The SEC’s “War On Crypto” Ends With Coinbase Case Dismissed, & More
Want more research from the ARK Team? Have feedback on our publications? Click here to help inform our content creation.
1. The SEC’s “War On Crypto” Ends With Coinbase Case Dismissed
By: Frank Downing | Director of Research, Next Generation Internet
Last week, Coinbase announced[1] that the U.S. Securities and Exchange Commission (SEC) has agreed to dismiss its enforcement case against the company. If incoming SEC Chair Paul Atkins approves, the dismissal will end the “regulation by enforcement” movement led by former SEC chair Gary Gensler.
Under Gensler’s reign from April 2021 through January 2025, the SEC mounted many cases against Coinbase and its peers, alleging that the digital asset exchanges and staking businesses were running afoul of US securities laws. Despite good faith efforts to comply with the ambiguity of securities regulations applied to digital assets, cases against those companies continued apace.
Now, a new crypto-friendly administration has taken center stage. Several stablecoin and digital asset market structure bills should move swiftly through the Republican-controlled Congress, providing regulatory clarity for digital asset companies. In our view, the shift to common sense legislation and regulation will accelerate the adoption of public blockchains, benefiting investment strategies with meaningful exposure to digital assets.
2. Figure AI’s Helix Humanoid Robot Embodies Artificial Intelligence
By: Sam Korus | Director of Research, Autonomous Technology & Robotics
Last week, Figure AI released a video[2] showing two of its robots working together to put away groceries that neither had seen before—all with the help of Helix, a generalist Vision-Language-Action (VLA) model that combines perception, language understanding, and robotic actuation to control the upper body of the robot, as shown below.
The VLA model comprises two systems: System 1 is a low latency policy that handles robotic action, while the higher latency System 2 scopes and understands the scene and language inputs. Helix’s generalist, end-to-end training is boosting its capabilities at a rapid rate, mirroring the methods and achieving the success associated with autonomous driving. Now that it is adopting AI more comprehensively, ARK expects the humanoid robotic age to accelerate at a rapid rate.
3. xAI's Grok 3 Is Catching Up With OpenAI’s Advanced Models
By: Jozef Soja | Research Analyst
Last week, xAI released Grok 3,[4] marking a major milestone in the competitive generative AI race. Powered by a new 150k H100 and H200 graphic processing unit (GPU) cluster—equivalent to 200k Nvidia H100 GPUs and assembled in just 214 days—Grok 3 still is in training mode but is accessible to Premium X subscribers. Competing directly with OpenAI’s o3 and DeepSeek R1, Grok 3 is catching up quickly.
Early benchmarks[5] demonstrate that Grok 3 and its sibling Grok 3-mini excel in coding, math, and PhD-level scientific reasoning, outperforming OpenAI’s o3-mini but trailing the flagship o3. Notably, xAI offers “Think” and “DeepSearch” modes for deeper reasoning and detailed reporting from both the web and X, underscoring the company’s focus not only on proprietary data but also on swift productization and real-world utility.
Perhaps most striking is xAI’s pace of innovation. Founded in 2023 and quickly rising to the upper echelons of AI development, Grok 3 already ranks at or near the top of both academic and consumer-focused benchmarks like the Chatbot Arena.[6] As measured by the Graduate-level Google-proof Q&A (GPQA) PhD science-reasoning benchmark, xAI's models have surpassed many Google and OpenAI models, as shown below.
As xAI continues to scale its compute resources to a massive 1.2GW cluster, competition among large language models is likely to intensify. ARK will be watching Grok 3’s trajectory—with a focus on the ways in which consumer feedback and adoption guide its evolution.
4. The Quantum Computing Tree Is Growing A Topological Qubit Branch
By: Brett Winton | Chief Futurist
Last week, Microsoft announced a significant breakthrough in quantum computing with its Majorana 1 chip design, one that differs significantly from existing techniques. The promise of this new approach lies in its use of topological qubits that could be more stable and more capable of achieving lower error rates over time.
Lower error rates are crucial in quantum computing, as they allow more complex programs to run successfully on a smaller number of qubits. Traditional approaches like superconducting qubits or trapped ions typically require thousands of physical qubits to support each logical qubit, which are useful for computation. Topological qubits could offer a more favorable physical-to-logical ratio, potentially 100:1 or better—fewer physical qubits to achieve the same computational power, in other words—which could bring improved stability.
For example, breaking RSA-2048 encryption might require 4,000 logical qubits.[8] Microsoft's topological approach should be able to achieve the same goal with as few as 400,000 physical qubits, whereas alternative technological approaches might require 10 million physical qubits or more.
Though exciting, the Majorana 1 chip’s potential should be evaluated in the context of an industry that has seen many widely touted breakthroughs that have not delivered any meaningful commercial utility. Moreover, while claiming to have created topological qubits, Microsoft has not produced any unimpeachable evidence. Even if its chip were to perform as expected, its 8 physical qubits would match a milestone that trapped ion quantum computing achieved 20 years ago.[9] While the use of topological qubits offers theoretical advantages in stability and error rates, translating those advances into meaningful commercial applications will take time—possibly decades.
[1]?Coinbase. 2025. “Righting a Major Wrong.”
[2] FigureAI. 2025. “HELIX: A VISION-LANGUAGE-ACTION MODEL FOR GENERALIST HUMANOID CONTROL.”
[3] Ibid.
[4] xAI. 2025. “Grok3 Launch.” X.
[5] xAI. 2025. “Grok 3 Beta — The Age of Reasoning Agents.”
[6] Chatbot Arena. 2025. “Chatbot Arena (formerly LMSYS): Free AI Chat to Compare & Test Best AI Chatbots.”
[7] See Google DeepMind. 2024. “Gemini 1.5: Unlocking multimodal understanding across millions of tokens of context.” Anthropic. 2024. “he Claude 3 Model Family: Opus, Sonnet, Haiku.” xAI. 2025. “Grok 3 Beta — The Age of Reasoning Agents.”
[8] Quintessence Labs. 2025. “Breaking RSA Encryption - an Update on the State-of-the-Art.”
[9] Quantum Optics. 2011. “Trapped ions smash entanglement record.” Physics World.