GitLab财报拆解：Flex定价、AI成本与收入确认机制

Four months ago GitLab looked like the clearest AI casualty in public B2B:

• The stock was in the high teens.
• In May the company announced it would cut roughly 14% of its workforce, about 350 people, exit 22 countries, shrink its geographic footprint about 37%, and strip out up to three layers of management.
• AI writes the code now, so why pay for the place the code lives?

Q2 FY27 came out after the close on September 1 and changed the entire narrative:

• Revenue of $286.3 million, up 21%, about five points ahead of its own guide and $13 million ahead of consensus
• Non-GAAP operating income of $42.6 million against a $31.2 million estimate
• Full-year revenue guidance raised
• The stock closed at $45.09, traded up 15-20% after hours, at roughly an $8 billion market cap, and then kept going

The reported 21% showed big acceleration — but was the slowest growth number in the quarter. Calculated billings grew 24% against 12% the quarter before, gross bookings were the largest in company history, and the company guided Q3 revenue below Q2 anyway. Understanding why those three sit together is worth more to most founders than the beat itself.

Here are the 5 Interesting Learnings:

#1. Revenue Grew 21%, Billings Grew 24%, and Q3 Revenue Is Guided Below Q2

Reported revenue: $286.3 million, up 21% from $236.0 million. Calculated billings: up 24%, double the prior quarter’s 12%. Gross bookings: the largest quarter in company history.

Then the Q3 guide comes in at $281-283 million, below the $286.3 million just printed, which works out to 15-16% year-over-year growth against the 21% they just did.

The cause is revenue recognition, and GitLab created it on purpose by moving customers onto Flex. Under the traditional self-managed model, about 15% of the contract value is license, recognized up front in the first quarter, with the rest ratable. Under Flex, the license component is recognized over the term because the customer re-elects the product mix every month. GitLab’s own illustration of a $100 annual contract shows 15% plus 21% landing in Q1 the old way versus a flat 20% per quarter under Flex.

The modeling guidance from the CFO: for every $50 million of self-managed available-to-renew that converts to Flex in FY27, about $5 million of revenue moves out of FY27 and into future periods, with a $13 million maximum estimated full-year impact. Bookings, billings, and cash collection are unchanged. Only GAAP timing moves.

One thing to be careful with here. The deck also says GitLab has not incorporated the potential impact of Flex into the guide at all. So the raised full-year number ($1.129-1.133 billion, up from $1.112-1.118 billion, 18-19% growth) sits on top of a headwind the company has quantified but not yet embedded. Faster Flex conversion means a better business and a worse revenue line, and the guide currently reflects neither.

A note on the number every headline led with. Both the CEO and CFO said “net ARR” grew 42%, and it got repeated everywhere as though total ARR grew 42%. It did not. GitLab’s investor deck puts +42% Net ARR growth on the highlights slide with a footnote pointing to the Appendix for definitions, and the Appendix defines eight metrics, none of them net ARR. It cannot be the ARR base either: revenue grew 21%, net retention was 117%, and customers over $5,000 grew 8%, so a base growing 42% is arithmetically impossible. On the call the same term appears as “new logo net ARR” and “first-order net ARR,” which only parses as an amount of ARR added. It is net new ARR booked in the quarter, and 42% is how much bigger that quarterly addition was than the year-ago quarter’s. That is a strong number. It is also growth in an increment rather than growth in a base, which is why it should never be printed next to a 21% revenue growth rate without a label.

#2. Flex Did More Than $20M From 130+ Customers in Six Weeks

GitLab Flex is one annual dollar commitment covering Premium and Ultimate seats, GitLab Credits for agent consumption, and new eligible capabilities as they ship. Customers choose the mix monthly and reshape it without contract amendments. Usage above the commitment is billed in the month it happens, with no re-contracting or true-up cycle. GitLab’s stated goal, in its own words on the slide, is for Flex to become the default way customers transact with it.

Six weeks in market: more than 130 customers, more than $20 million committed.

The metric the CEO told investors to track instead of revenue is Paid Consumption Run Rate, defined in the deck as GitLab Credit commitments, Flex commitments, and paid on-demand consumption, excluding trials and promotional credits. It exited Q1 at $15 million and exited Q2 above $40 million, against a target of more than $100 million by fiscal year end. Duo Agent Platform paid consumption alone grew about 50% sequentially.

CrowdStrike is running the same mechanic with Falcon Flex, which crossed $2.29 billion in ARR and about 39% of total ARR last quarter. One dollar commitment, drawn down across the portfolio, reshaped mid-term without a new contract. Two very different companies landed on the same structure within a year of each other, because the seat stopped being the unit that grows.

The buyer-side term to read closely is in GitLab’s own appendix: under the Reserved bucket, revenue is recognized ratably as delivered and the unused amount is forfeited monthly. Reserved is the floor GitLab expects to be 80%+ of FY27 Flex revenue. Commit-and-forfeit is how these pools work at CrowdStrike too. Flex is a better deal than fixed seats for a customer who consumes, and a worse one for a customer who commits optimistically and does not.

Ultimate is now 59% of total ARR growing about 35%, and 8 of the 10 largest Q2 deals bought Ultimate. SaaS is 34% of revenue growing 36%. The mix is moving toward the tiers where the consumption sits.

#3. Subscription Cost of Revenue Grew 76% While Revenue Grew 21%. AI Costs Are Way, Way Up

Non-GAAP gross margin: 86%, down from 90% a year ago. GAAP gross margin: 84%, down from 88%.

Subscription cost of revenue went from $21.8 million to $38.3 million, 76% growth against 21% revenue growth. Management guided full-year gross margin to 85-87% and attributed the compression to SaaS mix shift plus investment in the consumption products.

The nine-quarter series in the deck is the better view: 91%, 91%, 91%, 90%, 90%, 89%, 89%, 88%, 86%. Five points over two years, with the largest single drop in the quarter GitLab started selling agent consumption.

Where the AI bill lands differs by company, and the differences are large:

• Figma: roughly 500 basis points of gross margin, into COGS
• GitLab: 400 basis points, into COGS
• Box: 20 basis points, because Box sells the permissions and audit layer and buys little inference
• CrowdStrike: subscription gross margin up 100 basis points, while capex went from $30.5 million to $124.4 million, roughly 3% to 8% of revenue

Buying inference puts it in COGS and it compresses gross margin permanently. Building the infrastructure puts it in capex, where it depreciates over years and leaves gross margin alone. Selling governance around someone else’s compute barely shows up at all.

GitLab is buying. It has integrations with Anthropic’s Claude models, Amazon Bedrock, and Google Vertex, and an explicit model-neutrality strategy that makes the inference bill a permanent operating cost. Eighty-six percent is still a strong gross margin for a business selling agent consumption.

#4. 11,114 Customers Over $5K, Up 8%. And About 1,700 First Orders in the Quarter.

The customer counts:

• Customers with more than $5,000 of ARR: 11,114, up 8%
• Customers with more than $100,000 of ARR: 1,571, up 17%

The new business numbers:

• First orders: approximately 1,700, more than double a year ago, the largest first-order quarter in three years
• First-order net ARR: up about 40%