Is Meta Moving Hardware Toward Glasses-First?, & More

This week’s newsletter covers Meta's AR glasses, Google's Agent Payments Protocol, Nvidia's Rubin CPX GPU, and Coinbase's expansion of its Morpho partnership.

1. Is Meta Moving Hardware Toward Glasses-First?

By: Nick Grous | Director of Research, Consumer Internet and Fintech

During Connect last week, Meta unveiled Ray-Ban Display, its foray into consumer augmented reality (AR) glasses.[1] Priced at $799[2] and paired with the Neural Band for wrist-based input, the device includes a micro-display for texts, translations, and video snippets, enabling quick-glance interactions away from the phone.

While unlikely to replace the smartphone anytime soon, the glasses could be ideal for specific use cases—checking notifications, viewing photos, translating conversations—for which access in real-time matters most. AR glasses could become a platform that complements other mobile devices. In our view, Meta is hedging its distribution risk: refreshed Ray-Ban Gen 2 and the sport-focused Oakley Vanguard have broadened the glasses across style and function,[3] embedding AR into familiar consumer brands. Its focus on wearability suggests that Meta is prioritizing social acceptance as much as technical capability.

Among the questions we have:

• Will users continue to use the glasses after their novelty subsides?
• Will the Neural Band capture the right kind of input for output?
• How quickly will developers build apps that create reasons for consumers to wear them daily?

If Ray-Ban Display can deliver compelling answers to all three questions, Meta’s AR glasses could evolve from complementary devices to a disruptive force, siphoning time from smartphones and laying the foundation for the next major computing platform.

2. Google Introduces An Agentic Payments Protocol

By: Varshika Prasanna | Research Associate

Google’s new Agent Payments Protocol (AP2) appears to be making personalized agentic shopping a reality.[4] Whereas the traditional consumer shopping journey progressed from search to discovery and then purchase, generative AI is compressing the first two steps. Instead of sifting through countless links and open tabs, a single prompt can surface the perfect product within seconds. Embedding AP2 into every stage of commerce, the AI assistant should guide consumers to the right choices and complete purchases on their behalf.

AP2 is designed as an open, payment-agnostic standard that integrates seamlessly with the broader agentic stack, including Agent-to-Agent (A2A) and Model Context Protocol (MCP). With that flexibility, agents can pay instantly for purchases using cards, bank transfers, or even stablecoins. As agentic shopping takes hold, consumers are likely to delegate discovery and purchasing to their agents, transforming the consumer transaction process. Instead of minutes or hours browsing, searching, and comparing goods and services, agents can compress the shopping journey into a few seconds.

Agentic purchasing also should challenge marketplace moats by lowering the cost and friction of switching from one to another. To remain competitive, marketplaces will have to double down on operational excellence—delivering on price, selection, and speed—to appeal to agents that are satisfying consumer demand. Just as consumers gravitated toward marketplaces that offered seamless one-stop shopping, AI purchasing agents are likely to favor marketplaces with AP2. With higher traffic, the odds of successful transactions on the most competitive marketplaces should increase, driving market share gains.

3. NVIDIA’s Latest Chip Optimizes For Large Context Inference

By: Frank Downing | Director of Research, AI & Cloud

NVIDIA recently announced the Rubin CPX GPU (graphics processing unit), a purpose-built accelerator designed for massive-context inference.[5] Traditional GPUs optimized for training and short-context inference, but workloads are shifting. Coding assistants now must comprehend repositories spanning millions of tokens, while generative video requires models to “reason” over hours of compute-intensive content.

Rubin CPX focuses on the prefill stage of inference, during which compute demand is high and memory bandwidth relatively low. Trading costly HBM (High Bandwidth Memory) for Graphics Double Data Rate 7 (GDDR7), NVIDIA can deliver far more Floating point operations per second (FLOPs) per dollar, slashing inference costs. SemiAnalysis estimates that the component materials for Rubin CPX could be ~75% lower than those in the primary Rubin R200 chip.[6]

The disaggregation of inference into chips dedicated to “prefill”[7] and decode could improve the inference economics for agentic workloads, as model builders who charge per token can generate more tokens per dollar of hardware investment. NVIDIA estimates that customers could generate $5 billion in token revenue per $100 million invested.[8] If Rubin CPX gains traction, competitors like AMD could be forced to revisit their hardware roadmaps to reach performance per dollar parity with Nvidia.

Beyond GPUs, NVIDIA also announced a deal with Intel.[9] The company will invest $5 billion in Intel’s common stock and partner to develop custom data center and PC chips. Intel will design NVIDIA-custom x86 CPUs (central processing units) for AI infrastructure and x86 SoCs (system on chips) that integrate NVIDIA RTX GPU chiplets for PCs. The investment follows the US government’s decision to take a $10 billion stake in Intel and could help Nvidia curry favor with the Trump administration as it negotiates over chip sales to China.

4. Coinbase Is Expanding Its Morpho Partnership To Route USDC Into DeFi

By: ARK Digital Assets Team | Raye Hadi & Lorenzo Valente

Coinbase is leaning into a “DeFi mullet” strategy that offers a seamless user experience powered by DeFi (decentralized finance).[10] The company is building on its integration of bitcoin (BTC) lending into Morpho, a modular lending protocol that allows Coinbase to route user deposits to decentralized markets while preserving the familiar, user-friendly interface of its app.

Now, Coinbase is extending the Morpho integration to USDC[11] lending through an onchain setup on Base, Coinbase’s Layer 2. Base already accounts for more than 20% of Morpho vault balances. Users can lend their USDC on Coinbase and earn yields up to 10.8%, far above the 4.1% rewards available for holding USDC in the app, as of September 18.[12] Behind the scenes, Steakhouse Financial curates vaults that allocate liquidity across lending markets to optimize returns. In other words, Coinbase is becoming a user-friendly gateway to DeFi.

Morpho is surging. During the week of September 8–14, it generated a record $5.4 million in revenue, as shown below.

Source: Blockworks, data as of September 18, 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. Past performance is not indicative of future results.

The power of Coinbase’s distribution, paired with Morpho’s modular infrastructure, is creating a potent funnel of stablecoin liquidity: a regulated front-end channeling user deposits into decentralized vaults that could serve as the foundation for a broader suite of financial strategies. As liquidity deepens and integrations proliferate, Coinbase and Morpho could reshape the way stablecoins and lending markets are routed onchain.

SUBSCRIBE TO ARK'S RESEARCH

[1] Meta. 2025. “Meta Connect 2025.”

[2] Roth, E. 2025. “Meta Connect 2025: the 6 biggest announcements.” The Verge.

[3] Ibid.

[4] Parikh, S. and R. Surapaneni. 2025. “Powering AI commerce with the new Agent Payments Protocol (AP2).” Google Cloud Blog.

[5] NVIDIA. 2025. “NVIDIA Unveils Rubin CPX: A New Class of GPU Designed for Massive-Context Inference.”

[6] Patel, D. et al. 2025. “Another Giant Leap: The Rubin CPX Specialized Accelerator & Rack.” SemiAnalysis.

[7] Computationally intensive, “prefill” is the initial stage of large language model (LLM) processing.

[8] NVIDIA. 2025. “NVIDIA Unveils Rubin CPX: A New Class of GPU Designed for Massive-Context Inference.”

[9] NVIDIA. 2025. “NVIDIA and Intel to Develop AI Infrastructure and Personal Computing Products.”

[10] Khatri, Y. 2025. “Coinbase now lets users lend USDC onchain with current yields up to 10.8%.” The Block.

[11] USDC, a cryptocurrency pegged to the value of the US dollar.

[12] Based on data from Blockworks as of September 18, 2025.