Touchmark AI

Market and management platform for reserving future AI inference capacity with service-level agreements.

Website: https://touchmark.ai/

Cover Block

Public sources

Field Value
Name Touchmark AI
Tagline Market and management platform for reserving future AI inference capacity with service-level agreements [Touchmark, November 2026]
Headquarters San Francisco, US [FundUp.ai, November 2026]
Founded 2026 [NUS Computing, November 2026]
Stage Seed [Touchmark, November 2026]
Business Model Marketplace
Industry Deeptech
Technology AI / Machine Learning
Geography North America
Growth Profile Venture Scale
Founding Team Co-Founders (2), Roman Yanushevskyi and Ilia Bolgov [FundUp.ai, November 2026]
Funding Label Seed
Total Disclosed ~$3,500,000 [FundUp.ai, November 2026] [Touchmark, November 2026]

Links

Public sources

Executive Summary

PUBLIC Touchmark AI is building a marketplace for reserving future AI inference capacity under standardized contracts and service-level terms, a proposition that merits attention because rising inference demand is starting to look less like ad hoc cloud spend and more like capacity planning [Touchmark, November 2026] [Y Combinator] [Dealroom.co]. The company was founded in February 2026 by Roman Yanushevskyi and Ilia Bolgov, according to an NUS Computing profile and startup funding coverage, and it entered Y Combinator's Summer 2026 batch before opening to commercial customers in August 2026 [NUS Computing, November 2026] [FundUp.ai, November 2026].

The product, as described by the company and corroborated in part by NUS Computing, lets customers pre-purchase inference capacity by model, token volume, provider, and delivery window, then draw that capacity later through an API; the claimed differentiation is a neutral market structure with the ability to buy, sell, and rebalance commitments rather than rely solely on fixed reservations [Touchmark, November 2026] [NUS Computing, November 2026] [Touchmark]. That is a sharper idea than a generic infrastructure abstraction layer, but most operating claims remain company-supplied, including reported requests for quotes covering more than 4 trillion tokens and more than 30 providers listing over $30 million in contracts on the platform [Touchmark, November 2026].

The founder-market fit is plausible on paper. Yanushevskyi is described as an NUS Computing alumnus with prior quantitative research experience at Citadel Securities and engineering experience at Lovable, while Bolgov is identified as a former Revolut product employee and Imperial College London alumnus [NUS Computing, November 2026] [FundUp.ai, November 2026] [LinkedIn].

On capitalization, Touchmark disclosed a $3 million seed round in November 2026 after reported accelerator financing of $500,000 tied to Y Combinator participation, bringing disclosed funding to about $3.5 million [Touchmark, November 2026] [FundUp.ai, November 2026]. The business model is best read as a marketplace around forward procurement of inference capacity, and the next 12 to 18 months will likely test whether the company can convert early supply listings and quote volume into repeat demand, contract settlement credibility, and evidence that discounts and liquidity hold up outside initial launch conditions [Touchmark, November 2026] [Y Combinator] [NUS Computing, November 2026].

Company-stated, unverified -- This section relies materially on company disclosures, with partial corroboration from NUS Computing, FundUp.ai, and Y Combinator.

Taxonomy Snapshot

Axis Value
Stage Seed
Business Model Marketplace
Industry / Vertical Deeptech
Technology Type AI / Machine Learning
Geography North America
Growth Profile Venture Scale
Founding Team Co-Founders (2)
Funding Seed, total disclosed about $3.5 million

How the Company Got Here

PUBLIC Touchmark AI entered the market in 2026 with a narrow but intelligible premise: let companies reserve future AI inference capacity instead of buying entirely on demand [Touchmark, November 2026]. The company is based in San Francisco, and its founders are Roman Yanushevskyi and Ilia Bolgov [Y Combinator]. On its own website, Touchmark describes itself as a market for AI model inference, while its November 2026 announcement frames the product more specifically as a market and management platform for reserving future inference capacity with service-level agreements [Touchmark] [Touchmark, November 2026].

The public timeline is short, but it is unusually compressed. NUS Computing reported that the founders started the company together in February 2026, Touchmark opened to commercial customers in August 2026, and the company presented at Y Combinator Demo Day on September 10, 2026 [NUS Computing, November 2026]. By November 2026, Touchmark had announced a $3 million seed round, following the standard $500,000 Y Combinator financing associated with its Summer 2026 batch participation, according to FundUp.ai and the company announcement [FundUp.ai, November 2026] [Touchmark, November 2026].

What is missing matters almost as much as what is present. No legal entity name appears in the cited company website or Y Combinator profile, and the available public record does not establish customer names, valuation, or a longer operating history beyond 2026 [Touchmark] [Y Combinator]. That leaves the current picture straightforward: a newly formed marketplace startup, early in commercialization, with public momentum concentrated in product positioning and seed financing rather than in independently verified operating scale [Touchmark, November 2026].

Lightly corroborated -- This section relies primarily on the company website and company announcement, with partial corroboration from Y Combinator and NUS Computing.

Product and Technology

MIXED

The product claim is straightforward even if the market framing is unusual: Touchmark says it operates a marketplace where AI application companies can reserve future model inference capacity under service-level agreements, then draw that capacity down later through an API [Touchmark, November 2026] [Y Combinator] [NUS Computing, November 2026]. Public descriptions are consistent on the core workflow. Buyers specify a model, token volume, provider, and delivery window, while the platform standardizes those commitments as contracts rather than one-off reservations [Touchmark, November 2026] [NUS Computing, November 2026] [Touchmark].

The more differentiated part of the product is the secondary market logic that Touchmark attaches to those commitments. On its website and in its seed announcement, the company describes a shared ledger that lets customers buy, sell, and rebalance reserved inference commitments, and it positions the venue as a neutral market for standardized forward contracts with physical delivery to commercial buyers and sellers [Touchmark, November 2026] [Touchmark]. Y Combinator’s company page adds the commercial claim that customers may be able to lock in pricing at 10 to 30 percent below on-demand rates, although that pricing benefit appears to be company-supplied rather than independently verified [Y Combinator].

Public evidence on the underlying technical stack is thin, so the safer read is product architecture rather than implementation detail. The company says it uses dynamic capacity procurement to match demand with a live, dynamically priced pool of inference capacity, and it reports more than 30 providers listing over $30 million in inference contracts alongside requests for quotes covering more than 4 trillion tokens, but those figures come from the company announcement and should be treated as directional until corroborated independently [Touchmark, November 2026]. There is no verified public demo or engineering documentation in the cited materials that would support a deeper claim about orchestration, routing, settlement infrastructure, or model-serving internals.

Company-stated, unverified -- This section relies primarily on company website and company announcement materials, with partial description-level corroboration from Y Combinator and NUS Computing.

Where the Demand Sits

PUBLIC

This market matters now because the cost and availability of AI inference are becoming budget line items for application companies, while supply remains fragmented across model providers and infrastructure vendors [Touchmark, November 2026] [NUS Computing, November 2026].

The available public record does not support a conventional TAM, SAM, and SOM build for Touchmark’s exact category, which is narrower than general AI infrastructure and newer than most third-party market maps. What can be said from the cited evidence is that the company is positioning itself around a specific purchasing problem: buyers want to secure future inference capacity for a chosen model, token volume, provider, and delivery window, rather than buying only on demand [Touchmark, November 2026] [NUS Computing, November 2026]. That places Touchmark at the intersection of AI infrastructure procurement and marketplace software, with its own framing closer to forward contracting than to standard API resale [Touchmark, retrieved 2026] [Dealroom.co, retrieved 2026].

Public demand signals are still early and mostly company-supplied, but they point to a real budgeting and planning use case. Touchmark reported requests for quotes covering more than 4 trillion tokens, and said more than 30 providers were listing over $30 million in inference contracts on the platform as of its seed announcement [Touchmark, November 2026]. Those figures do not establish revenue or durable liquidity, but they do suggest that both buyers and sellers are testing a market structure for reserving future capacity instead of relying only on spot purchasing [Touchmark, November 2026].

The nearest adjacent markets are easier to define than Touchmark's core market itself. One adjacent market is cloud and GPU capacity procurement, where enterprises already reserve infrastructure in advance to manage price and availability risk. Another is API-based model inference, where costs scale with usage and can become volatile for applications with uneven demand [Touchmark, November 2026]. A third substitute is direct bilateral contracting with model or infrastructure providers, which may remain the default for larger buyers if they can negotiate favorable terms without using a neutral venue [Y Combinator, retrieved 2026].

Macro conditions also appear supportive, at least directionally. The company's own materials and the NUS profile both frame inference as a recurring operational expense rather than a one-time model training cost, which matters because recurring spend tends to create demand for procurement tooling and price certainty [Touchmark, November 2026] [NUS Computing, November 2026]. The regulatory picture is less developed in the available sources. There is no cited evidence yet of sector-specific regulation governing forward contracts for AI inference capacity, so the more immediate external variables are likely commercial ones: model pricing moves, provider reliability, and whether customers are comfortable treating token commitments as a planned inventory input rather than purely variable spend [Touchmark, November 2026] [Touchmark, retrieved 2026].

Market lens What the public evidence supports
Core category Marketplace for future AI inference capacity sold via standardized contracts [Touchmark, retrieved 2026]
Buyer need Reserve tokens in advance by model, volume, provider, and delivery window [NUS Computing, November 2026]
Supplier need List capacity and potentially sell future commitments into a shared market structure [Touchmark, November 2026]
Adjacent market Cloud and GPU capacity reservation, plus API-based model inference procurement [Touchmark, November 2026]
Substitute workflow Direct contracting with providers rather than transacting through a market venue [Y Combinator, retrieved 2026]

The table underscores the main limitation in this section: the category is legible, but not yet well-covered by independent market research. For now, the market case rests more on an observable procurement friction than on a clean third-party sizing dataset.

Company-stated, unverified -- This section relies primarily on company materials, with partial corroboration from NUS Computing and Y Combinator for product framing and market context.

Competitive Landscape

MIXED Touchmark is positioning itself less as another inference endpoint and more as a market layer that sits between AI application buyers and model-capacity suppliers, which places it against a patchwork of substitutes rather than a clean list of direct, named peers [Touchmark, November 2026] [Y Combinator].

The competitive map is easiest to read by function. On one side are inference providers themselves, the suppliers Touchmark says can list contracts on its platform, with customers choosing model, token volume, provider, and delivery window directly [Touchmark, November 2026] [NUS Computing, November 2026]. On another side are procurement workflows inside larger AI application companies, which can still buy capacity bilaterally rather than through a neutral venue, particularly if their demand is predictable enough to negotiate directly with providers. The adjacent substitute is the standard on-demand API model, where buyers pay spot prices and avoid forward commitments entirely, even if that means giving up the discounts of up to 30 percent that Touchmark and third-party profiles say may be available for advance purchases [Y Combinator] [Dealroom.co].

The company’s edge, such as it is in public evidence, appears to come from product framing and founder fit rather than from entrenched distribution. The product is described as a neutral market for standardized inference contracts, with the ability to buy, sell, rebalance commitments, and later consume reserved tokens through an API, which is a meaningfully different workflow from a typical direct provider relationship if it works as advertised [Touchmark] [Touchmark, November 2026]. Roman Yanushevskyi’s reported background in quantitative research at Citadel Securities and engineering at Lovable also fits the market-design thesis better than a conventional infrastructure-sales profile would [NUS Computing, November 2026] [LinkedIn]. That edge is still perishable. Most of the visible differentiation rests on market structure and execution, not on a public dataset, owned hardware footprint, regulatory license, or exclusive channel.

The exposed flank is just as clear. If large inference suppliers or well-capitalized infrastructure platforms decide to package their own reserved-capacity products, they would control both inventory and customer relationship, while Touchmark would still be proving it can be the venue through which those trades clear. Public sourcing does not name a direct startup rival, but the threat need not come from a startup. The more standardized inference purchasing becomes, the easier it is for suppliers to internalize the margin and keep buyers inside their own commercial stack. Touchmark also does not appear, from the available material, to own a proprietary demand channel or a named anchor customer base that would make disintermediation difficult [Touchmark, November 2026] [NUS Computing, November 2026].

The most plausible 18-month scenario is a sorting between market makers and vertically integrated sellers. Winner if standard contracts gain real liquidity: Y Combinator, as the most visible backer in the public record, benefits reputationally if Touchmark turns fragmented procurement into a repeatable category and converts its reported early quote flow into recurring traded volume [Touchmark, November 2026] [FundUp.ai, November 2026]. Loser if suppliers keep capacity sales in-house: Touchmark, because the current public traction signals, more than 4 trillion tokens in quote requests and more than 30 providers listing over $30 million in contracts, are company-reported and do not yet show that the platform owns the transaction layer over time [Touchmark, November 2026]. In other words, the near-term contest is not plainly startup versus startup. It is venue neutrality versus supplier control.

Company-stated, unverified -- The section relies primarily on company and company-profile descriptions, with limited independent corroboration from NUS Computing and FundUp.ai,

Opportunity

PUBLIC

If Touchmark executes, the prize is not simply another AI infrastructure tool, it is a position at the pricing and allocation layer of inference, where buyers and providers increasingly need a market rather than a one-off procurement workflow [Touchmark, November 2026] [NUS Computing, November 2026].

The headline opportunity is straightforward: become the default venue where AI application companies secure future inference supply and where model providers monetize spare or planned capacity through standardized contracts [Touchmark, November 2026] [Y Combinator]. That outcome is still early and company-asserted in important places, but it is not purely aspirational. The product described across the company site, Y Combinator profile, and NUS profile is already framed around concrete transaction primitives, model, token volume, provider, delivery window, and later API consumption, rather than a vague brokerage concept [Touchmark, November 2026] [Y Combinator] [NUS Computing, November 2026]. Commercial availability reportedly began in August 2026, the company entered Y Combinator's Summer 2026 batch, and it announced a $3 million seed in November 2026 after reporting quote requests covering more than 4 trillion tokens and more than 30 providers listing over $30 million in contracts on the platform [NUS Computing, November 2026] [Touchmark, November 2026]. Those operating claims remain company-sourced, but they suggest the initial behavior needed for a marketplace, both demand inquiry and seller participation, may already be present.

The most credible paths to outsized scale depend on whether Touchmark becomes a procurement system, a trading venue, or both.

Scenario What happens Catalyst Why it's plausible
Treasury layer for AI apps AI application companies begin treating inference as a budgeted input that can be reserved in advance, with Touchmark handling forward purchases and later API drawdown Large swings in inference demand and pricing push finance and infrastructure teams toward pre-purchase tools with service-level agreements [Touchmark, November 2026] The product is already described as letting customers reserve future capacity by model, provider, token volume, and delivery window, then consume later through API access [Touchmark, November 2026] [NUS Computing, November 2026]
Neutral exchange for model providers Providers use Touchmark to list future capacity and reach buyers without building bespoke sales and contracting infrastructure More providers seek utilization on planned or underused inference capacity, making a shared venue more efficient than bilateral selling [Touchmark, November 2026] The company says more than 30 providers are already listing over $30 million in contracts, which, if sustained, would indicate early supply-side willingness to standardize around the platform [Touchmark, November 2026]
Embedded market infrastructure Touchmark's contracting and ledger layer becomes embedded into the workflows of infrastructure buyers, brokers, or AI platforms that want pricing certainty without building a market internally Y Combinator distribution and a simple commercial framing, buying AI tokens in advance, make the product easier to slot into adjacent procurement or developer tooling stacks [Y Combinator] [NUS Computing, November 2026] Third-party descriptions from NUS and Dealroom both reduce the product to an intelligible buyer use case, pre-purchasing inference at a discount, which improves odds of adoption beyond sophisticated traders alone [NUS Computing, November 2026] [Dealroom.co]

What compounding looks like here is a classic but difficult marketplace loop. More buyer demand makes it worthwhile for providers to list future capacity; more provider inventory improves price discovery and contract availability; better price discovery gives buyers a reason to route a larger share of spend through the system rather than staying on on-demand pricing or ad hoc bilateral negotiation [Touchmark, November 2026]. If the shared ledger and rebalancing features work as described, Touchmark could also deepen from a booking tool into a system of record for inference commitments, which is where switching costs begin to matter [Touchmark, November 2026]. The earliest sign of that loop, again company-reported, is the combination of quote demand, provider listings, and a product built around standardized contracts rather than custom consulting [Touchmark, November 2026] [Touchmark].

The size of the win is hard to anchor because no credible public market sizing figure or direct public comparable is in the supplied record. Even so, the upside can be framed in structure rather than in unsupported precision. If Touchmark became the recognized market layer for reserved inference capacity, with meaningful transaction volume across many models and providers, it could resemble the economics of an exchange, broker, or infrastructure marketplace rather than a pure software point solution (scenario, not a forecast) [Touchmark, November 2026] [Y Combinator]. That matters because market operators that sit at a critical coordination point often capture durable value through take rates, data, and workflow control, provided liquidity forms on both sides. Here, the company has at least articulated the right wedge, future capacity reservation with standardized contracts and SLA-backed delivery, and shown enough early commercial motion to make the upside legible, even if the evidence base is still thin and heavily company-led [Touchmark, November 2026] [NUS Computing, November 2026].

Company-stated, unverified -- This section relies materially on company-sourced product and traction claims from Touchmark, with partial corroboration from NUS Computing and Y Combinator on company framing, launch timing, and batch participation.

Sources

Public sources

  1. [Touchmark, November 2026] Touchmark raises $3M to solve the AI economy | https://touchmark.ai/blog/touchmark-raises-3m-to-solve-the-ai-economy/

  2. [FundUp.ai, November 2026] Touchmark $500K Seed Funding (2026) | Investors & Contacts | https://fundup.ai/recently-funded-startups/company/298674af39e5c58b424d3ee4c7aab923e8a1e79e059d0b5c6443a4b62ac624fb/touchmark

  3. [NUS Computing, November 2026] With US$3 million in seed funding, NUS Computing alum Roman Yanushevskyi is selling AI tokens the way farmers sell wheat | https://www.comp.nus.edu.sg/news/with-us3-million-in-seed-funding-nus-computing-alum-roman-yanushevskyi-is-selling-ai-tokens-the-way-farmers-sell-wheat/

  4. [Touchmark] Touchmark | Market for AI model inference | https://touchmark.ai/

  5. [LinkedIn] Zuzanna Bochenek - Google | LinkedIn | https://www.linkedin.com/in/zuzanna-bochenek/

  6. [Y Combinator] Touchmark: Market for future inference capacity | https://www.ycombinator.com/companies/touchmark

Articles about Touchmark AI

View on Startuply.vc