9 min read
TLDR
Tap for the short version
- Rezolve Ai says it will scale through acquisitions, cloud relationships and systems integrators instead of adding customers one salesperson at a time.
- The approach is spreading across enterprise AI because partners already hold the contracts, trust and delivery teams needed to reach large companies.
- Rezolve Ai reported $130.8 million in H1 2026 revenue, up from $6.3 million a year earlier, while gross margin fell from 95.2% to 48.9%. The comparison now covers a much larger business after acquisitions and does not measure organic growth alone.
- Partnership announcements describe enormous reach but rarely disclose revenue contribution, customer retention, commercial terms or who verified the product claims.
Contents
Rezolve stops selling alone
On September 1, Rezolve Ai’s chief executive told investors the company is “not building this one customer and one salesperson at a time.” The results help explain the shift. Rezolve reported $130.8 million in first-half revenue, up from $6.3 million a year earlier, while gross margin fell from 95.2% to 48.9%.
The comparison now covers a much larger business after acquisitions and does not measure organic growth alone. Partnerships cannot explain the margin decline from the published figures either. By midyear, Rezolve’s revenue included an acquired loyalty platform, a large services operation and products sold with companies already working inside large enterprises.
The results raise a wider question. Enterprise AI companies can reach customers through cloud marketplaces, consultancies, distributors and systems integrators. These partners may already hold contracts, security approvals and relationships inside large accounts. The AI company gains access while giving up some control over the economics and customer relationship.
Rezolve used several routes at once. In February it paid $230 million in cash for Reward Loyalty, a UK platform with relationships across banks, retailers and payment networks. Rezolve said the deal would add about $90 million of revenue and expand its access to tens of millions of cardholders. The Reward announcement describes an acquisition of revenue, customer access and operating capability. It is more concrete than the partnerships that followed because the price and expected contribution were disclosed.
Google’s announcement needs separating from the sales story. The company selected Rezolve’s distributed database for infrastructure supporting Google Cloud Web3. The Google deployment covers about 100 terabytes across ten blockchain networks. It gives Rezolve a substantial reference deployment, though the disclosed work does not distribute its commerce products.
Tech Mahindra offers the more direct route to enterprise demand. Its alliance can take Rezolve’s commerce products into a business serving more than 1,100 clients with about 146,000 professionals across 90 countries. Rezolve also cites Microsoft and Tata Consultancy Services. I could not find disclosed revenue shares, conversion figures, exclusivity terms or margin contributions for those relationships.
Put side by side, the announcements tell three different stories. Reward brought an operating business and customers inside Rezolve. Google adopted infrastructure for a defined use. Tech Mahindra can introduce, integrate and deliver Rezolve products to its clients. Each removes a different obstacle on the way to an enterprise customer.
Enterprise AI finds the channel
Selling AI to a large company takes more than proving the model works. The buyer may need security approval, data integration, procurement review, staff training and someone willing to own the deployment. A new AI company must earn each permission. A cloud provider or integrator may already have them.
Rezolve is not alone in using this route. OpenAI launched a formal partner network in June with $150 million in support and a target of 300,000 certified consultants by the end of 2026. C3 AI promotes co-selling through AWS, Microsoft and Google Cloud. Its partner programme lets other companies build and deliver applications on its platform. Claude Enterprise is available through AWS Marketplace, letting customers buy through an existing cloud account. Anthropic still sells directly, while the marketplace removes procurement friction for AWS customers. In each example, the partner removes a specific procurement or delivery step.
Growth changes the revenue mix
Rezolve’s figures show how sharply the business mix changed. Gross profit was $63.9 million on $130.8 million of H1 revenue, producing the 48.9% margin. Net loss reached $139.5 million, compared with $101.4 million across all of 2025. The company also raised about $250 million in equity during the half.
The revenue mix changed as the company grew. Rezolve said its customer count rose from about 950 at the end of 2025 to more than 1,640 by midyear. It also operates a services team of roughly 700 people, mainly in India, handling data preparation and catalogue enrichment. That work carries more labour than a software licence and is likely to produce a lower margin.
I could not separate the effects of Reward, services work and partner-led sales from the public numbers. Rezolve’s finance chief said second-half guidance assumes no further acquisitions and is “purely organic.” The company did not disclose how much of its $360 million full-year target depends on Google, Tech Mahindra, Microsoft or TCS.
Product claims enter the channel
These channels carry the product and the claim used to sell it. Rezolve presents its commerce AI as more reliable than a general model, and part of that case rests on published research.
A four-page ACM paper presented at UMAP 2026 tested whether chatbots could report three checkout states correctly. The researchers used 90 sessions from two merchants and four third-party foundation models. Instead of asking the model to infer a customer’s status from conversation history, they supplied the computed state from the transaction log. Accuracy increased from lows of 86% to between 99.5% and 100% across the tested models.
The paper supports a narrow conclusion. Verified transaction state helped the models report checkout status more accurately. It did not compare Rezolve with another commercial commerce product or establish a general cure for hallucinations.
The accompanying user study involved 42 people recruited through the organisation and personal networks. In one scenario, 71% accepted a response containing a subtle checkout error, while 29% accepted the verified version. The authors reported the unexpected result and suggested that presentation can override accuracy when an error is easy to miss.
The gap appears in Rezolve’s wider marketing. A company release connected the checkout research to a claimed “26% AI distortion crisis” using figures from separate research. Its self-published brainpowa paper says the model beats GPT-4, Claude and several open models, although the comparisons rely on an AI judge rather than verified ground truth. The reported 52.4% win rate against GPT-4 sits slightly above an even split. Only the checkout paper has peer review.
A direct customer can ask the product maker to explain a benchmark, provide test data or run a pilot. A customer buying through an integrator may put those questions to the company managing the deployment. That company understands the client’s systems but may have limited visibility into how the underlying model was evaluated.
The partner model complicates that exchange. An integrator can reduce procurement and technical risk without testing every accuracy claim. It also places another layer between the product team and its users. When the product fails, market reach says nothing about who understood its limits before deployment.
Europe divides the roles
The EU AI Act assigns duties according to the system, its risk classification and the role each company performs. A provider, deployer, importer and distributor do not carry identical obligations. A systems integrator may also become a provider in some circumstances if it substantially modifies a high-risk system or puts it on the market under its own name.
For a European buyer, a familiar partner does not settle the underlying questions. A retailer or bank still needs to identify the system, the evidence supporting it and the responsibilities written into the arrangement. The legal duties will vary, and procurement through a large integrator does not turn a product claim into independent evidence.
I went looking for evidence that could measure Rezolve’s partner channel and found little. Its October 6 Investor Day could add early revenue contribution, commercial terms and margin figures for Google and Tech Mahindra. Across the market, useful evidence would show signed customers, retained revenue, repeat deployments and margin after revenue sharing. It would also show who supports the product, verifies its claims and carries user feedback back to the product team. Most partnership announcements disclose none of this.
Partnership networks are becoming a serious route into enterprise accounts, though the public evidence does not establish them as the default. Rezolve’s next disclosures should show whether its channel is producing durable growth or moving the difficult part of the sale somewhere less visible.
Sources behind the analysis
This analysis uses Rezolve Ai’s H1 results, transaction and partnership announcements, the full ACM paper, Rezolve’s self-published model paper and public material from the other companies discussed. Financial figures attributed to Rezolve come from its disclosures. Industry Contents has not independently audited them.
FAQ
Rezolve reported that gross margin fell from 95.2% to 48.9% as its revenue mix changed. The business added acquired revenue and operates a roughly 700-person services team doing data and catalogue work. Its disclosures do not isolate the effect of each factor or establish that partnerships caused the decline.
Rezolve said its second-half guidance excludes further acquisitions and is organic. It has not disclosed how much of the $360 million full-year target depends on Google, Tech Mahindra, Microsoft or TCS.
A narrow checkout-state result was peer reviewed and presented at ACM UMAP 2026. Broader claims about the brainpowa model come from a separate self-published paper without peer review.
No. OpenAI, C3 AI, Anthropic and Vida also use marketplaces, consultancies, resellers or systems integrators to reach enterprise customers. Public evidence does not establish partner distribution as the dominant route across the market.
Related reading
- Nordic SaaS, on how AI changes who captures value inside an existing platform relationship, the same underlying question in a different sector.
- Agentic business risk, on what happens when an AI system embedded in a business relationship starts working against the company that deployed it.
Spotted an error, have a correction, or want to pitch a story? Contact [email protected].
Spotted an error in this piece? We correct publicly, tell us via the contact page. Read our corrections policy.
