Launchbay Newsletter 1.08.2025
Newsletter
31 July

This Summer? All Gas, No Chill in Capital Markets

Figma just became the second most richly valued software company in the public markets—after Palantir

Its IPO was a blowout: shares surged 250% on day one, pushing enterprise value to $67B and fully diluted market cap to $69B. That puts Figma at a 55× forward revenue multiple (2026), more than 4× the software sector median—and more than double Cloudflare or Crowdstrike.
Only Palantir, trading at 80× next year’s revenue with a $400B EV, ranks higher on valuation multiple, according to the Bessemer Cloud Index.
Figma’s pitch: category dominance, fast growth, and real cash generation. Investors snapped up the IPO after four years of thin, low-conviction tech listings. One banker put it plainly: “Four years of pent-up demand for high-quality software IPOs is playing itself out in one trade.”

📊 Key numbers:

  • IPO price: $33
  • Day one close: $118
  • Projected 2026 revenue: $1.2B, up from $749M in 2024 (Goldman estimates)
  • Shares sold: 37M (~8% of total outstanding)
  • Shares traded on day one: 47M+ (lots of flipping)

Figma now trades well above Adobe’s failed $20B acquisition price, which regulators blocked. Employees who sold shares in June 2024 at $23.19—more than 75% below current levels—may be watching this IPO with mixed emotions.

Klarna and Firefly hint at a broader IPO thaw

Klarna is reportedly considering reviving its New York IPO as soon as September, eyeing the rebound in U.S. fintech stocks
Firefly Aerospace plans to price its IPO at $35–$39/share, which would give it a $5.5B valuation. Firefly’s story is one of explosive top-line growth: revenue surged 6× YoY, from $8.3M to $55.9M as of March. But losses widened too—$60.1M net loss, up from $52.8M a year prior.

Anthropic is raising again — this time at a $170B valuation

The round is expected to bring in $3–5 billion, as the company nears $5 billion in annualized revenue—up from $4B just earlier this month.
Anthropic needs massive capital to keep training and running its Claude models. Earlier this year, it told investors it expects to burn $3B in 2025, after burning $5.6B last year.
One of the biggest drivers behind this revenue acceleration? Code.
Anthropic has captured an estimated 42% of the $1.9B code generation market, compared to OpenAI’s 21%, according to Menlo Ventures. That share includes tools like Cursor, which rely on Claude. Claude Code alone is now generating $400M in annualized revenue, doubling in just a few weeks. It’s available via Claude subscriptions or API access—and its explosive growth is starting to test the limits of Anthropic’s economics.
In fact, usage is so high that Anthropic plans to cap some Claude chatbot subscribers later this month. Some $100–$200/month users are burning way more compute than they’re paying for. Claude Code launched just two months ago and may have been underpriced to pull users away from competitors like Cursor—which recently sparked outrage by moving to usage-based pricing with little notice.
Investors are watching not just growth, but gross margins. Anthropic claims around 60% gross margin, including direct and partner-driven sales (via AWS and Google Cloud). That’s notably higher than OpenAI’s ~40% last year and projected 50% this year.
There’s a twist: Anthropic told some investors that if it counted model training costs under cost of revenue (instead of R&D), it would be breaking even. That’s not how most model builders report—but it paints a picture of a company with more sustainable economics. For context: if OpenAI used the same accounting logic, its 2024 gross margin would be around –12%, due to ~$7.5B in training costs.
Framed that way, Anthropic isn’t just scaling fast—it’s positioning as the most capital-efficient foundation model company on the board.

OpenAI isn’t slowing down either

Annualized revenue has hit $12 billion, nearly double its pace from the start of the year—putting it well on track to meet its $12.7B projection for 2025.The growth is being fueled by surging usage. OpenAI now reports 700 million weekly active users across its ChatGPT products, up from 500 million in late March. That includes both consumer and enterprise users—driven in part by its expanded enterprise push and model integrations into Microsoft products.
But with scale comes cost. OpenAI now expects to burn roughly $8 billion in 2025, up $1B from earlier forecasts. The jump reflects growing inference demand, expanding model capabilities, and the sheer weight of serving hundreds of millions of users weekly.

Cohere is stepping up too

The company is in talks to raise $300M–$500M in equity at a $6.3B pre-money valuation.
Cohere expects to cross $200M in annualized revenue by year-end, up nearly 3x from $70M in February. And it’s not stopping there—management is projecting $4–5 billion in ARR by 2029, according to investor materials.
Unlike OpenAI or Anthropic, Cohere doesn’t just sell API access. It gives customers the option to run models on their own servers, a major draw for enterprises and government agencies handling sensitive data. That “bring-your-own-infra” model positions Cohere as a go-to for compliance-heavy sectors—something other foundation model companies can’t yet match at scale.

Hollywood today, robots tomorrow

Runway and Luma AI, two of the top players in AI video generation, are gaining real traction in entertainment—and already eyeing a far bigger market.
Imax announced it will screen winning films from Runway’s AI film festival in its theaters—a mainstream nod for generative video. Meanwhile, Netflix revealed it used AI-generated footage (a collapsing building in The Eternaut) for the first time in a show.
Screenshot 2025-08-01 at 16.54.32.png
But this is just Act I.
Both Runway and Luma are now in talks with robotics and autonomous vehicle companies to use their video-generation models for training. The logic: realistic, diverse, synthetic video is a goldmine for training perception models in robotics, especially where real-world footage is limited or risky to collect.
Luma AI CEO Amit Jain says robotics could become the company’s largest revenue stream. Runway agrees—adding that real-time video generation for robotics and gaming could eventually eclipse the film and VFX market.
In short: don’t think “AI video = movie effects.” Think simulation infrastructure for everything with a sensor.

Neuralink is thinking bigger—much bigger

Elon Musk’s brain-computer interface company expects to implant chips in 20,000 people a year by 2031, generating at least $1B in annual revenue. That’s a massive leap from today—where fewer than 10 people are known to have received the device through early trials.
A recent investor presentation outlines a bold roadmap:
By 2029, Neuralink aims to secure U.S. regulatory approval for its first product, Telepathy, which enables brain-to-machine communication. The company projects 2,000 surgeries/year at that stage, generating $100M in revenue.
Then comes the scale-up:
In 2030, Neuralink expects to launch Blindsight, a device aimed at restoring vision, and perform 10,000 surgeries/year, generating $500M+ in revenue. By 2031, a third device—Deep, designed to treat tremors and Parkinson’s—would help Neuralink hit 20,000 implants/year and cross $1B in revenue.
The company also plans to operate five large clinics by then, offering all three devices. But despite the vision, most of this remains speculative. Clinical trials are early, and none of the vision-restoring or Parkinson’s-targeting devices are in patients yet.

Kraken is eyeing a $500M raise at a $15B valuation

The crypto exchange is back in growth mode, rebounding hard from the 2022–23 bear market. In FY2024, Kraken hit $1.5B in revenue (+128% YoY) and posted $424M in adjusted EBITDA. Trading volume reached $665B last year, with Q2 2025 volume at $186.8B. Kraken now serves 15 million users globally and holds $43B in customer assets.
That growth is getting attention—but so is Kraken’s business model shift.
While spot trading fees still drive most revenue, Kraken has expanded into derivatives, staking, tokenized equities, and payments. Its 2025 acquisition of NinjaTrader accelerated its U.S. futures push. The company also launched “Krak,” a global money transfer app, and commission-free U.S. stock trading, signaling ambitions well beyond crypto.
Kraken is now the #2 U.S. exchange after Coinbase, but trails globally—#8 by spot volume with ~2% market share vs. Binance’s 39%. Its edge lies in fiat-to-crypto rails, where it commands over 40% of global stablecoin-to-fiat volume. And unlike Binance, Kraken wears its compliance badge proudly—no major hacks to date, and a broad, proactive regulatory footprint.
Notably, the SEC dropped its staking lawsuit in 2025 with no penalties, clearing a major cloud from Kraken’s IPO runway. It’s also the first exchange licensed under Europe’s MiCA framework (via Ireland) and holds U.S. and Canadian licenses.
Financially, Kraken’s raise implies a ~10x revenue / ~35x EBITDA multiple, ahead of Coinbase’s. At $15B, that’s $1,000/user valuation—about 2× Coinbase. The IPO thesis leans on continued top-line growth, product breadth, and a “compliance premium” in regulated markets.

Ramp is racing up the valuation curve

The fintech is in talks to raise $350 million at a $21B valuation, just a month after closing a round at $16B. Back-to-back raises like this have become common in AI—think Perplexity or Anysphere—but it’s rare air for a fintech.
What’s driving the demand? A hybrid revenue engine.
Most of Ramp’s revenue still comes from interchange fees, but its software revenue is growing fast—thanks to products like its travel management tool and a new wave of AI-powered finance agents. These agents aim to automate rote tasks for finance teams, like expense reviews and approvals.
Ramp’s overall business has scaled quickly. As of January, it was pulling in $700 million in annualized revenue—a steep climb for a company in a sector that’s historically struggled to command AI-era multiples.

Fireworks AI is raising at a $4B valuation

The company rents out GPU servers to developers running AI models, and it’s riding the inference infrastructure wave hard. Annualized revenue just crossed $200M, with expectations to hit $300M by year-end. That’s nearly $17M/month today, powered by customers like Cursor and Perplexity.
Fireworks doesn’t buy Nvidia servers itself—it aggregates supply from third parties and offers API access to open-source models like LLaMA, DeepSeek, and Qwen, similar to how developers tap into paid models like GPT-4o. Its pitch: run open-source models cheaper and faster than AWS or Google Cloud.
This puts Fireworks in a fast-growing, capital-hungry category of “inference providers” alongside Together AI and Baseten. Together AI, for example, was last valued at $3B and had $150M ARR as of March.
Fireworks' gross margin is ~50%, and it’s targeting 60% as it optimizes GPU usage. But the economics are tricky—resellers often have to keep up to 50% of server capacity idle to accommodate demand spikes. That’s a real tension between margin expansion and flexibility.
A looming threat? Nvidia itself. The chipmaker acquired Lepton in March and launched its own GPU cloud marketplace—a direct move into Fireworks’ lane. For now, Fireworks is riding the surge in demand for faster, cheaper inference—but the long-term moat depends on how well it can optimize, differentiate, and avoid getting squeezed by upstream suppliers.

Cerebras Systems is seeking up to $1B in fresh funding—potentially postponing its IPO

The AI chipmaker is one of the few serious challengers to Nvidia’s hardware dominance. But instead of going public this year as planned, Cerebras is now opting for another private raise—up to $1 billion.

Groq is in talks to raise $600M at a $6B valuation

Another AI chip challenger, Groq is leaning into investor appetite for Nvidia alternatives. The company has drawn attention for its Language Processing Units (LPUs)—custom chips optimized for ultra-low latency inference, particularly for LLMs.
Groq’s pitch: while Nvidia dominates training, Groq can win in real-time inference, especially in edge and enterprise use cases where latency matters more than throughput. The raise would mark a big step up from its previous valuation and comes as demand for specialized AI hardware continues to fragment beyond GPUs.

Surge AI, a data labeling platform powering the accuracy of LLMs, is reportedly in talks to raise $1 billion at a $25B+ valuation.

Cognition, the team behind Devin—the AI software engineer—plans to raise $300M at a $10B valuation. Devin made waves as the first end-to-end coding agent able to autonomously plan and execute software tasks. That has made Cognition a standout in the next-gen coding space, even as Anthropic and OpenAI battle for market share via APIs.

Oura quietly crossed $500M in annual revenue last year, growing 120%+ as demand for health-tracking wearables—and subscription-backed wellness—continues to accelerate. The company’s positioning at the intersection of biometric data, consumer hardware, and habit formation gives it strong recurring revenue fundamentals in a category often dominated by one-off hardware sales.

Cleo, the AI-powered fintech, doubled its revenue in 2024, driven by the rise of its conversational assistant. The chatbot—known for a bold tone and Gen Z appeal—encourages better financial habits but is also becoming a serious revenue engine. Cleo saw a 50% jump in subscription revenue, alongside a 3× increase in transaction fees, including interchange and cash advance-related charges.

# Popular Secondaries
Publications
31 July
IPO Barometer
17 July
Launchbay Newsletter 18.07.2025
16 July
State of VC Secondary market 1H2025 report
10 July
Launchbay Newsletter 11.07.2025
3 July
Launchbay Newsletter 4.07.2025