ARK • Disrupt

Last week, xAI hit another AI milestone with the release of Grok 4.¹ Now available in two tiers, users can tap into Grok 4’s standard model either for $30 per month or through premium X memberships, while power users can access Grok 4 Heavy for $300 per month. Grok 4 Heavy is a clear competitor to OpenAI’s ChatGPT Pro, available at $200.

xAI’s tiered pricing approach not only democratizes access to cutting-edge AI but also highlights a surge in the value ascribed to compute-intensive reasoning. In other words, the “consumerization” of advanced AI is accelerating, lowering the barriers to entry for both entrepreneurs and developers.

Grok 4’s performance has leapfrogged the competition on critical benchmarks like the Graduate-Level Google-Proof Q&A Benchmark (GPQA), as shown below. xAI’s strategy of scaling massive compute clusters—potentially the largest currently in use—with a ten-fold increase in the computing power associated with reinforcement learning have led to a hockey stick increase in its performance relative to the competition. Its integration of tool use during pre-training is revolutionary. By empowering Grok to integrate external tools earlier in the training process, xAI has unlocked efficiency gains in real-world tests: on Humanity’s Last Exam, Grok 4 achieved 38.6% accuracy with tools, 13.2 percentage points higher than the 25.4% without tools. Other leading models like Gemini 2.5 Pro and OpenAI’s o3 have not benefited as much from tool use: Gemini 2.5 Pro and o3 scored 24.9% and 21.0% without tools, respectively, but only 26.9% and 21.6% with tools, underscoring the power of Grok’s unique approach.

Source: ARK Investment Management LLC, 2025. This ARK analysis draws on a range of external data sources as of July 11, 2025, which may be provided upon request. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security. Past performance is not indicative of future results.

Based on Application Programming Interface (API) pricing with a 3:1 input/output ratio, Grok 4 costs $6 per million tokens,² slotting neatly between OpenAI’s o3 at $3.50 and o3 Pro at $35.³ That said, as AI teams productize their models, Elon Musk’s vision extends beyond raw performance: xAI is empowering Grok to interact with complex engineering and design software—the same class of tools used by engineers at Tesla—with plans to roll out a coding agent, a multimodal agent, and even a video-generation model, all by year-end.

Grok 4 also is leveraging the distribution of Elon’s other companies and will roll out to Tesla cars soon, potentially turbocharging the productivity, creativity, and adoption of Tesla’s vehicles and aligning with ARK’s conviction that the symbiosis between human common sense and embodied AI will define the next decade.

Last week, Chai Discovery published its new Chai-2 model, marking a watershed moment for computational biologics. With only a 16% success rate, the antibody design improved upon previous methods by 100-fold, collapsing months of lab screening into two weeks of computation.⁴

The model's architecture unifies sequence generation and atomic-level structure prediction, producing functional antibodies, nanobodies, and miniproteins that bind specified targets. In tests across 52 proteins that had lacked known antibodies, half yielded viable binders, some achieving picomolar affinities rivaling optimized therapeutics.

Importantly, Chai-2 incorporated known protein interaction sites: it did not discover novel epitopes, so the model followed nature's existing roadmap instead of charting new territory. While significant validation, that constraint did raise questions about Chai-2’s real-world discovery capabilities.

Even so, Chai Discovery’s timing couldn't have been better from a regulatory perspective. The U.S. Food and Drug Administration (FDA) is shifting away from mandatory animal testing toward "New Approach Methodologies"—including AI models and organ-on-chip systems—creating fertile ground for computation-first discovery platforms. This convergence between technology and regulations suggests a fundamental reshaping in the development of biologics.⁵

Although its potential for novel target discovery is unproven, Chai-2's success with epitopes suggests that AI will accelerate therapeutic development for known binding sites. Notably, companies like Absci have advanced AI-designed antibodies into human trials: its ABS-101 for inflammatory bowel disease (IBD) recently entered Phase 1 a year after conception.⁶ As regulatory frameworks evolve and incorporate in-silico workflows, tools like Chai-2 position early adopters to capitalize on a future of more efficient and ethical drug discovery.

On July 4, the Bitcoin blockchain recorded an 80,000-BTC transaction, approximating $8.6 billion USD—or 419 million coindays destroyed⁷ or 1.1 million coinyears of time-weighted dormant supply spent.⁸ The massive spike, as shown below, has attracted the attention of both long-term investors and speculators.

Source: ARK Investment Management LLC, 2025, based on data from Glassnode as of July 12, 2025. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security or cryptocurrency.

Two theories about the wallet's identity are taking shape. One attributes the transfer to early bitcoin investor and evangelist Roger Ver, a prominent figure in the cryptocurrency ecosystem who had advocated for Bitcoin Cash.⁹ Last April, the U.S. Department of Justice (DOJ) charged Ver with tax fraud,¹⁰ suggesting to some observers that he could be consolidating or moving funds to facilitate a legal settlement. That said, no reliable evidence has connected him to the transaction.

A second theory, originating from researcher CryptoTux,¹¹ highlights the use of the “OP_RETURN” field that included four messages on the Bitcoin blockchain, a rare occurrence in standard bitcoin transfers:

LEGAL NOTICE: We have taken possession of this wallet and its contents Not abandoned? Prove it by an on-chain transaction using private key by Sept 30
[…]

According to CryptoTux, this transaction could be exploiting old wallet software vulnerabilities—not a deliberate or premeditated move by the original owner. Because these coins had been untouched since 2011, they may have been secured using outdated cryptographic practices such as low-entropy randomizers in the wallets of that era. An attacker could have inserted the OP_RETURN data to signal control of the funds, aiming to leverage common law—specifically the abandoned property doctrine¹² and adverse possession¹³— to obtain legal digital ownership of the coins.

Given the transaction’s size, market participants are sensitive to the implications. Large dormant coin movements often precede increased bitcoin volatility, particularly if linked to potential selling pressure. Instead, markets digested the news with little volatility, as the coins remained unmoved. Perhaps the September 30 deadline in the OP_RETURN message could increase bitcoin’s volatility during the next few months.

We will continue to monitor on-chain flows for signs that the coins are entering exchange wallets, signaling an intent to sell. Persistent metadata analysis and wallet clustering research could help narrow the identity associated with the transfer. As bitcoin’s ownership structure evolves, understanding rare but significant movements will be key to assessing market risk and opportunity.

Robinhood’s latest product suite¹⁴ marks a significant moment in crypto adoption. With the launch of tokenized equities, a proprietary Layer 2 blockchain, and an expanded crypto offering in the US and the EU, Robinhood is transforming from a trading app into a full-stack, crypto-native financial platform.

In Cannes, the company unveiled tokenized versions of US equities and ETFs for EU users. Issued on Arbitrum, a Layer 2 based on Ethereum, these assets will support dividend payments and enable 24/5 trading without commissions or spreads. By tokenizing equities, Robinhood is providing frictionless fractional ownership¹⁵ of US markets, signaling that institutions in TradFi (Traditional Finance) are migrating onchain.

Equity tokens are just the beginning. Robinhood announced it is developing its own Layer 2 blockchain, built on the Arbitrum stack, purpose-built for real-world asset (RWA) tokenization, seamless bridging, self-custody, and eventually 24/7 settlement. Like Coinbase’s Base, built on the OP (Optimism) Stack, Robinhood’s L2 will allow the company to control both the relationship with the end-user and value creation.

Robinhood also announced broader crypto offerings:

• Perpetual Futures: Launching in the EU with up to 3x leverage and an intuitive user interface (UI) to simplify margin management.
• US Crypto Staking: Ethereum and Solana staking now live for US users.
• Cortex AI Assistant: Coming to Robinhood Gold in the US to deliver real-time, token-level market analysis.
• Smart Exchange Routing: Fee tiering and best execution via order routing across multiple partner exchanges.
• Tax Lots & Advanced Charts: the sale of tax-specific lots, and the launch of charting for mobile.

Tokenization offers clear benefits: the potential of 24/7 liquidity, global distribution, programmable compliance, and digitally-native ownership. As it enters the digital asset space with stock tokens and its own blockchain, Robinhood is likely to strengthen the bridge between traditional and decentralized finance.