Figma (FIG.US) Gross Margin Under Pressure; Arm (ARM.US) Revenue Hits Record High; Wall Street Admits AI Returns Are Difficult to Predict

According to Woofun AI, Figma (FIG.US) AI features have penetrated the workflows of 80% of its major clients, while Arm (ARM.US) achieved a record $1.44 billion in revenue for its Q2, setting the industry tone. Although technological penetration has notably increased, the pressure on Figma’s gross margin and the changing cost structure behind Arm’s business transformation together illustrate the complex picture of AI infrastructure and application layer commercialization in its early stages. Wall Street executives have openly admitted the difficulty in predicting AI’s long-term returns. Investors are now demanding that companies demonstrate exactly how technological advancements can be translated into profits, rather than relying solely on narrative-driven valuations.

Figma released its Q3 FY2026 financial results, reporting revenue of $391.4 million, up 43% year-over-year and $16.4 million above its prior guidance range. The company raised its full-year revenue forecast to $1.497 - $1.499 billion and its non-GAAP operating profit guidance to $158 - $163 million.

However, the market did not fully buy in: Figma’s share price dropped about 7% in after-hours trading. The company’s Q4 revenue guidance of $402 - $404 million implies growth will slow from Q3's 43% to around 33%, which is the direct cause of after-hours downward pressure. The market is now scrutinizing not just gross margin but also changes in growth rate itself. Woofun AI’s compilation shows Figma’s Q3 GAAP gross margin stood at 80%, with non-GAAP gross margin at 82%, lower than 86% from a year ago. The drop in gross margin was mainly due to increased inference costs from higher AI usage. The company is increasing the proportion of tasks handled by in-house models to improve cost efficiency, but has not yet disclosed specific savings. Figma has already proven that major clients are willing to continue using AI; the next step is to prove how much incremental profit this additional usage can generate.

If AI credit revenue grows faster than inference costs, Figma could establish a new source of income; otherwise, AI could become a continuously cost-intensive service that erodes gross margin.

AI product usage data was the most noteworthy aspect of this earnings report. As of October 31, over 80% of paid customers with annual contract value above $10,000 consumed AI credits weekly, and 65% used Figma Agent weekly. Figma MCP Server’s overall usage in Q3 tripled quarter-over-quarter, with write-back operations up four-fold. The company also launched a public beta for Code Layers in October, allowing users to build and edit code-based prototypes directly on the design canvas. These changes are redefining Figma's product boundaries.

The core of traditional design tools is to help users draw interfaces more efficiently, while Agent, MCP, and Code Layers enable the software to directly participate in design, code generation, and modification. Figma is attempting to evolve from a design editor into a work platform that connects design and development. The business model is also changing. In the past, Figma mainly derived revenue from seat subscriptions; now, AI credits have introduced a usage-based charging mechanism. The Code Layers public beta released in October has further blurred the line between design and development, with 80% of high-value clients now not only using AI-assisted design, but also beginning to rely on AI for interactions at the code level.

This transition from “tool” to “platform” has made Figma's revenue structure more complex, and has raised the bar for cost control. Though the non-GAAP gross margin of 82% remains high, the retreat from the historical peak of 86% indicates that cost pressures from AI scaling are starting to show in the early stage of wide-scale application.

Arm released its Q2 FY2027 earnings as of September 30, with revenue reaching $1.44 billion, up 27% year-over-year—a record high for a company second fiscal quarter. License and other revenue grew 50% year-over-year to $773 million, while royalty revenue rose 8% to $670 million. Importantly, data center royalty revenue doubled year-over-year. Arm’s business is undergoing more fundamental change. In the past, the company mainly earned income by licensing its architecture to chip designers; now, it is directly providing its self-designed Arm AGI CPU. The company stated it has secured purchase orders from multiple clients extending to calendar year 2027, with initial commercial deliveries planned for the fourth quarter of FY2027. Management expects the product could contribute up to $2 billion in revenue combined over FY2027 and FY2028. This is a company estimate, not confirmed sales revenue. Arm also noted that, among major hyperscale cloud providers, the share of Arm architecture servers increased from 15% in 2024 to over 50% in 2025.

This data applies only to selected major cloud providers and cannot be directly treated as Arm’s global server market share. Nonetheless, these figures reflect spreading demand for AI infrastructure. While GPUs handle massive parallel computation, CPUs are still responsible for general computing, task scheduling, and system management. As AI Agents need to run programs, invoke tools, and execute complex workflows, CPU performance and efficiency will also affect overall data center costs. For Arm, both opportunities and risks are expanding. Selling self-designed CPUs could bring far higher revenue per customer than licensing alone, but also means the company must handle chip delivery, supply chain, and product competition risk. Arm is shifting from primarily selling chip design IP to also participating directly in the chip product market.

At the Hong Kong International Financial Leaders’ Investment Summit, financial executives including Goldman Sachs (GS.US) CEO David Solomon, Citadel Securities CEO Peng Zhao, and TPG (TPG.US) founding partner Jim Coulter discussed the long-term prospects of AI. One key observation was that the most notable aspect of the summit was not the optimism about AI, but the fact that several executives openly acknowledged the enormous uncertainty in making long-term predictions. Solomon believes that short-term demand for computing power remains strong, but that long-term development paths are difficult to forecast.

Peng Zhao raised a more direct question: The pace of technological advancement has already exceeded investment cycles, forcing capital markets to make massive funding decisions in the absence of reliable long-term forecasts. Coulter reminded investors that the lesson of the dot-com era is that even if you are right about a technology changing the world, you can still lose money by betting on the wrong company. This reveals a structural contradiction in AI investing. Data centers, chips, and power infrastructure require capital commitments years ahead, but model costs, usage cases, and eventual business models can shift within months. The investment cycle of infrastructure is not aligned with the technology iteration cycle.

At the summit's November 2 welcome dinner, Chinese AI firms provided another perspective. Moonshot founder Yang Zhilin, during a closed-door “Technology for Good” conversation hosted by Shen Nanpeng, responded to the open vs. closed source debate, and discussed recursive self-improvement for models as well as generalizing to the physical world. Compared to simply increasing computing power, embedding AI into existing industries may provide a more direct path to validating commercial value—the perspective of Hong Kong Financial Secretary Paul Chan at the luncheon was that factories, hospitals, and logistics networks in mainland China are fertile ground for AI applications. But use case value does not automatically translate into investment returns. Model providers, cloud companies, chip manufacturers, and application developers all may create value, but the resulting profit is not necessarily distributed evenly. This is also the key investment issue currently facing AI: the market already believes AI will transform the economy, but it remains unclear who will receive returns matching the scale of capital invested.

America’s five tech giants saw their market cap increase by nearly $800 billion in a single day. Following the Democratic Party’s win of both chambers of Congress, Amazon (AMZN.US), Apple (AAPL.US), Microsoft (MSFT.US), Alphabet (GOOGL.US), and Meta (META.US) all saw a collective stock price surge. Investors believe that if new data center construction is restricted, tech giants with existing massive computing infrastructure may actually benefit. Amazon rose about 6.1%, Apple gained about 5.3%, in contrast to some data center operators’ declines. With clear election results and sharply falling oil prices, the Dow Jones index rose 1,569 points (3.03%) on November 4, and the S&P 500 gained about 1.6%. However, the energy sector fell 4.77%—the day’s worst-performing group—due to plunging oil prices, while consumer discretionary and financials led the gains.

Brent crude fell about 8.7% to $94.62 per barrel. Lower energy prices have eased market inflation concerns, but the Iran war and US political changes could still affect future trends. US voters have begun to directly reject AI data center expansion. In the 2026 midterm elections, voters in many places supported candidates or policy measures restricting data center construction. Texas Democratic gubernatorial candidate Gina Hinojosa unexpectedly defeated incumbent Governor Abbott, becoming the state’s first Latina governor and first Democratic governor since 1990. She advocates pausing approvals of new data centers, making her victory a focal point of attention for the AI infrastructure industry. Issues of data center power consumption, water usage, and contributing to higher residential power prices are evolving from local disputes into a national political topic.

Uber (UBER.US) faces new variables in the autonomous driving field, but not because it is scaling back its business. Waymo Co-CEO Tekedra Mawakana confirmed that the exclusive Robotaxi partnership with Uber in Austin and Atlanta will end in January 2028, with Waymo switching to accepting rides via its own app ahead of the original agreement’s May 2028 expiration. Uber stated that opening a dedicated Robotaxi entry point within its own app “was not Uber’s choice,” and emphasized it continues to collaborate with several autonomous driving companies: on its Q3 earnings call, Uber said about 700 fully driverless vehicles would be deployed by year-end, rising to over 2,000 by 2027. The real pressure comes from supply-side and core mobility business. CEO Dara Khosrowshahi said high fuel prices and constrained driver supply suppressed supply growth in many international markets; Q3 mobility bookings reached $29.512 billion, up 18% year-over-year, but revenue fell 2% to $7.520 billion, mainly due to changes in the UK business model. Overall, total bookings grew 20% for the quarter, non-GAAP EPS was $0.88, up 35% year-over-year, both above market expectations. The key to autonomous driving commercialization remains operational efficiency, not just technical maturity.

California’s Proposition 40 proposed a one-time 5% wealth tax on residents with net assets over $1 billion. Preliminary results show 55.6% opposed, suggesting the measure is likely to fail. The proposal drew the attention of Silicon Valley entrepreneurs and investors, with debate centering on the trade-offs among wealth taxation, capital migration, and local fiscal policy.