, celebrates the ongoing AI revolution with regular meetups across the city, experimenting with the latest iterations of DeepSeek and Claude.

The country offers a gateway for Chinese firms to connect with the Western world and for US companies to enter Asia. Anthropic and OpenAI are aggressively expanding their labs in the city-state. China’s tech behemoths with significant AI initiatives, such as Tencent, Alibaba or ByteDance, already have a visible presence here. It could be an ideal location for a world-leading conference one might think.

The SuperAI event already lined up senior representatives from nearly all relevant firms in the sector, discussing the latest in the ongoing infrastructure build-out and model development. The actual AI superstars and thought leaders however didn’t show up (yet).

The role of headline speaker fell to tech investor and innovator Balaji Srinivasan who also has a favorable view on the region, to an extent that he moved his domicile to Asia.

Srinivasan, who is a regular in Singapore’s tech event scene, runs his latest venture, the Network School, from the fringe of the city, just across the border in Johor, Malaysia. The school hosts a bunch of digital nomads trying to solve current-world problems. It organizes talks such as the Quantum Day, earlier this month, analyzing how quantum computers might threaten the underlying cryptography of bitcoin.

Another top event for Singapore?

Similarly to SuperAI, Srinivasan organizes his Network State conference in Marina Bay Sands, which celebrates tech utopists and libertarian ideas, debating concepts like seasteading, private money and longevity. The event is held in proximity to the Formula One race, Milken Summit and Token2049 that make Singapore for a few days the global center of the conference circus attracting high-caliber speakers from all over the world.

Token2049, the sister event of SuperAI, has turned into the Woodstock of crypto, drawing 25,000 visitors to the Marina Bay area. It has in past editions also been luring celebrities like Donald Trump Jr., the Winklevoss twins or Vlad Tenev, founder of the brokerage app Robinhood.

AI is likely to be a main focus at this year’s meeting in early October. So are prediction markets, another hot topic du jour, with Polymarket’s Shayne Coplan already confirmed as speaker.

The next SuperAI conference, meanwhile, has been scheduled for September 2027, likely to make it easier for the industry megastars to participate. It might provide the city with another shiny event and possibly help cement Singapore’s vital role for the global tech scene.

We have our “roaring 20s” after all, with the astonishing rise of AI, the prospect of a robotic era, with unprecedented capital spending by the leading tech behemoths, notably in the US and China. We are likely living through another epoch-defining technological revolution, comparable to the printing press, steam engine, and more recently the computer and the internet. Very soon AI agents will have taken over routine tasks in offices and homes, transportation will have become largely autonomous and we will discuss space exploration that could include regular trips to the moon.

There’s also a lot of skepticism about the general benefits of the new tech. Autonomous vehicles, to start with one of the most amazing innovations of late, these days receive a bad rap. In the US, where the technology is advancing at a fast pace, robotaxis have obstructed ambulances and driven passengers into flooded areas. They also take the jobs of hardworking drivers, we are being told. While much can be improved, rejection has in many cases turned ideological and categorical. De-growth zealots are pushing arguments that sound similar to those that accompanied the advent of the automobile or electrification of our cities, expressing the fear of facing the new thing that vastly improves our standards of life.

AVs have been showing convincing numbers when it comes to lessen human tragedy on the road. Waymo claims to be able to reduce injury-causing crashes by up to 85%, based on data it collected over its more than 200 million driverless miles. Regarding serious injuries or deaths, the data is even better. And their vehicles are still at the beginning of their learning curve…

We first experienced self-driving cars and Waymo in San Francisco in late 2024, when the service was only available to the public for a few months. The first encounter with the technology is striking. There is a fleeting moment of trepidation when the vehicle sweeps into motion for the first time. But the moment the journey ends, only one thought remains: let’s do it again!

Reviewers of AVs like to describe how the car is able to change lines on its own, stop at traffic lights and let pedestrians cross the road. Of course that’s what it does! The revolution here is much more profound. There’s no conductor, so you, the passenger, are in charge. You adjust the temperature, put on your own music and sit back like a boss, getting chauffeured around flawlessly by a powerful computer.

Stepping into the future

When visiting Shenzhen in May, we had to try Pony.ai, which operates in the city. Although the company has an English app, signing up was a bit difficult. Google’s app store is not available in China, so we had to fall back on a local-language clone. After finding and downloading the app, everything else went smoothly.

Pony.ai, which operates 1,700 robocabs in over 20 cities globally, hasn’t had any reported injuries—or worse—involving customers. One empty robotaxi caught fire last year. Its AVs use Nvidia’s Drive platform and Orin X chips, which do the actual magic. Pony.ai claims that the latest version can perform about 4 quadrillion AI calculations per second.

At the core of Pony.ai’s philosophy is its world model. It assumes that the AI is showing driving behaviors that are different from a human driver, the surrounding traffic will therefore react differently. The only way to learn is to put it on the road. More driverless operations lead to better data, a stronger onboard model and more level 4 deployments. Once that loop starts, it compounds, the company says. Improvement becomes self-guided.

The Hong Kong and Nasdaq-listed firm just reported that its revenue is forecast to surpass by more than 3.5 times that of last year and that it will expand its fleet to more than 3,500 vehicles by the end of this year. It also said that its Gen-7 robotaxi already achieved a positive per-vehicle operating profit in Guangzhou. In February, operations in Shenzhen also returned a profit at a per-vehicle basis.

Pony was unaffected by a Chinese government “rectification campaign” that followed a major disruption of market leader Baidu-owned Apollo Go in Wuhan, which had left hundreds of its robotaxis stranded during rush hour.

During our Shenzhen ride, the Pony.ai car stopped abruptly a couple of times—for a short while—when it was passing a crowded city centre with cyclists and pedestrians crossing left and right (captured on video). The virtual assistant on the screen promptly apologized.

In Singapore, Pony.ai in April started testing with a safety driver/assistant in the Punggol district, in collaboration with local transport firm ComfortDelGro. Similarly, its competitor WeRide, number three in China, is running trials in the same neighborhood, together with Grab. It’s baby steps for the AV companies that are already able to deliver much more. We hope for a swift rollout across the island.