Summary

Anthropic’s $13 billion raise at a $183 billion post-money valuation is defensible only if its extraordinary growth persists. Rory O’Driscoll’s reported-number math moves from roughly $1 billion of opening ARR to perhaps $8–9 billion at year-end, then assumes $30 billion exiting next year; that would imply about $20 billion of FY26 GAAP revenue and an 8–9× forward multiple. “High absolute number, absolutely,” but at that trajectory, “it’s not crazy.”
Canva is approaching $4 billion of revenue, growing close to 40%, re-accelerating and still profitable—making its roughly $42 billion valuation look restrained beside AI peers. Its employee secondary was oversubscribed despite offering no primary capital: Canva has held more than $1 billion of cash and has been profitable for eight years. Cliff Obrecht described Canva’s journey from a 50× multiple in 2021 to roughly 10× now; Rory’s framing was that markets are “the madman in the back seat,” while durable growth can “cover a multitude of sins.”
The bigger AI risk is not first-year demand but whether products cross from early adopters into mainstream distribution and renew in year two. Cliff expects enterprise buyers to consolidate today’s “spray and pray” portfolios over the next 12–24 months; Canva itself runs four coding tools and lets employees choose two major LLMs. Its product rule is “workhorses, not gimmicks,” and Cliff attributes only about 20% of Canva’s re-acceleration to AI rather than the core organic and international flywheels. Canva also says it is deliberately optimizing for LLM discovery: ChatGPT-originating images rose from 0.02% to more than 5% of uploads.
OpenAI’s $1.1 billion all-stock Statsig acquisition and Meta’s $14 billion Scale transaction show two radically different forms of deal engineering. Statsig’s late investors received the same headline value as its last round but rolled into OpenAI stock. Jason speculated that Vijay would gain a major operating role, calling the transaction “perfectly engineered to check everyone’s boxes.” Rory sees Scale as the opposite: predictable talent conflict, weakening asset value and a possible future write-down after Meta paid richly for what may become “the empty husk of Scale.”
Lovable’s jump from $1.8 billion to a proposed $4 billion, alongside a reported $9 billion Vercel round apparently in progress, reflects both real execution and venture FOMO. Harry said Lovable was already around $125–130 million ARR and could finish the year at $185–200 million, but Rory argued rapid markups must come from new information, an initially underpriced round or social validation from the prior investor. Jason’s counterexample was Replit: similar product and roughly $100 million ARR, yet valued at $3 billion only 13 days after Lovable’s $1.8 billion round.
Public SaaS is not roaring back, but modest AI-driven re-acceleration is enough to produce violent upside when expectations are depressed. MongoDB’s GAAP growth returned to roughly 24% and its shares jumped 40–45%, even though that merely restored a growth rate and revenue multiple seen two years earlier. The call was less “new golden age” than Mark Twain: “Reports of my death were greatly exaggerated.”
Jensen Huang’s $3–4 trillion AI-infrastructure opportunity requires economics far beyond today’s software pass-through. Rory calculated that a 20% return on $4 trillion demands $800 billion of annual profit—roughly “another four Microsofts, another four Facebooks”—while existing cloud revenue is only about $150–200 billion. Canva already spends around 10% of revenue on AI, but expects routing, distillation, self-hosting and on-device models to reduce unit costs; it is therefore moving to unified credits and hybrid seat-plus-consumption pricing.
The investors’ closing rule was to treat a missed early investment as information, not a permanent prohibition. Two positive observations separated by time are vastly stronger than one fresh snapshot, especially when an AI founder repeatedly evolves the product and reveals a “survivor gene.” Rory’s prescription: “That is the tax you pay for being stupid. Pay the tax and just get off the stupid train”—then keep funding proven winners when company progress, not merely price, justifies it.

Deep dive

1. Anthropic’s $183 billion price rests on growth duration

Rory reconstructed the reported trajectory cautiously: roughly $100 million two years ago, a $1 billion run rate entering this year, around $5 billion now and perhaps $8–9 billion by year-end. Averaging opening and closing ARR could produce approximately $4–5 billion of current-year GAAP revenue.
His valuation case assumes another sharp deceleration that is still extraordinary: ARR moving from $9 billion to $30 billion next year, versus nearly 10× this year. That would average to about $20 billion of FY26 revenue, making $183 billion post-money only 8–9× forward sales—“if the growth lasts,” with a heavy underline under if.
Rory saw an institutional imperative among growth funds: they either make “a big balls call that this isn’t gonna work” or secure exposure to at least one of the two dominant LLM companies.

2. Canva’s fundamentals have caught back up with its valuation

Cliff expects Canva to finish the year very near $4 billion of revenue, grow close to 40% and continue re-accelerating. Jason’s comparison was blunt: at roughly $42 billion, Canva is “barely 10X,” despite growth at scale that breaks conventional assumptions about TAM.
Canva’s round was also super-oversubscribed; Cliff said it could have raised roughly ten times the amount of funds it took in. The current transaction is entirely secondary because Canva has more than $1 billion of cash and has been profitable for eight years. Fidelity is the round’s anchor and largest check; investor selection is being treated as preparation for long-term ownership through a potential IPO, not an exercise in squeezing out the highest mark.
Figma’s IPO reduced employee and investor willingness to sell, creating a supply shortage even before Canva satisfied incoming demand. Cliff tries to counter single-company anchoring by showing employees a spectrum of public comparables, their growth rates and the long-term consequences of each valuation framework.
Canva was valued at $40 billion in 2021—about 50× revenue—then fell to $26 billion in 2022 without its business stopping. Rory’s framing: Wall Street is “the madman in the back seat”; finance can be wrong by 5×, but sustained compounding eventually bails out that error.

3. Mainstream distribution will separate AI workhorses from gimmicks

Cliff worries that early-adopter enthusiasm has pulled forward revenue for many AI products. Reaching users in “Middle America” and across Europe is a different distribution problem from selling to people watching Product Hunt and X, making the climb from $50–100 million to $1 billion a genuine chasm.
Rory sharpened the concern into year-two retention: apparent product-market fit may be temporarily concealed by enthusiasm. Cliff agreed consolidation is coming, but distinguished Canva’s approach—put AI inside an established workflow to make visual creation “quicker, faster, and better,” building “workhorses, not gimmicks.”
Internally, Canva deliberately funded experimentation: Cliff was willing to open another $10–50 million budget, runs four coding tools and lets employees choose two among Gemini, OpenAI and Anthropic. The current “spray and pray” phase should narrow to clear winners over the next 12–24 months.
Cliff attributes only about 20% of Canva’s re-acceleration to AI. The larger mechanism was renewing core flywheels, international expansion and organic acquisition—90% of new users arrive organically—rather than treating the installed base “like a wet tea towel that you need to wring out.”
SEO is about 15% of Canva’s 90% organic acquisition; word of mouth, user flywheels and design sharing are now larger channels. Cliff said Canva is the number-one productivity app on ChatGPT and the fifth-highest domain ChatGPT refers to. Images uploaded to Canva from ChatGPT rose from 0.02% to more than 5%, and Canva has deliberately built a team for LLM optimization.

4. Statsig’s flat-price sale satisfied almost every constituency

Harry called OpenAI’s $1.1 billion all-stock purchase of Statsig “cheap” against roughly $75 million ARR and an exceptional team. The price exactly matched Statsig’s last financing led by Iconiq, making the transaction look flat in dollars but potentially attractive in what shareholders received.
Rory’s late-investor interpretation was pragmatic: “I thought I was investing in Statsig. Now I’m investing in OpenAI. Worse things can happen.” With OpenAI stock potentially rolling into another valuation event, the effective revenue multiple may not differ dramatically.
Jason saw a carefully balanced bargain: late-stage preference holders receive their money’s worth, Vijay may obtain a major OpenAI operating role, and Iconiq gains nine-figure OpenAI exposure that Jason argued it could not otherwise obtain after leading Anthropic’s round. Angels wanting another independent growth card may be the least satisfied group.

5. Meta’s Scale structure contains the problems critics expected

Rory treated Scale’s personnel departures as predictable rather than revelatory. Paying one recruit $100 million, another $1 billion and another $10 million, then asking them to share authority, virtually guarantees ego conflict—even if everyone involved is talented and Meta is imposing some deliberate structure.
The asset problem is more fundamental: frontier training now demands sophisticated biology, mathematics and reasoning data, not merely “this is a dog, this is a cat.” With competitors available, Meta’s teams will buy the best output rather than favor Scale simply because it is Meta’s largest venture investment.
Rory’s reconstruction was that Meta put $14 billion into Scale as if it were worth $14 billion, investors took out the proceeds, and much of the key team moved or departed. He expects auditors may eventually question whether the remaining business still supports that carrying value, producing an entertaining but painful write-down.
Jason called the recruits “a pack of mercenaries” assembled in weeks, perhaps months, with foreseeable casualties and power struggles. Rory’s hedged verdict was that the bet feels more like Meta’s metaverse miss than WhatsApp or Instagram: “My gut is this is more like the latter than the former. I could be wrong.”

6. Rapid AI markups mix new evidence, mispricing and social proof

Cliff chose FOMO as the dominant explanation for Lovable’s reported $4 billion valuation and the apparent $9 billion Vercel round in progress: investors believe the AI curve has more room, even if it is nearer the top than 12–18 months ago. He still worries about mainstream adoption, while judging Lovable well positioned if execution continues.
Harry supplied the bull case: Lovable had reached roughly $125–130 million ARR, planned to end the year at $175 million, and might instead land at $185–200 million. At 2–2.5× next-year growth, that could become $450–500 million, making a $4 billion entry less absurd than the speed of the markup suggests.
Rory’s three-bucket framework asks whether new performance justified the increase, whether the prior round was underpriced, or whether one prestigious investor’s commitment merely validated a higher bid from the next. Anthropic’s re-acceleration, aided by Claude Code, may qualify as genuine new information; many quick follow-ons probably do not.
Jason highlighted market inefficiency: Lovable closed at $1.8 billion on July 17, 2025, then Replit—“basically the same company,” with roughly the same $100 million ARR—closed at $3 billion 13 days later. Different lead investors appeared sufficient to produce dramatically different prices.

7. Excess capital helps only when founders preserve optionality

Harry argued that if Lovable becomes a $20–50 billion company, the difference between entering at $2 billion and $4 billion may barely matter. Rory’s logical reply: if nothing changed operationally, that simply means the first round was underpriced—the private-market equivalent of an IPO doubling immediately.
Rory rejected the blanket warning against large balance sheets. A shakeout is likely in many markets within two years; capital squandered on performance marketing is destructive, but cash held for a decisive opportunity can be valuable. “The really great ones” take the money without letting it distort operating discipline.
Cliff said Canva historically minimized dilution, raised as little as possible and sometimes ran “to the bones” while targeting the highest feasible valuation. It worked, but he now calls that a high-risk maneuver and tells founders to be somewhat overcapitalized instead.

8. Public SaaS has escaped the death sentence, not regained hypergrowth

Jason estimated that perhaps half of leading public B2B companies are finally receiving an AI tailwind. Box, Zoom and MongoDB showed evidence; Salesforce has demand but not yet the revenue, while Atlassian, Dropbox and Asana had not shown the same acceleration. The hosts also cited Snowflake among the week’s strong results.
His standard was unforgiving: a company with more than $1 billion of installed distribution had 18 months to exploit AI and has “no excuse” for failing to re-accelerate by the end of 2025. Even so, outside MongoDB and Snowflake, he characterized the improvement as modest rather than a return to hypergrowth.
MongoDB’s year-on-year GAAP growth returned to about 24%, and the stock rose 40–45%. Rory noted that this merely restored a rate and revenue multiple seen two years earlier; depressed expectations turned a moderate beat and stronger guide into an outsized repricing.
The broader lesson was relative performance, not euphoria. Once investors conclude “SaaS is dead,” merely doing reasonably well can trigger a violent reversal: “Reports of my death were greatly exaggerated.” Strong CEOs still needed to tighten execution rather than rely on AI as a universal explanation.

9. Canva is preparing for an IPO but distrusts the direct-listing experiment

Cliff said late-stage private scrutiny already resembles public-market reporting, while public investors now award higher multiples than private pools. Canva is becoming IPO-ready and hired Kelly, Zoom’s former CFO who led its IPO, although readiness and the actual listing date remain separate decisions.
Capital access is not Canva’s binding problem; employee liquidity is. After 13 years, annual secondaries remain “pretty janky” and can be nearly impossible in some jurisdictions. Cliff believes employees who created the value deserve straightforward access to it.
Canva’s 240 million monthly active users also create unusual retail demand. Cliff wants customers who contributed to its success to share in it, while Rory argued that forcing ordinary investors to access Canva through fee-charging private funds is structurally absurd.
Cliff remains skeptical of a direct listing because historical examples have not performed well immediately and the long-duration institutions likely to own half the stock are “pretty anti-direct listing.” Harry estimated that, had Figma used a direct listing, it might have started around $36–40 rather than $75. Rory conceded the practical point: founders rarely want to be “an experimental baby” when their life’s work is landing.

10. AI’s infrastructure bill still lacks a convincing return bridge

Harry cited Jensen Huang’s expectation that Blackwell and Ruben could address a $3–4 trillion AI-infrastructure opportunity over five years. Rory’s return math was daunting: a 20% return on $4 trillion requires $800 billion of annual profit—effectively “another four Microsofts, another four Facebooks.”
Jason resisted betting against the build-out because adoption has barely begun. He cited Salesforce nearing $50 billion of revenue while AI penetration remained around 0.1%, and imagined as much as $200 billion of eventual AI attachment; Canva already records billions of AI uses per month, with usage accelerating.
Rory compared the thesis with today’s $150–200 billion combined cloud market. Even if every software vendor spends as much on AI as on cloud, that implies only about $200 billion of revenue and perhaps $100 billion of profit—insufficient to justify $3–4 trillion without far greater consumption or economics.
Cliff added an energy thesis: AI’s enormous electricity requirement may force faster progress in zero-emission generation, particularly nuclear. Rather than treating consumption as purely environmental damage, he expects necessity to make cleaner power more economically viable and accelerate the broader transition.

11. Canva is turning AI compute from an open-ended cost into metered value

Cliff said Canva’s AI expenditure already reaches roughly 10% of revenue, including foundational-model training and product inference. He expects that percentage to decline: a frontier image may initially cost four cents, while Canva is banking on reducing it to 0.02 cents within six months and treats the interim gap partly as marketing expense.
The long-run architecture routes only premium queries to expensive OpenAI or Anthropic APIs; Cliff expects perhaps 90% to run on-device or through self-hosted models. The counterforce is richer products consuming more tokens, so falling unit costs do not necessarily mean falling aggregate infrastructure demand.
Canva Code already has 20 million active users and is Canva’s most expensive product to serve. Expanding it toward a “Lovable Prime” experience could consume dramatically more tokens, illustrating why Canva cannot expose every capability indiscriminately across 240 million users.
Ahead of a planned October wave of deeply integrated AI products, Canva is modelling adoption at 20%, 50% and 80% of users. Unified AI credits will cap included usage, while heavy users move into consumption pricing; one marketer generating tens of thousands of assets cannot economically remain a flat $20 seat.

12. Time-series evidence is the antidote to venture regret

Rory urged investors to revisit companies they previously rejected. After passing on Amature in 2003, he bought aggressively a year later at twice the price; after passing on Box in early 2010, he returned nine months later once he recognized it as “the dumbest thing I did all year.”
Two positive data points separated by time carry “almost infinite” additional information over one snapshot. In AI, the strongest signal may be a product that changed three times while its founder kept adapting faster than competitors: “This guy has a survivor gene. Run, don’t walk.”
Cliff extended the argument to follow-ons: Canva’s best early investors created SPVs and new vehicles to preserve or expand ownership. Rory agreed that a validated outside-led up-round near a fund’s normal strike zone is often the moment to “do every dime,” provided operating evidence—not prestige alone—supports it.
Harry preserved the key objection: a price inflection without a company inflection is not validation; he would rather pay at $800 million–$1 billion after substantive progress. Rory accepted the distinction but retained the closing discipline: when execution proves the original pass wrong, “pay the tax and just get off the stupid train.”

Is Non-Consensus Investing Overrated?49. What Can China Learn from Israel’s Tech Innovation? | A 10-Year Retrospective from a Hard-Tech VC