Amplitude also had a banging quarter, and we take a look at Meta's new AI pricing gambit. #
- Welcome to *. Cautious Optimism, a newsletter on tech, business, and power. Modestly upbeat
Thursday. Today we’re focused on earnings from Figma, and what its results (alongside Amplitude’s own) can tell us about the state of SaaS in the AI era. Both companies have credible AI stories. And yet one dove, while the other rose. If you’re into blizzards of numbers, today’s issue is for you.
Not to worry, there’s piles more to talk about! To work! — Alex
📈 Trending Up #
Shade… revenue growth atSandisk,Western Digital…Ukrainian strikes against Russian oil assets…neo-defense primes…shock, awe…knowing when you are listening to AI…Tesla and SpaceX’s chip plans…Omilia, now flush with $67 million more…media consolidation…
Meta’s new AI business model: Meta announced a new AI model yesterday, dubbed Meta Muse Spark 1.2. A coding-focused model, it follows Muse Spark 1.1, a general LLM that received generally strong reviews at launch for its balance of performance and cost. The coding model appears strong in benchmark terms for its price, much like its sibling. Muse Spark 1.2 costs the same as Spark 1.1 ($1.25 per million input tokens, $4.25 per million output tokens), but comes with an interesting pricing twist:
Ah, the glories of having enough compute. Meta can offer its coding model for effectively zero cost, trading near-term revenue for explicitly approved training data. Recall that major AI labs are viewed by many technology companies — including some of their own investors, Microsoft — as too cozy with customer data. This setup solves the issue.
- OpenRouter only lists the non-training version of Spark 1.2 on its pages, so all we can tell is that on its first full day of availability the model will
serve around ~13 billion tokensworth of inference demand. The Series A crunch (again): Venture investor Gil Dibner argues that the Series A stage of startup fundraising is broken. This is not the first (fifth?) time that we’ve heard this refrain. But it’s important to understand what’s busted today. Dibner argues that Series A investors are either putting capital in too late (following obvious PMF), too early (overvaluing pre-PMF companies), or investing ever-earlier (Pre-Seed/Seed) in hopes of catching a hot startup early.
- Dibner correctly points out that, historically, Series A rounds were bestowed onto startups that had reached (some) product-market fit and needed capital to scale. That’s not the case today.
Massive, multi-stage funds which need to write huge, late-stage checks (your fund size is your strategy) are part of the problem. Their early-stage checks are thus price-insensitive scouting missions, and the giant firms have little incentive to continue backing startups that fail to go instantly vertical. So, you wind up with lots of poorly priced deals, stuck startups, and a king-made set of later-stage unicorns. Manufacturing your own magic is a good way to generate above-average luck, but are the best startups truly winning? Or is capital (tail) wagging the startup (dog)?
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html) Trending Down
[📉](https://finance.yahoo.com/news/servicenow-pledges-1-5bn-investment-110000403.html)
Cybersecurity…AppLovin, after earnings…data center polling…Slate Auto’s niche dominance? …challenges to Paramount-WBD…personal courage…working in media…
AI prices: OpenAI’s surprise price cuts from a few days back are here to stay. Codex daddy Tibo told X that “the GPT-5.6 Luna price reduction by 80% is not a temporary stunt, it’s permanent,” pointing to the durability for “efficiency gains” for the new, lower price point. In response to the discount — and a further 50% coupon from OpenRouter — folks are using the hell out of Luna. The same dataset lists it as the fifth most popular AI model on OpenRouter, beating open-weight models from China like GLM-5.2, MiniMax M3, Kimi K3, and others. A few *incredibly *cheap models beat it on token demand at present, but, clearly, OpenAI has shown that American closed-source AI labs can compete on the price/performance axis with anyone. (At least with a promotion running!)
Figma and Amplitude
Figma and Amplitude
This morning, let’s slow down and spend a little time with earnings results from Figma and Amplitude. The two companies are software-as-a-service companies that are evolving to both take advantage of new, AI-powered capabilities and accelerate from new AI-predicated products. After reporting second-quarter results, however, Figma shares are down sharply (-15%) while Amplitude’s stock is heading higher.
Figma’s killer quarter: Design and product giant Figma reported revenue of $370.08 million in the second quarter ($350.8 million expected), and non-GAAP earnings per share of $0.08 ($0.04 expected). Figma grew 48% in the quarter, an incredible pace for a public company, and the company’s third consecutive quarter of accelerating revenue growth:
Net retention for the period came in at 136%, another eye-popping figure for any software company, let alone one of Figma’s scale. Free cash flow came in at $53.2 million, good for a 14% margin. And if we look at Figma customers with spend of more than $100,000 (annually), the cohort grew by 46% over the last year. That in hand, let’s vet Figma’s AI progress:
**Figma’s paid AI push is bearing fruit: **The second quarter was Figma’s first with a full three months of paid AI credit revenue. The company began charging customers for AI usage over set caps in March. Figma said in its prepared remarks that AI credits attached to every paid seat “make those seats more valuable,” helping drive two-thirds of $10,000+ annual spend customers to add seats at renewal. So much for the end of seat pricing.
Credit demand is real: Charging for incremental AI usage is a real business (Figma’s gross margins improved in the second quarter from the first). Figma’s larger customers are opening their wallets. The company reported that at the end of the second quarter, “over 80% of Paid Customers with more than $10,000 in ARR were consuming AI credits weekly.”
- Figma reports that its beta agent is being used by half of its customers that spend more than $10,000 with it each year; 20% of “weekly credit-consuming users on paid plans” exclusively used the agent.
**Now, the bad news. **Why did Figma drop so sharply after reporting earnings? Yes, it crushed expectations and turned in a stellar bill of health. Hell, Figma even raised its full-year outlook thanks in part to “strength in AI credit consumption for products being monetized today [and] positive early signal on the back of our new launches.”
**So what went wrong? **Figma’s Q3 revenue guidance was soft (about 1% higher than what the company reported in Q2). Compared to its strong second quarter, projecting third-quarter revenue of $373 million to $375 million was a surprisingly conservative estimate. Investors balked.
Asked about its modest third-quarter guidance, Figma CFO Praveer Melwani said his company does not include “products that are in early access programs or betas that are not drawing down paid credits” in its forecasts, content to wait “until those transition over to [generally available] products where we can observe how they actually monetize before taking credit for it in the guide.” Figma is also lapping year-ago pricing changes in the coming two quarters. (Figma doesn’t charge for AI credit usage for products in beta; more here.)
The lack of claimed credit for yet-to-fully-launch products is classic CFO caution. We need to peer through the grass to see what’s coming down the product pike that will convert beta usage to revenue in time. A lot, as it turns out. Here’s Melwani from earlier in the call:
Our products and features that are in beta and are still rolling out to customers—including Figma agent, Figma Make on local code, Motion, generative plugins, and Code Layers—do not currently consume paid credits. Early usage of these new products and features is trending ahead of expectations. We’re already seeing that the Figma agent is drawing new users to consume AI credits as well as driving existing users to deepen their usage.
The selloff says investors will not pay today for future revenues trapped in beta. Bulls will argue that products currently being tested will quickly graduate and monetize — perhaps soon enough to turn Figma’s falling value into a buying opportunity. The market is voting quickly. Weighing takes longer.
Then there’s Amplitude: In the second quarter, digital analytics and experimentation company Amplitude reported revenue of $100.89 million, ahead of an expected $97.81 million result. The company’s non-GAAP loss of a penny per share matched expectations, and it bested its own top-line guidance for the period (revenue between $96.9 million and $99.1 million).
Amplitude grew 21% in the quarter, ahead of 17% in Q1 2026, 17% in Q4 2025, and 18% in Q3 of last year. Its pickup of Statsig’s customer base (OpenAI bought the team) helped juice ARR growth in its most recent quarter by $16 million. The company’s backlog grew faster than revenue, with remaining performance obligations expanding to $483 million (+35% YoY). In raw terms, Amplitude generated free cash flow of $23.7 million, just over a 23% margin when we consider the company’s revenue result.
- Looking ahead, Amplitude expects revenue growth of 21% (YoY) in its current quarter to between $105.6 million and $108 million, or mid-single digits (percentage growth). Even better, Amplitude raised its full-year guidance to $407.2 million to $411.2 million (
from $397 million to $403 million). What’s driving the company’s accelerating growth? Demand from AI companies (“More than 40 AI native companies now pay us over $100,000 a year”), demand created by use of AI (“The more our customers build with AI, the more they need to measure”), and new pricing and packaging “leading to average ARR increasing, higher multi-product attach, and longer contract duration.”
And there’s more on the horizon. Amplitude is building a product — demoed during its earnings call — called Wave, which is designed to constantly scan customer data (analytics, experiments, replays) to come up with product recommendations, create plans to build them, and then help with their construction. If Wave works as claimed, it could prove a powerful tool that Amplitude can use to attract new customers and more revenue from its existing customer base.
**Summing: **Amplitude beat both internal and external expectations in the second quarter, raised its guidance, reported that the fastest-growing companies in the world (AI-native firms) use its products, and teased a future product that could make its platform far more valuable to its customers. All that was worth a 5% share-price bump.
- Figma and Amplitude showed us in the second quarter that SaaS companies can thrive in the AI era. Both companies beat trailing expectations, but Amplitude promised more looking forward. AI creates growth, but it does not buy investor grace. The difference was worth billions.