Back to Home

AI Startup Scorecard: $4.5B, $110M, $47M in a Week

Three AI startups announced three very different deals inside the same 48-hour window this week. Each one prices a different theory of where the money in AI actually sits right now.

Palo Alto Networks agreed to buy Applied Compute for about $4.5 billion in stock. Bessemer led a $110 million Series A for Boltz, a biomolecular design company born at MIT. Accel led a $47 million Series A for Bangalore-based Humyn Labs, which sells real-world training data to robotics companies. Same week, same asset class, three buyers with three different reasons.

The Three Deals, Side by Side

Applied Compute is the outlier. It handles post-training and inference for open-weight models, was founded less than 18 months ago, and according to its own November 2 blog post went from zero to a $100 million annualized revenue run rate in that time. It is also one of Nvidia's training partners for tailoring open models. Co-founder and CEO Yash Patil becomes Palo Alto's co-chief AI officer when the deal closes in fiscal Q3 2027.

The strategic logic is blunt. Palo Alto has made seven AI-related acquisitions in 2026, including CyberArk, Chronosphere and Embrace. CEO Nikesh Arora framed the purchase as a hedge against model diversity: "Every enterprise is going to run a mix of open and closed models, and every one of those models needs to be secured." He told Calcalist it could eventually become larger than the cybersecurity business that built Palo Alto.

Boltz sells something harder to price. It grew out of open-source models from MIT's CSAIL lab: Boltz-1 in November 2024 claimed AlphaFold 3-level accuracy, Boltz-2 in June 2025 added binding affinity prediction at 1,000 times the speed of physics-based methods, and BoltzGen designed protein binders. All three shipped with code and weights.

Commercialization began in January with the Boltz Lab platform, a $28 million seed round and a Pfizer collaboration. In June it released proprietary BoltzMol-1 and BoltzProt-1 and started charging in "molecular tokens." Boltz says paying-customer token volume has grown 50-fold since June, and that six of the world's 20 largest pharma companies are customers, with Pfizer, GSK, Takeda and BMS named. The Series A brings total funding to roughly $138 million, with no valuation disclosed.

Humyn Labs is the smallest check and the strangest business. It builds its own capture hardware, records work in factories, warehouses and homes across more than 20 countries, and turns that footage into labeled datasets across sight, sound, motion and touch. Co-founders Manish Agarwal and Ishank Gupta say it passed $50 million in annualized revenue within six months and serves about a dozen customers. Only $26 million of the raise is equity; the rest is venture debt, largely from Lighthouse Canton, with Prosus Ventures and Jump Trading in the equity.

Here is the scoreboard at a glance:

CategoryApplied ComputeBoltzHumyn Labs
Deal value$4.5B all-stock$110M Series A$47M ($26M equity, $21M debt)
Company ageUnder 18 monthsAbout 2 yearsUnder 1 year
Revenue signal$100M ARR, about 45xUndisclosed, usage up 50x$50M ARR, raise under 1x
MoatResearch team, platformOpen models plus pharma dataHardware, field ops, sites

Scoring the Categories

Price discipline. Applied Compute is the only deal where a multiple is visible, and 45 times revenue is a frontier-software number for an infrastructure company. Humyn looks cheapest, raising less than a year of revenue, but a fifth of that is debt to service. Boltz is unscoreable: no valuation, no revenue figure, only a steep usage curve.

Buyer quality. Palo Alto pays in equity, which keeps the seller aligned with the stock and leaves its balance sheet intact. Bessemer and a returning syndicate of Andreessen Horowitz, Amplify, Zetta and Obvious are pricing a science bet with real logos and no disclosed revenue. Accel is underwriting a services business with software-margin ambitions, the least fashionable story of the three.

Defensibility. Applied Compute's asset is a research team and a platform, in a category crowded by Nvidia's partner network. Boltz has the cleverer structure, free open weights as a distribution funnel and paid proprietary models on top, but rivals can download those weights too. Humyn has the most physical moat, since capture rigs and site relationships take years to replicate, and the lowest ceiling, because data vending is a contract business.

Timing risk. Palo Alto's deal does not close until fiscal Q3 2027 and needs regulatory clearance, so a year of integration risk sits between announcement and revenue. Boltz is racing a market that just got pricier: Chai Discovery raised $400 million at a $3.8 billion valuation in July, and Isomorphic Labs closed a $5 billion Series C on November 17. Humyn depends on robot makers continuing to spend.

The Verdict

Applied Compute wins on consequence. It is the clearest signal that security budgets and AI infrastructure budgets are merging, and a 45x multiple on 18 months of revenue tells every founder building model tooling that the strategic exit door is open.

Boltz wins on interest. It built a research audience first with open models that reached every top-20 pharma before the seed round, then converted that into usage-based revenue. Its $110 million is small next to its rivals, which is why the next 12 months matter.

Humyn wins on efficiency. No other company here turned real-world labor into a $50 million run rate in six months, and none sells the same input that Tesla, Figure, Agility, Agibot, UBTech and Unitree are bidding for.

None of the three sells a chatbot. All three sell inputs: compute for post-training, models for molecules, data for robots. If you want to know where AI money goes next, watch what these three do with their first year of capital.

  • Watch whether Palo Alto keeps Applied Compute as a platform or folds it into its security products.
  • Watch Boltz's first disclosed valuation, and whether its paid models stay ahead of its free ones.
  • Watch Humyn's customer concentration as it opens offices in San Francisco, London and Seoul.

Comments

No comments yet. Be the first to share your thoughts!