Leemon Baird first published his work on hashgraph consensus in 2016, positioning it as an alternative to traditional blockchain architectures. With a background in computer science and a career spanning both academia and industry, Baird co-founded Hedera to commercialize the technology.

His academic trajectory is notable for his early work in neural networks and reinforcement learning during the 1990s, a period when AI research was navigating what would later be called the «AI winter.» Since then, the Hedera project has evolved in a landscape crowded with competing distributed ledger approaches, each claiming technical superiority and targeting different segments of the market.

Baird, a speaker at Consensus 2025, transitions easily between technical explanations and business strategy, reflecting the dual challenges of building both a novel technology and a viable ecosystem around it.

This interview has been condensed and lightly edited for clarity.

CoinDesk: Your hashgraph algorithm emerged in 2016, during a period when many alternative consensus mechanisms were being proposed. What technical limitations of earlier approaches were you specifically trying to address?

Baird: I love computer science and the math side of it—inventing things and solving problems. When I became an entrepreneur 25 years ago, it was the same process. The core of what I always do is trying to understand the fundamental problem we’re trying to solve. What is the real question? What are we really trying to accomplish? And then you build on that and solve that problem. In blockchain, the fundamental question I asked was: Bitcoin is cool, but it’s slow and not as secure as it could be with ABFT [Asynchronous Byzantine Fault Tolerance]. It burns a lot of energy and isn’t as flexible as we might want. I wondered if, at the very bottom layer in the consensus itself, there might be a way to avoid burning lots of energy while still being fast and secure. Is it possible to achieve ABFT—the strongest kind of security—while also being super fast and not burning electricity or dumping carbon into the atmosphere?

I started working on this in 2012 as one of many math problems I was exploring. Initially, I was convinced it couldn’t be done. I’d pick up the problem, play with it, and convince myself it was impossible—over and over again. But in 2015, I realized that by throwing in two hashes, suddenly it all falls into place. You can have ultimate speed—essentially at the speed of the internet—while also having ultimate security with ABFT. And it’s proof of stake, so you don’t waste electricity.

From a business perspective, the question was: what is the right way to govern this? When we look at blockchains, they often claim, «We won’t have any governance. Anyone can help do it.» But power over time can consolidate—you end up with a handful of developers or people behind the scenes controlling everything.

With Hedera, we started differently. We made governance decentralized from the very beginning. We brought in some of the biggest organizations in the world—top universities and businesses spread globally that people trust and that have reputations to protect. They balance each other, creating checks and balances, and together they govern the system.

It was about addressing the fundamental question: what do you really want in governance? What will give you true trustlessness, or at least a lower bar of trust needed to fully trust the system? That was our answer—approaching business questions with the same rigor as mathematical ones.

CoinDesk: You’ve mentioned RWA tokenization, carbon credits, and stablecoins as key use cases. These are areas where nearly every major blockchain is focusing. What specific implementations on Hedera have demonstrated meaningful transaction volumes or user adoption?

Baird: I would highlight four key areas:

First, AI is extremely exciting right now. The dangers of AI are also concerning, which is why we need to establish provenance, governance, and version control for AIs. People need to know if they can trust what’s happening. Hedera helps with AI in multiple ways, including permissioning data and potentially handling royalties for people providing training data. The work that EQTY Lab is doing with NVIDIA and Intel on Hedera is particularly exciting.

Second, real-world asset tokenization is transforming how we handle valuable assets. We have numerous projects tokenizing real estate, gold, diamonds, carbon credits, and even carbon emissions on Hedera. From the beginning of blockchain technology, I’ve maintained that what’s important isn’t pictures of monkeys or games—it’s that all things of value on the planet will ultimately be put onto these ledgers.

Third, stablecoins are essential if you want real-world adoption. We’ve created a Stable Coin Studio to make stablecoin development easy on Hedera. The Hedera Council includes many financial institutions doing impressive work with stablecoins.

Fourth, immutable data records are almost unique to Hedera through our Hedera Consensus Service. This allows messages to be sent to topics with access controls and immutable recording. Companies like Hyundai and Kia use this for emissions tracking throughout their supply chain.

CoinDesk: UCL’s research on energy consumption compared various networks, but methodology matters significantly in these comparisons. Proof-of-stake systems generally have similar profiles, with differences coming down to node count and hardware requirements. Does Hedera’s approach differ fundamentally from other PoS networks, or is the efficiency primarily from the current network configuration?

Baird: We’ve been thoughtful about energy consumption from the very beginning. Everything—from our algorithms to how nodes are run and governed, and the fact that we use proof of stake instead of proof of work—laid the foundation for low emissions from day one.

This created a virtuous cycle. Early adopters looking to tokenize carbon credits chose the green blockchain. Then, people who wanted to tokenize emissions and credits wanted to use the same blockchain where everyone else was doing similar work. This flywheel effect has made Hedera perhaps the most popular blockchain in the green technology space.

According to University College London, Hedera has the lowest carbon emissions per transaction of any blockchain. We also purchase carbon credits to be carbon negative. Being green was baked into our structure from the start, which is why we’ve become a leader in this area.

CoinDesk: We’re seeing a lot of projects attempting to combine AI and blockchain technologies. Given the different computational paradigms these systems operate under, what realistic integration points do you see beyond the marketing narratives?

Baird: The intersection of AI and blockchain is more significant than most people realize. On Hedera, we’re seeing real traction in several areas:

Providence and governance is critical. As we enter a world where everything will be AI-generated, we need to know we can trust the AI. This requires digital signatures to verify origins—whether human or AI-created.

Data permissioning is another crucial intersection. When thousands of people contribute small amounts of data to train an AI, each person needs control over their data—the ability to give or withdraw permission.

Looking forward, I’m most excited about using Hedera for identity and incorporating identity into AI systems. We’re reaching a point where you can’t distinguish AI-generated media from reality. The only solution is digital signatures—content needs to be signed by the photographer or reporter. But then you need trusted identity systems to verify those signatures.

CoinDesk: Having studied neural networks in the 1990s, long before the current AI boom, what’s your perspective on today’s large language models? Has something fundamentally changed in the technology, or are we just seeing the results of scale?

Baird: Many AI developments have unfolded exactly as I expected. When AlphaGo defeated the world’s Go champion, when AlphaZero mastered chess, when AIs conquered poker—I had anticipated all of it. They even used nearly the same techniques I’d envisioned. We simply needed faster computers.

Self-driving cars, too, are progressing precisely as I predicted, using the methods I foresaw.

But ChatGPT and Large Language Models (LLMs) have utterly astonished me. The transformer architecture from 2017—described in the paper «Attention is All You Need»—represented a breakthrough that no one could have anticipated. Back in the 90s, we were completely stumped by language processing—trying various approaches, but failing at every turn.

The capabilities of today’s LLMs still astound me, and their future remains unpredictable. Will they reach superintelligence? Or will they hit a ceiling? I don’t know—and I’d contend that nobody does.

Humanoid robots have also surpassed expectations. While their physical development has matched my predictions, their conversational abilities—powered by LLMs—have far exceeded what I imagined possible. In the coming years, they’ll begin with basic factory work before advancing to skilled trades like welding, plumbing, and electrical work.

These technological advances will make the Industrial Revolution look minor in comparison. Most people haven’t grasped the magnitude of these changes or how rapidly they’re approaching.