Why should AI investors reconsider Solana?

Author: Jeff Park Source: X, @dgt10011 Translation: Shan Ouba, Golden Finance

Since I entered the crypto industry, I have always held two goals in my heart—both independent and tightly connected: first, to reshape the underlying principles of sound money by relying on a censorship-resistant store of value (Bitcoin); second, to use technology to build a capital market that is better suited to the digital age, with higher efficiency and greater precision. And I believe that Solana today is precisely able to help advance this second mission. Before we go into the discussion, let’s first talk about a revolutionary technology company that is now well known to everyone, but was once quiet and unnoticed in its early years—NVIDIA.

In 2013, NVIDIA began developing its first supercomputer built specifically for AI, DGX-1. It launched a comprehensive overhaul starting from the underlying chip architecture, introducing the Pascal architecture chips using the FinFET process. This chip gathered three core ideal transistor characteristics: high precision, high energy efficiency, and controllability. Although this was the most significant hardware upgrade in the transistor field since the 1970s, more than ten years ago, the general public knew nothing about it. The reason Jensen Huang was able to foresee that the industry’s core bottleneck would shift from pure compute operations to data transmission between chips was precisely the inspiration brought by the 2014 NVLink interconnect technology. Even after the DGX-1 officially became available in 2016, NVIDIA still spent another ten years before growing into one of the world’s most legendary companies with a top market capitalization.

Many people mistakenly think NVIDIA’s success came from hardware circuits that are superior to those of competitors, but the real core factor lies in its well-developed software ecosystem. After Moore’s Law gradually began to fail, computer scientists used numerical algorithms to accelerate matrix operations, giving GPUs an unprecedented new instruction set. In Thinking Machines, Stephen Witt calculated that hardware transistors contributed only a 2.5x performance improvement, while the remaining nearly 400x speedup came entirely from a math tool library called CUDA.

Sharing NVIDIA’s journey serves two purposes. First, the path from scientific discovery to commercialization and deployment is long and full of twists and turns. Second, the simultaneous pursuit—on both the software and hardware layers—of the three core traits of high precision, high efficiency, and controllability is the central appeal of today’s breakthrough technologies. In this regard, NVIDIA and Solana share deeper common ground far beyond what meets the eye.

In my view, it is by no means a coincidence that Solana founder Toly comes from Qualcomm’s GPU R&D system. The project’s initial core bet was the Sealevel parallel execution environment: transactions across multiple shard states can be processed synchronously and in parallel on multi-core processors. At its essence, this logic is transplanting the underlying core idea of CUDA into the blockchain ledger system. The complex timing coordination logic required to achieve efficient parallelism gave rise to the consensus mechanism we are familiar with—Proof of History (PoH). Although Solana’s early architecture was already adapted to run on GPU-to-TPU hardware, the deeper lesson NVIDIA left for the industry is worth learning: the long-term core moat lies in a mathematical tool library that continues to iterate. The most direct evidence is that its underlying consensus mechanism has already been comprehensively rewritten: the Alpenglow upgrade will completely eliminate PoH and introduce a new consensus path that is faster.

Today, the market is generally bearish on the Solana mainnet, mainly because general-purpose public chains lack real application value. People judge the fit between its product and market by block-space utilization, but this approach has a hard flaw. However, there is a story from 2009 that is highly worth referencing: Geoff Hinton led graduate students (including Sutskever) to carry out machine learning experiments using NVIDIA graphics cards, and they applied to NVIDIA for free GPUs, only to be refused. At the time, AI was simply not included in the parallel-computing application scenarios planned by NVIDIA’s chief scientist. Yet in the end, AI became the most explosive deployment scenario for parallel computing—parallel computing itself is a general-purpose underlying technology. This early technology first gained initial momentum through large-scale purchases of GeForce GPUs by academia and interconnection among multiple cards, but Jensen Huang had already noticed this trend as early as 2004, long before the rest of the industry paid attention.

How did he catch the opportunity? This also explains how all frontier research that has no short-term commercial returns manages to obtain funding: by relying on a niche group of core enthusiasts. NVIDIA’s first dual-use chip for both graphics and parallel computing faced intense internal controversy because it raised the production cost of GeForce GPUs, making them more expensive than the competitor Radeon—this is what the industry calls the “CUDA tax.” The brilliance of CUDA is that it makes gamers bear the huge chip R&D costs. Correspondingly, Solana’s “CUDA tax” undoubtedly comes from the core loyal user group in the trading circle—a circle that blends gaming, entertainment, and finance. An objective fact is that all early cutting-edge technologies require a group of early developers and users who pursue extreme performance and deeply explore general-purpose scenarios. This is also a hallmark feature of how product-market fit takes shape. As long as you are willing to observe, this trend is already clearly visible today.

I have ample indications that a major breakthrough in crypto technology is still in its early stages of emergence. Solana is the first asset to have an on-chain trading market go live simultaneously with its listing: within the first 24 hours after the asset SPCX went live, Solana completed $52 million in trading across 51 liquidity pools. Solana is gradually becoming the settlement layer for AI, covering decentralized compute inference markets, DePIN data networks, and autonomous on-chain agents; it has also entered the top tier of the stablecoin ecosystem, with monthly on-chain transaction volume consistently ranking among the industry’s best. All of this is rooted in its underlying architectural principles from day one—strictly designed with a relentless pursuit of general-purpose performance. Of course, at this stage it is still only early data and there is not yet a conclusion. The key questions that remain are: can these specific segmented scenarios complete a commercially mature closed-loop, building a sustainable shared network economic model that is not limited to a single use case? And can related applications break free from dependence on short-term incentive subsidies, forming a self-reinforcing long-term positive cycle?

Jensen Huang’s friends have long said that his most core ability is not a talent for predicting the future, but a kind of “resonance perception”: based on what he sees in the present, combined with continuous and repeated communication with customers and employees, he uses logical reasoning to anticipate the industry’s long-term direction. I believe that, whether in terms of accumulated development or underlying thinking, the Solana community is already fully prepared and has the comprehensive capability to drive the next generation of industry opportunities to real-world implementation. To be part of this journey fills me with anticipation and honor, and I will do my utmost to contribute to the future of this promising industry.