Capital Is Concentrating on Applied AI Companies That Can Show Revenue

Revenue first, model second

The clearest signal in the latest funding news is speed to revenue. Applied Compute says it reached a 100 million dollar annualized revenue run rate in less than 18 months and raised 350 million dollars at a 3.25 billion dollar valuation, in a round led by Elad Gil. The company plans to use the funding to expand compute capacity, grow its team, and accelerate enterprise deployment. m:…0908

Usage figures accompany the revenue claim. Applied Compute says its inference service handles more than 1 trillion tokens a day, customers have run more than 1,700 training runs on its AC2 system since July, its team has tripled since January, and it has more than doubled the number of enterprises it serves since its last fundraise. Forty percent of customers use the platform for both training and inference, according to the company, which suggests buyers are consolidating work onto one platform rather than sampling it. m:…0908

Customer examples give the numbers a concrete shape. Clay used Applied Compute to train and deploy a custom router that reduced inference costs by up to 47 percent, according to the company, and Harvey used the platform to train a model for its Review Table product. The pitch is not a better model in isolation. It is cheaper, faster delivery of a model a customer already wants to run. m:…0908

Applied robotics follows the same logic

Tutor Intelligence fits the same pattern in physical work. Andreessen Horowitz led its Series B, and the company sells robotics on an hourly pricing model that shifts spending from upfront capital cost to operating cost. Implementation times run as low as 8 hours, according to the company, which lowers the barrier for smaller manufacturers that could not justify a large robotics purchase. m:…5947

The company plans to have more than 1,000 robots under contract by the end of 2027, and nearly one-third of customers who received robots in the past 12 months were existing customers. Its robots are trained by human teleoperators, and its single-arm robot Cassie handles case picking and palletizing while its two-arm robot Sonny targets e-commerce fulfillment. Repeat buying, not a demonstration, is the evidence investors are being asked to weigh. m:…5947

Bilt acquisition of LAVA for 300 million dollars extends the pattern beyond infrastructure. The deal brings relationships with 56 professional sports franchises into a network of more than 7 million members and more than 50,000 merchant partners, and Bilt plans to combine its AI agents with LAVA real-time engagement technology. The value being bought is distribution and live customer relationships, not a research breakthrough. m:…6654

The premium must be earned in the workflow

An Andreessen Horowitz analysis explains why revenue discipline now decides who keeps a premium. An application is vulnerable, the analysis argues, when a customer can reproduce its core output with a general-purpose model and little ongoing effort. To defend its price, a vendor must outperform the customer alternative across the full workflow, even as model progress continually improves that alternative. m:…7754

The analysis offers measured examples rather than slogans. In a Repsol pilot, Harvey outperformed Microsoft Copilot and Lexia in blind evaluation by the legal team, and Repsol lawyers saved four to six hours per week using Harvey. Decagon reported up to a 34 percent improvement in ticket resolution time for Notion. Those are workflow results a buyer can test, and they are harder for a general model to match without the surrounding product. m:…7754

Pricing follows the same logic. The analysis notes that 94 percent of enterprise AI decision-makers surveyed in August 2026 expected total spending on large language models to increase, with 37 percent expecting it to at least double in the next year. Lower inference costs can improve vendor margins, and one vendor separated data credits from actions to avoid overcharging simple workloads. Capital is therefore flowing to companies that can show a workflow gain, a pricing model tied to outcomes, and usage that grows with customers, because those are the defenses that survive cheaper and better general models. m:…7754