Skip to main content
PROFITVISIONLAB
Equity Deep Research

Figma (FIG) Q3 2026: More AI Usage, Yet Figma Cut Prices

Figma's Q3 2026 revenue was US$391.4 million, up 43% and above the high end of guidance. But NDR fell from 136% to 131%, non-GAAP gross margin fell to 82%, and Q4 guidance implies only 33% growth. This report breaks down the slowdown: part is the base effect of last year's price change, and part is Figma's own cut to the unit price of AI credit add-ons. AI usage is established, AI pricing power is not; gross margin recovery depends on the share of work handled by Figma's first-party post-trained model, and a real improvement in shareholder value comes in 2028 at the earliest.

FIG

📌 Key Conclusions

  • Q3 revenue was US$391.4 million, up 43% year over year and above the high end of guidance; but NDR fell from 136% to 131%, non-GAAP gross margin fell from 85% to 82%, and Q4 revenue guidance implies growth of only 33%, with leading indicators weakening at the same time.
  • The slowdown came mainly from pricing rather than competition: part was a base effect from the Q1 2025 price adjustment, and part came from the August reduction in the unit price of AI credit add-ons and the shift of large customers to drawdown; gross retention remained around the mid- to high-nineties in percentage terms.
  • There is ample evidence of AI usage (credit consumption grew 2.5 times after the agent became a paid product, and MCP had about 1 million weekly active users), but AI credit revenue has not been disclosed; Claude Design and Figma are not on the same scale, a point the market misunderstands most.
  • Whether gross margin can recover depends on the share of work handled by first-party post-trained models, rather than market token prices; the company has confirmed the introduction of first-party models but has not disclosed that share.
  • Stock-based compensation was approximately 40% of revenue, and all free cash flow in the first three quarters went toward tax withholding payments for employee shares; a substantive improvement in shareholder value will not come until 2028 at the earliest. For Q4, watch whether revenue exceeds US$415 million, NDR holds at 130%, and the billings proxy returns to 25% or above.

Revenue exceeded guidance, but the key message this quarter is that Figma chose to lower prices. Q3 revenue was US$391.4 million, up 43% year over year and US$16.4 million above the high end of company guidance; AI usage also grew substantially, with weekly AI credit consumption growing 2.5 times after Figma agent became a paid product in October. In the same quarter, the company lowered the per-credit price of AI add-ons (providing more credit for the same price, with credit included in seats unchanged), shifted large customers to drawdown, and absorbed the inference costs incurred before the agent's general release; the effect of the Q1 2025 price adjustment is also fading. Leading indicators weakened at the same time: net dollar retention (NDR) fell from 136% to 131%, non-GAAP gross margin fell from 85% to 82%, remaining performance obligations (RPO) fell 1.4% quarter over quarter, and Q4 revenue guidance implies growth of only 33%. This report concludes: This is not weakening demand, but Figma trading a lower AI unit price for usage; Q4, a major renewal quarter, will provide the first test of whether this trade-off is justified.

Previous research review: From "no cracks in demand" to this quarter

The August Q2 earnings research concluded that there were "no cracks whatsoever in demand" for Figma: revenue growth accelerated for three consecutive quarters, NDR was 136%, and large customer expansion accelerated. That report also set a test: if Claude Design were indeed eroding Figma's budgets or use cases, one should expect to see "slowing growth, weakening NDR, or a slowdown in large customer expansion."

Three months later, all three signals have appeared. Revenue growth fell from 48% to 43%, and Q4 guidance fell further to 33%; NDR fell from 136% to 131%, returning to its level in the same quarter last year; year-over-year growth in customers with ARR above US$100,000 fell from 46% to 36%. Once the conditions are triggered, the cause must be identified. The previous report set these conditions to test whether tools such as Claude Design were taking budgets; the Q3 data instead point mainly to pricing: part is a base effect from last year's price adjustment, and part is Figma's own reduction in the price of AI credit add-ons.

Two items on the previous report's watchlist can be closed. First, beta products began charging: Figma agent became generally available on October 6 and consumes paid credit, and Code Layers, which remains in beta, has also been included in credit consumption. Second, stock-based compensation as a share of revenue did not continue to decline: it was 39.9% in Q2 and 40.5% in Q3. The previous report noted that the 129% figure for 2025 was a base distortion caused by a one-time recognition at the IPO, but still interpreted the move from 129% to 39.9% as "converging downward"; approximately 40% is Figma's current normal level, and it did not decline this quarter.

Growth: Revenue reflects past contracts, while leading indicators slow across the board

Metric

Q3 2025

Q4 2025

Q1 2026

Q2 2026

Q3 2026

Revenue growth year over year

38%

40%

46%

48%

43%

NDR

131%

136%

139%

136%

131%

Year-over-year growth in customers with ARR of US$10,000 or more

—

32%

37%

34%

28%

Year-over-year growth in customers with ARR of US$100,000 or more

—

46%

48%

46%

36%

RPO growth year over year

—

—

—

53%

31%

Revenue plus change in deferred revenue, year-over-year growth

—

55%

44%

34%

24%

non-GAAP gross margin

86%

86%

82%

85%

82%

Sources: Quarterly earnings releases and 10-Q filings. The Q4 2025 year-over-year customer growth rates are derived from the Q1 earnings release's statement that they accelerated by 5 and 2 percentage points; the second-to-last row is a billings proxy calculated by this report as revenue plus the change in deferred revenue for the quarter, not a company-disclosed figure.

Revenue is recognized over time from contracts signed in the past; the lower half of the table reflects new contract momentum. The leading indicators above slowed at the same time: NDR fell from 136% to 131%, growth in the number of large customers fell from 46% to 36%, RPO growth fell from 53% to 31%, and the billings proxy fell from 55% in Q4 to 24%. Management offered three explanations for the quarter-over-quarter decline in RPO: the Q1 2025 pricing adjustment made Q1 and Q4 the main renewal quarters; some business was already billed monthly and would not enter deferred revenue in the same way; and the company did trade away some short-term revenue, with credit repricing as one example.

The first point can explain a quarter-over-quarter decline, but not year-over-year growth. The billings proxy compares the same quarter: deferred revenue increased by US$40.4 million in Q3 last year and decreased by US$0.2 million in Q3 this year, a substantive slowdown in a same-quarter comparison, not a seasonal effect. The second point concerns measurement: monthly and pay-as-you-go revenue is not collected in advance and recorded as deferred revenue in the same way as annual contracts, so the higher the share of usage-based billing, the more this proxy understates the figure. This report therefore treats it as a directional signal rather than a precise figure.

Assessment: Q3 revenue outperformance was a continuation of contracts signed in the first half of the year, not a new acceleration; leading indicators have already turned. Figma exceeded the midpoint of its guidance by US$21.1 million and US$17.4 million in the first two quarters, respectively; if Q4 exceeds guidance by the same amounts, revenue would be approximately US$420 million to US$424 million, up 38% to 40% year over year, still below Q3's 43%. Conservative guidance can explain the 33% figure, but not the slowdown itself.

Reasons for the slowdown: Base effects and deliberate price concessions

On the earnings call, management attributed the Q4 slowdown to three factors: one was the base effect of last year's pricing changes, and two were this year's AI price concessions.

  1. The pricing and plan changes introduced in Q1 2025 are rolling off, and their impact will be fully absorbed in Q1 2027; Q4 is the main renewal season, making the impact particularly pronounced.
  2. AI credit add-ons were repriced in August, providing more credit for the same price and effectively lowering the price per credit in the add-ons; the credit included with seats remained unchanged.
  3. Large customers shifted to drawdown arrangements to purchase AI add-ons, allowing them to use credit in batches over a longer period. The company did not explain how revenue is recognized; this report infers that the recognition period will also lengthen accordingly.

CFO Praveer Melwani stated explicitly that the company would trade short-term growth for the durability of long-term growth if necessary, and acknowledged that this could depress NDR in the near term. Management also noted that pay-as-you-go consumption is excluded from NDR; the greater the share of usage-based billing, the more NDR understates customers' actual spending.

There is little clear evidence of customer losses to competitors. The company said gross retention remained in the mid-to-high 90s, approximately two-thirds of customers spending more than US$10,000 added Full seats at renewal, and the share of total ARR from customers with ARR above US$10,000 rose from 64% at the end of Q1 2025 to nearly 70%. If AI design tools were taking Figma's customers at scale, gross retention should decline first; that has not happened.

The market's biggest misunderstanding: Claude Design and Figma are not in the same league

Figma's stock fell approximately 7% on the day Anthropic launched Claude Design in April 2026, as the market interpreted it as a direct threat to Figma. Gizmodo. This report considers that a fundamental misunderstanding of how the two are positioned.

Claude Design serves the everyday visual needs of general employees: presentations, proposals, one-page briefs, and simple interface sketches, with exports to PDF, PPTX, HTML, and Canva, making it akin to an extension of office document tools. Anthropic's official positioning also encompasses exploration by designers, interactive prototypes, reading codebases to build design systems, and handoff to Claude Code. Anthropic (2026-04-17) Figma, by contrast, serves professional product and website design: collaborative design systems, component libraries, brand and accessibility standards, and the complete workflow for handoff to engineering teams through Dev Mode, Code Connect, and MCP. The large-customer examples in the Q3 prepared remarks involve connecting enterprise design systems to Make, MCP, and Code Connect to maintain brand consistency, accessibility standards, and governance.

The two differ in their primary users, outputs, and paying customers. Even tools such as Base44 and Lovable, which focus on AI-generated websites and applications, primarily produce rapid prototypes and small and midsize applications, making it difficult for them to replace professional design teams' work on design system governance, cross-team collaboration, and quality control. Figma's gross retention in the mid-to-high 90s and the continuing rise in the share of large customers are inconsistent with the narrative that AI design tools are replacing Figma.

Assessment: The slowdown is driven mainly by pricing, not competition. However, the two types of pricing factors must be considered separately: the roll-off of last year's price changes is a base effect, not a decision made this year; NDR fell from 139% to 136% before the August repricing and cannot be attributed entirely to AI price concessions; customer growth also slowed, and management did not directly address it. The AI credit concessions are the actual deliberate decision made this year, showing that AI credit is currently a tool for cutting prices, not a tool for exercising pricing power at Figma. Starting March 18, Figma imposed AI credit usage limits on all seats, requiring customers to purchase add-ons or switch to pay-as-you-go for excess usage, while Q2 revenue and gross profit accelerated together, demonstrating pricing power; five months later, Figma lowered the price per credit in add-ons in exchange for higher usage. Management's strategy is to establish usage habits first and then expand contract commitments. The logic holds, but it also means Figma does not yet have the conditions to charge premium prices for AI at this stage.

Precedents for cutting prices to gain market share

Large software companies have several precedents for cutting prices or bundling to drive AI adoption and monetizing later. The outcome depends on whether they have leverage to raise prices afterward:

Company

Approach

Subsequent outcome

Microsoft 365

Copilot was originally an add-on at US$30 per person per month; Copilot Business for small and midsize businesses launched in December 2025, and prices for most commercial suites increased starting in July 2026

Paid Copilot seats totaled approximately 15 million in January 2026, approximately 3.3% of 450 million commercial seats; approximately 20 million in April

Google Workspace

In January 2025, the Gemini add-on charge was eliminated while Business Standard rose from US$12 to US$14

Alphabet said Workspace revenue growth came from higher average revenue per seat and an increase in seats, without quantifying each contribution

Salesforce Agentforce

Added Flex Credits at US$0.10 per action alongside the US$2 per conversation price (May 2025)

Agentforce ARR was US$800 million, up 169% year over year; more than 60% of Q4 orders for Agentforce and Data 360 came from expansion by existing customers (FY26 Q4)

Adobe Firefly

Generative credit combined with a freemium strategy

Firefly-related ARR grew rapidly; management acknowledged that freemium suppressed ARR in the short term, while the full-year ARR growth target remained at 10.2%

Cursor

Switched to a credit allowance system in June 2025 at the same US$20 monthly price

Users pushed back; the CEO publicly apologized and issued refunds

Sources: Directions on Microsoft (2026-01-29), Microsoft (2025-12-02), The Register (2025-12-05), TechCrunch (2026-04-29); Constellation Research (2025-01), Zacks (Alphabet 2025 Q4); Salesforce press release (2025-05-15), Salesforce FY26 Q4 earnings release (2026-02-25); Adobe Q1 FY26 earnings call transcript (2026-03-12); Yahoo Finance (2025-07).

The successful cases share pricing leverage later in the process: Microsoft and Google recoup costs through suite price increases, while Salesforce converts usage directly into revenue through per-action pricing, with expansion coming primarily from existing customers. A disputed case such as Cursor arose from changing what a fixed price included without sufficient communication; Adobe's case shows that usage growth from freemium may not necessarily be reflected in ARR in the short term. This report found no public example of an AI credit price being cut, subsequently restored to its original level, and followed by a revenue recovery.

Assessment: Figma's approach falls into a higher-risk category. Figma increased the value of its add-ons this time without a corresponding seat price increase; the seat pricing lever was used in Q1 2025 and its effect is rolling off, leaving the company without the most effective lever in the precedents in the near term; the remaining recoupment channels are repeat add-on purchases, excess usage, and drawdown commitments. Moreover, pay-as-you-go revenue is excluded from NDR, so even if the strategy succeeds, NDR will understate or reflect the results later. A positive signal is that commitment amounts in three large-customer examples increased to 6 to 8 times their previous levels, resembling Salesforce's path of expansion primarily through existing customers; Q4 must show whether such expansion can extend to more customers.

AI monetization: Strong evidence of usage, but revenue evidence remains insufficient

The evidence of usage is the strongest in the past year:

  • As of October 31, 65% of paying customers with US$10,000 or more in spending used Figma agent weekly, and more than 80% consumed AI credit weekly; in Q2, more than 50% used agent.
  • After agent moved from beta to paid availability, weekly credit consumption excluding beta usage grew 2.5 times; among Org and Enterprise users who had used agent in beta, more than 80% continued to use it after charges began.
  • From early July to the end of October, the number of users consuming credit weekly among customers with US$10,000 or more in spending rose nearly 50%; after repricing in August, usage among customers who had purchased an add-on rose nearly 40% on average.
  • The MCP server was the fastest-growing product in Q3, with about 1 million weekly active users, total usage growing 3 times quarter over quarter, and write-back usage to Figma growing 4 times quarter over quarter; about 5% of credit consumption during the quarter was triggered by MCP requests.
  • Management cited three large customer examples: a technology infrastructure company, a global payments company, and a global pharmaceutical company increased their AI credit commitments within 5 to 6 months to about 6 to 8 times their initial purchases.

Explainer: Completing a website directly through MCP

MCP (Model Context Protocol) is an open standard introduced by Anthropic in November 2024 and works like a universal socket for AI: an AI coding tool that supports MCP can connect to external software in the same way, read data, and perform operations. Figma introduced its own MCP server in 2025, allowing tools such as Cursor, Claude Code, and VS Code to connect directly to Figma files.

The actual process is broadly as follows:

  1. A designer completes a web design in Figma, including layout, components, colors, font sizes, and other design variables.
  2. A developer selects a screen from the design in an AI coding tool and gives an instruction such as "implement a web page based on this design."
  3. The AI agent obtains structured information about the screen from Figma through MCP: the layer hierarchy, components, spacing, and design variables, rather than guessing from a screenshot alone.
  4. The AI agent generates front-end code from that information, and the developer can deploy it after checking and adjusting it.
  5. The reverse is also possible: an agent can write content back to Figma through MCP, for example by creating or modifying a design, which is the "write-back" usage that grew 4 times quarter over quarter.

For Figma, the significance of this channel is that the design becomes a source of design data when AI writes code, preserving Figma's place in the AI development workflow; but the agent and model completing the work may come from third parties, and users may not need to operate within Figma, which is the seat pressure described in the 10-Q risk factors below.

Revenue disclosure remains lacking. Figma has not disclosed the amount or proportion of revenue from AI credit, and the three large customer examples provide only multiples, without a base amount. The previous report identified this gap, and Q3 did not fill it.

Two risk factors in the 10-Q warrant comparison. First, third-party AI agents that create and modify Figma files directly through MCP may reduce the paid seats needed to produce the same results, and the Q2 10-Q contained the same wording. Second, the company has begun introducing plans that change who bears AI costs (the Q2 10-Q said it "may offer" them, while Q3 changed this to "has begun introducing" them), such as allowing customers to bring their own third-party AI keys, which will reduce usage billed through Figma credit. Q3 10-Q

MCP: Moat or leakage channel

Figma's MCP server is integrated with major AI coding tools, and the official list of supported tools includes Cursor, Claude Code, Codex, VS Code/GitHub Copilot, Gemini CLI, and Xcode; in October, it added 20 MCP partners, including Notion, Xcode, and Meta's AI agent Muse. Figma Help Center; Q3 earnings call transcript

Evidence supporting the moat:

  • Management said about half of new Pro teams used MCP in Q3, and customers using MCP grew their Full seats faster than those not using it (specific figures were not disclosed).
  • Full read access is tied to seats: View and Collab seats and the Starter plan have only 20 calls per month, while Dev and Full seats have 200 per day (600 for Enterprise); writing to files outside personal drafts requires a Full seat. Figma developer documentation; Figma Help Center
  • Write-back usage grew 4 times quarter over quarter, indicating that canvas is beginning to become a destination for agent writes, rather than only a source of data.

Evidence supporting the leakage channel:

  • The tool for writing to canvas is free and has no call limits during beta, Figma says it will charge based on usage in the future, and agent writes to canvas are not currently billed. Figma developer documentation; Rate limit documentation
  • Community-developed Framelink (Figma-Context-MCP) reads files through the Figma REST API, bypassing the official MCP credit and call limits, but the REST API has separate seat-tiered call limits (up to 20 per month for View/Collab seats); it has about 16,000 GitHub stars, and its npm package had about 146,000 weekly downloads (November 7 to 13, 2026). GitHub; Figma REST API rate limits

Assessment: At this stage MCP is closer to a moat, and leakage risk exists but has yet to appear in the data. Management's claims about adoption by new teams and seat growth are unquantified company statements that require verification through subsequent disclosures; the key points to watch are how canvas writes are priced after beta ends and whether the share of credit triggered by MCP can rise from 5%.

Assessment: AI has become a usage habit for Figma users, but it has not yet become a revenue engine for Figma. MCP has two effects: it makes Figma a source of design data for external agents, bringing in new customers and seat upgrades; the same channel also lets customers complete work through third-party agents and even their own model keys. Figma's response is to keep work on canvas through its own agent and first-party post-trained model; from early July to the end of October, the share of work handled by its first-party post-trained model more than doubled. The direction is right, while its effectiveness remains to be assessed by watching whether the share of credit triggered by MCP can continue to rise from 5%.

Profitability: The 82% Gross Margin Is a Deliberate Choice, While the Implied Q4 Margin Falls by Half

Non-GAAP gross margin fell from above 90% in the first half of 2025 to 82% to 85% this year. The 10-Q disclosed that technology infrastructure and hosting costs related to AI and paid user consumption increased US$29.5 million year over year in Q3; non-GAAP cost of revenue rose US$14.2 million sequentially, while revenue rose only US$21.3 million. The CFO said gross margin fell 3 percentage points from Q2 to Q3 because the company decided to cover inference costs for customers before the agent's general availability.

This explanation is reasonable and supported by evidence: the share of work handled by first-party models doubled, the average cost per agent message declined, and optimization of internal AI spending helped Q3 non-GAAP operating income reach US$47 million, with an operating margin of 12%; full-year operating income guidance was raised by US$30.5 million, slightly less than the US$33 million increase in revenue guidance.

Q4 is the key. The midpoint of full-year non-GAAP operating income guidance is US$160.5 million; subtracting US$135.2 million for the first three quarters leaves implied Q4 operating income of only about US$25.3 million and an operating margin of about 6%, half of Q3's level and below the 14% in Q4 last year. Management did not explain this gap on the earnings call, instead saying that Q3 investments would begin translating into gross profit dollars in Q4 and that cost to serve continued to improve. The figure may simply reflect conservatism: full-year guidance after Q2 results likewise implied an operating margin of only about 5.5% in the second half, yet Q3 alone reached 12%. The 6% therefore has two interpretations: Figma consistently guides low, or Q4 inference costs or expenses are indeed rising. Until actual Q4 figures are released, this report regards it neither as evidence of deteriorating profitability nor as a risk that has been resolved.

Can Falling Inference Costs Restore Gross Margin?

The decline in inference prices per unit is clear. Epoch AI estimates that the price of inference at GPT-4-level capability falls about 40-fold per year (with different capability thresholds ranging from 9-fold to 900-fold per year); a16z estimates that inference costs for equivalent capability fall about 10-fold per year. Epoch AI (2025-03); a16z (2024-11). The price war continued in 2026, with OpenAI cutting GPT-5.6 Luna prices by 80% in July. VentureBeat (2026-07-30)

Three factors, however, offset the decline in unit prices:

  1. Top-tier model prices have stopped falling. In the same price war, GPT-5.6 Sol remained at US$5 per million input tokens and US$30 per million output tokens, while Anthropic Opus 4.5 through Opus 5 remained at US$5 / US$25 (Opus 5.5 was US$4 / US$20); price cuts were concentrated in mid- and lower-tier models. Anthropic pricing
  2. Usage is growing faster than unit prices are falling. Research from the Stanford Digital Economy Lab shows that agent coding tasks consume about 1,000 times as many tokens as ordinary coding conversations. Stanford Digital Economy Lab (2026-05). Figma's own credit consumption also grew 2.5-fold after the agent's general availability.
  3. Pricing decisions reduce revenue per unit. Both the August repricing and drawdown reduce revenue per unit of compute.

What can improve gross margin is the model mix and pricing method, rather than market token prices. The most consequential variable is how much of Figma's work shifts to first-party post-trained models.

Does Figma use first-party post-trained models? Confirmed, but the share has not been disclosed. The Q3 press release explicitly states that the company has expanded its use of first-party and open-weight models in production; from July 1 to October 31, the share of work handled by its first-party post-trained design models more than doubled. As recently as Q2, management said it had begun introducing first-party models into Figma agent, making this a transition that began only in mid-2026. The company also disclosed three quality metrics: compared with the previous production version of the agent, the new version won or tied in 62% of head-to-head tests by professional designers; it also disclosed that user ratings were 33% higher and users undid agent actions 60% less often, but did not specify the comparison baseline for those two figures. In the Q&A, management said first-party models can deliver comparable quality with lower latency while also managing costs.

Two limitations must also be considered. First, first-party models are used alongside third-party models, not as a replacement; the Q3 10-Q risk factors still state that AI features on the platform generally rely on third-party models. This report infers from this that most inference is still paid for through external model providers (the same statement also appeared in the Q2 10-Q and is an ongoing risk disclosure). Second, the company has not disclosed the actual share of first-party models, the source of their base models, or the reduction in cost per message; the press release lists first-party and open-weight models together. This report infers that its first-party models may be post-trained on open-weight base models, but the company has not said so explicitly.

The order of magnitude of the cost gap can be estimated from peers. Cursor's Composer 2, post-trained on the open-source base model Kimi K2.5, is priced at US$0.50 per million input tokens and US$2.50 per million output tokens, about one-seventh of the previous generation Composer 1.5 (US$3.50 / US$17.50) and about one-tenth of Anthropic Opus 4.5 through Opus 5 (US$5 / US$25). Tessl (2026-03); VentureBeat. Intercom also claims that its post-trained model Fin Apex 1.0 costs about one-fifth as much to run as frontier models (company claim). VentureBeat (2026-03)

On this basis, if the unit cost of first-party models is 10% to 20% of that of third-party frontier models, the blended average inference cost would decline substantially as the first-party share increases:

First-party model share

0%

25%

50%

75%

Average inference cost (first-party models at 10% of third-party cost)

100

78

55

33

Average inference cost (first-party models at 20% of third-party cost)

100

80

60

40

Source: This report's hypothetical calculation, indexed to 100 for exclusive use of third-party models; Figma has not disclosed the actual share or cost ratio, and the figures in the table do not represent the company's actual situation.

The calculation shows that even with the same doubling of share, an increase from 10% to 20% lowers average cost by only about 9% to 10%, while an increase from 30% to 60% can lower it by about 32% to 37% (at a cost ratio of 10% to 20%). The key to a recovery in Figma's gross margin is not market token prices, but the share of work handled by first-party models. Until the company discloses the absolute share, a doubling of share cannot be directly translated into the extent of gross margin improvement; three signals should be tracked: the specific share of first-party models, the reduction in cost per agent message, and whether growth in non-GAAP cost of revenue is lower than revenue growth.

On balance, part of the Q3 gross margin reflects a one-time cost (the company covered inference costs before the agent's general availability), and the product shifted to consuming paid credit beginning in October. A partial recovery in gross margin from 82% is reasonable; there is not yet evidence to support a return to the 86% level before AI credit charges began (the second half of 2025). Management has not provided gross margin guidance and uses gross profit dollars rather than gross margin as its core objective, indicating that the company is willing to accept a lower gross margin.

Assessment: The gross margin decline was a deliberate choice, and management's explanation is acceptable; however, Figma has yet to demonstrate that AI can deliver operating leverage. Revenue grew 43% year over year, while the operating margin remained at 12%, the same as in the year-ago quarter, and Q4 guidance is lower still. Gross profit dollars, the CFO's chosen core metric, grew 36% year over year in Q3, below revenue growth of 43%. If gross profit dollar growth continues to lag revenue next quarter, “invest first, charge later” will become “continuous investment.”

Stock-Based Compensation and Cash: All Free Cash Flow in the First Three Quarters Went to Employee Stock Tax Withholding Payments

The gap between GAAP and non-GAAP results comes almost entirely from stock-based compensation. Q3 stock-based compensation was US$158.6 million, or 40.5% of revenue; the total for the first three quarters was US$475.1 million, or 43.4% of revenue. This is a noncash expense, but shareholders bear the economic cost, which can be examined in two ways.

The first is cash. Figma handles employee RSU through net share settlement: it withholds some shares upon vesting and pays employee taxes in cash to limit the increase in shares outstanding. The company spent US$210.7 million on these payments in the first three quarters, more than the US$183.3 million in free cash flow over the same period; Q3 tax withholding payments alone were US$49 million, also above free cash flow of US$41.5 million. Employee option exercises brought in US$89.4 million in cash, offsetting part of the spending. In other words, all of Figma's free cash flow in the first three quarters went to employee stock tax withholding payments.

The second is share count. At the end of September, unvested RSU totaled 63.89 million shares, about 11.9% of shares outstanding, excluding the CEO's stock award. Class A shares increased 0.6% from the end of July to the end of October, or about 2.3% annualized.

On a full-year basis, dilution is higher: according to the statement of changes in shareholders' equity in the 10-Q, common shares increased from 513 million at the end of 2025 to 535.1 million at the end of September, a 4.3% increase over the first three quarters, including about 9.52 million shares from employee option exercises; shares increased 0.8% in Q3 alone, or about 3.4% annualized.

Employee Compensation Is High. When Will Shareholder Value Improve?

Shareholder value in this section means per-share value and returns actually received by shareholders, rather than total equity on the balance sheet. Measuring returns actually received by shareholders requires accounting for both cash flow and share dilution. Annualizing the first three quarters, free cash flow was about US$244 million against a market capitalization of about US$10.4 billion, for a free cash flow yield of about 2.3%; over the same period, annualized share count growth ranged from about 3.4% (Q3) to 5.7% (the first three quarters). The free cash flow yield is insufficient to offset dilution; moreover, free cash flow in the first three quarters fell 10% from US$204.2 million in the same period last year to US$183.3 million, while the share count increased 4.3% over the same period, resulting in a marked decline in free cash flow per share. Measured as free cash flow less stock-based compensation, the first three quarters came to negative US$292 million; Q3 stock-based compensation of US$158.6 million was 3.4 times non-GAAP operating income of US$47 million for the same period, and GAAP profitability remains a considerable distance away. As of Q3, the company had not announced any public share repurchase program (the repurchases disclosed in the 10-Q consisted only of 22,000 shares repurchased at par value under vesting agreements).

There is no clear trend yet in the absolute amount of stock-based compensation: the first three quarters were approximately US$169 million, US$148 million, and US$159 million, respectively, with Q1 high, Q2 low, and Q3 up 7.5%; annualized from Q3 alone, the amount is about US$634 million. Assuming the amount stops increasing, a lower ratio can come only from revenue growth diluting it. The following calculation uses the midpoint of 2026 revenue guidance, US$1.498 billion, as the base:

Stock-based compensation as a share of revenue (assuming the Q3 annualized US$634 million remains unchanged)

2027

2028

2029

2030

Revenue growth of 30% per year

32.6%

25.1%

19.3%

14.8%

Revenue growth of 20% per year

35.3%

29.4%

24.5%

20.4%

Source: This report's calculation based on the Q3 10-Q and full-year guidance, not a company forecast. If stock-based compensation increases 10% annually as headcount expands, the ratio would not fall below 20% until 2031 (about 18.4%) under the 30% annual revenue growth scenario.

Improvement in shareholder value can be divided into three milestones:

  1. Stock-based compensation falls below 30% of revenue. Under both growth scenarios, this occurs in 2028, provided the amount of stock-based compensation stops increasing.
  2. Free cash flow consistently exceeds employee stock tax withholding payments, and annual share count growth falls below 1%. Only then would free cash flow begin to remain with shareholders rather than being used to offset employee shareholdings.
  3. Stock-based compensation falls below 20% of revenue, or share repurchases begin to offset dilution. This occurs around 2029 with annual revenue growth of 30%, and around 2031 with annual growth of 20%. Cash on hand of US$1.72 billion provides capacity for repurchases, but there is currently no such plan.

Assessment: Employees currently receive most of the benefits of Figma's growth, and substantive improvement in shareholder value will come no earlier than 2028. Cash on hand keeps this issue from posing a financial risk, but it also means Figma's valuation cannot be measured by free cash flow margin alone. The timeline depends on two variables: whether revenue can sustain growth above twenty percent, and whether the amount of stock-based compensation can remain flat. The former depends on progress in AI monetization, while the latter can be tracked directly through quarterly stock-based compensation amounts and changes in share count; if stock-based compensation grows by double digits in 2027, the timeline above must be pushed back.

Valuation: EV is approximately 5.8 times this year's revenue

Item

Value

Notes

Share price (11/13 close)

Approximately US$19.45

Approximately US$23.73 when the previous report was published, down approximately 18%

Shares outstanding

Approximately 535.6 million shares

10-Q cover, Class A plus Class B shares as of October 30

Market capitalization

Approximately US$10.4 billion

—

Cash and marketable securities

Approximately US$1.72 billion

As of the end of September, with no financial debt

Enterprise value (EV)

Approximately US$8.7 billion

Market capitalization less cash and marketable securities

EV / trailing 12-month revenue

Approximately 6.2 times

Revenue of US$1.399 billion

EV / midpoint of 2026 revenue guidance

Approximately 5.8 times

Guidance of US$1.498 billion

The closing price on the day after earnings (November 5) was US$18.44, down 16.7% from US$22.13 the previous day, a decline comparable to the one following Q2 earnings. JPMorgan lowered its price target from US$28 to US$24 and maintained a neutral rating; Citi trimmed its target from US$35 to US$34 and maintained a buy rating. MarketBeat; Yahoo Finance

The 9.9 times figure in the previous report was market capitalization divided by revenue and did not subtract cash; using the same cash-adjusted basis as this report, EV / trailing 12-month revenue fell from approximately 8.5 times to 6.2 times. For a software company growing at 40% with gross retention around 95%, a valuation of 6 times revenue is not high. However, two factors warrant a discount to this multiple: stock-based compensation represents 40% of revenue, and the growth rate is falling from above 40% into the 30% range.

Assessment: The share price already reflects the slowdown, but does not yet reflect the possibility that AI usage will translate into revenue. If Q4 proves that usage can translate into renewal amounts, 6 times revenue will be a relative low; if Q4 renewal performance is lackluster, the market will begin to price Figma as a company with 30% growth and stock-based compensation equal to 40% of revenue, and 6 times may not be the bottom.

Conclusion: AI usage is established, but AI pricing power is not

Returning to the central theme of this research series, the conclusion has three layers.

First, demand has not eroded. Gross retention remained around 95% to the high 90s, and approximately two-thirds of customers spending at least US$10,000 added seats at renewal; this customer group's share of ARR continued to rise. The previous report concluded that the threat from Claude Design had not yet appeared in the data, and that assessment remained valid in Q3.

Second, AI usage is established, but AI pricing power is not. Clear data supports growth in agent, MCP, and credit consumption, but Figma chose to lower the unit price of credits in August to increase usage. The previous report interpreted the first full quarter of AI credit billing and the recovery in gross margin as indirect evidence that AI monetization was beginning to materialize. That assessment must be revised: the March credit cap proved that customers were willing to pay for usage above the cap, while the August repricing showed that Figma had not yet been able to sustain higher prices.

Third, growth is shifting gears, and profitability has yet to keep pace. Revenue growth fell from 48% to 43%, with Q4 guidance at 33%; operating margin held steady at 12%, with Q4 implying approximately 6%; all free cash flow in the first three quarters went toward tax withholding payments for employee stock awards. The previous report stated that non-GAAP operating income and free cash flow should be monitored to assess financial health; this report adds that their actual balances after stock-based compensation must also be examined.

Follow-up will focus on three indicators, all to be revealed in Q4 earnings:

  1. Q4 revenue exceeds US$415 million, representing approximately 37% year-over-year growth. This level is only US$12 million above the midpoint of guidance, less than the beats in the preceding two quarters; failure to reach it would mean the cost of repricing is higher than management estimated.
  2. NDR is no lower than 130%. Management emphasized that NDR has remained above 130% for five consecutive quarters. A drop below that level in Q4 would show that price concessions are beginning to affect expansion among existing customers.
  3. Year-over-year growth in revenue plus the change in deferred revenue returns to above 25%, meaning the two items combined exceed approximately US$532 million in Q4, indicating that the slowdown in billings has stabilized. Q4 is the main renewal quarter, and large customer contracts using drawdown should be reflected here if billed upfront.

If all three are achieved, the strategy of trading unit price for usage will have worked, and this report will upgrade its assessment of AI from a "usage habit" to a "revenue engine"; if both revenue and NDR miss their targets, this report will instead conclude that Figma's AI price cuts were not an investment, but a forced response to competition. If only the third indicator falls short, the first step will be to examine whether a higher share of monthly payments and usage-based billing caused an understatement; that result alone will not change the assessment.

This report is updated through November 14, 2026. Financial data is through Q3 2026, and the share price is through the November 13 close. Sources: Q3 earnings press release; Q3 10-Q; Q2 earnings press release; Q1 earnings press release; 2025 Q4 earnings press release; Q3 earnings call transcript; Q3 prepared remarks. Sources for industry and peer data are cited in the relevant sections.

This report is based on public information for research and investment education purposes and does not constitute investment advice.

ProfitVision LAB|Independent Research