MACHINA

What 2026 capital is rewarding

Three separate funding waves are cresting at once, and Machina sits in the overlap. That overlap is the asset. Telling all three stories simultaneously is the mistake.

Physical AI is the capital magnet

Robotics startups raised roughly $18.8B in the first half of 2026, already past the $15B of all of 2025 and the $14.1B of the 2021 peak. Figure re-rated past $39B, Apptronik to about $5.5B, and Germany's Neura to roughly $7B.

Crypto capital is already inside robotics

Neura's approximately $1.2B Series C was led by Tether, alongside Nvidia, Amazon and Qualcomm. This matters more than it looks: the crypto allocator and the robotics allocator are no longer separate audiences with separate pitches. They are increasingly the same cheque.

Prediction markets re-rated roughly 4x in a year

Kalshi went from $11B to $22B and is reportedly seeking $40B. Polymarket is seeking $15B, with ICE, the owner of the NYSE, committing $2B. Sector volume hit about $25.7B in March 2026 alone.

Robot training data became a funded category with no incumbent

XDOF launched with $70M from Thrive, Spark, a16z, Lux and WndrCo purely to collect robot teleoperation data. Vision Lab raised a seed to pipe factory data to frontier labs. OpenAI restarted its robotics programme and Meta bought a humanoid team into Superintelligence Labs. Demand is funded and supply is scarce.

The data buyers have standing budgets

The top twenty hedge funds each spend roughly $40M to $60M a year on alternative data, 94% of buy-side firms plan to increase that spend, and more than 55% of it is shifting toward proprietary, custom-negotiated licences rather than commodity feeds.

Pick the comparison you win

Every pitch is implicitly benchmarked. The single highest-leverage decision in this raise is which set of companies an investor mentally files you next to.

Lose this one

Kalshi at a reported $40B and Polymarket at $15B. Framed as an exchange, Machina is a seed-stage competitor to two companies with billions in committed capital and an IPO on deck.

Win this one

Sportradar at roughly $1.48B of 2025 revenue, growing 17%, and Genius Sports heading to about $1.1B in 2026. A two-company duopoly built entirely on official data rights, with no equivalent in robot combat.

Win this one too

XDOF at $70M raised pre-scale for robot data collection. Machina produces a data type that teleoperation fundamentally cannot generate.

Where the current thesis needs to move

The underlying strategy is sound. Three specific positions should change before it meets investors.

1. Lead with data, not the exchange Trading fees are the visible business but the weakest fundraising story in this market, because the comparison set is brutal and well capitalised. Data licensing survives every regulatory regime, carries software margins, and has a proven billion-dollar template. Put it first in every conversation and let the exchange be the mechanism that produces the data.
2. Do not run an oracle, be the source oracles read Chainlink holds roughly 70% of the oracle market, is actively marketing itself as "The Prediction Market Oracle," and already provides settlement for Polymarket alongside Pyth. Building a competing oracle is a fight against an entrenched network. Publishing Machina as a first-party data source through Chainlink and Pyth achieves identical lock-in, converts the incumbent into a distribution channel, and costs a fraction of the effort. Nobody rewires their oracle, but nobody has to when you are the data inside it.
3. Promote the AI-lab buyer from third to first This is the only part of the thesis where 2026 capital is visibly chasing supply that does not exist. Factory and teleoperation datasets capture robots succeeding at routine tasks. Competitive combat produces instrumented, labelled data on robots operating past their limits until something breaks: joint failure, thermal shutdown, balance recovery, structural fatigue under adversarial load. That is a category of one, and it cannot be synthesised cheaply.

Why the window is roughly five months

This is the part of the strategy with an expiry date, and it is the strongest argument for moving now rather than after a raise.

Five moves, in order

Each phase creates the precondition for the next. Skipping ahead to monetisation before owning citation is the common failure mode for data businesses.

1. Secure the rights before the final Non-exclusive telemetry, results and sublicensing rights from all three leagues, signed inside the current season. Non-exclusive is the concession that makes this easy to say yes to, and it costs nothing, because the moat is being first and being wired in, not being sole.
2. Become the citation layer Give away schedules, results, records and rosters with attribution required. Within months the objective is that every article, every stream graphic and every AI assistant answering a robot combat question cites Machina. This is distribution disguised as generosity, and it is what makes the later licences defensible.
3. Wire into settlement through the existing oracles Ship a free, well-documented API with self-serve keys, then get the feeds published via Chainlink and Pyth. Crypto builders integrate in days and never migrate. Every onchain robot combat market then settles on Machina data by default, without ever fighting an oracle network for that position.
4. Sell the three premium products Verified low-latency settlement and live in-play state to books, prediction markets and broadcasters. Adversarial failure datasets to AI labs. Hardware reliability signal to funds. Three different buyers, one collection pipeline.
5. Pay the leagues from day one Data revenue share plus advertising revenue share on hosted streams. This converts a rights agreement into a partnership that renews, and makes the anchor league an active defender of the arrangement.

One asset, three pitches

The same company should be introduced differently to each pool of capital. These are not different strategies, they are different first sentences and different proof points.

Tier-one venture

Lead with

The data layer for physical AI. A proprietary source of adversarial robot failure data that no factory dataset can produce.

Proof points

XDOF's $70M launch, OpenAI restarting robotics, Meta acquiring a humanoid team, $18.8B into robotics in six months.

Framing

Category creation with a live, dated wedge event, not a bet on whether robot combat becomes popular.

Avoid

Any mention of a token in the first meeting. It reframes the company from infrastructure to crypto and narrows the room.

Hedge funds and alternative data

Lead with

A proprietary signal on humanoid hardware reliability, measured under adversarial stress, ahead of and through the first public robotics listings.

The specific product

Unitree hardware powers the US leagues and Unitree is about to be listed. Component failure rates by manufacturer, measured in combat, are fundamental data on a public company that cannot be bought anywhere else.

Proof points

$40M to $60M annual alt-data budgets at the top twenty funds, 94% increasing spend, and the shift of most spend toward proprietary custom licences.

Framing

Sell the subscription first and the equity second. A paying data customer is the most credible investor you can have.

Crypto funds and treasuries

Lead with

The first-party settlement source for an entirely new asset class, distributed through the oracle networks they already use.

Proof points

Prediction market volume at roughly $25.7B a month, Kalshi and Polymarket re-rating, and Tether leading a $1.2B round into humanoid robotics.

Framing

Infrastructure that other protocols depend on, with the token as a later alignment mechanism and never as the reason to invest.

Sequence the money

The order of these conversations determines the terms of all of them.

• Equity first, on the physical-AI data narrative. This is the highest-multiple story available and the one with the most active buyers.
• Land a paying data customer before or during the round. One AI lab or one fund on a signed licence changes the conversation from thesis to traction.
• Take crypto capital as robotics capital. Tether into Neura is the precedent. Approach crypto treasuries with the infrastructure story, not the token story.
• Token last, if at all. Introducing it early costs the institutional investor and gains nothing, because the crypto investor will still be available later. Alignment and fee discounts only, never performance-linked rewards.

Monthly settled volume, plus four leading indicators

Settled volume stays the north star, but it is a lagging measure. In the first year these four predict it.

Rights secured

Leagues under contract with telemetry and sublicensing, out of the three that exist.

Citation share

Share of articles, broadcast graphics and AI assistant answers about robot combat that cite Machina.

Oracle integrations

Feeds live through Chainlink and Pyth, and markets settling on them.

Data ARR

Contracted annual revenue from labs, funds, books and broadcasters. The number that survives regulation.

What would break this

• 01The hardware makers keep their own data. EngineAI runs URKL and builds the T800; Unitree powers the US leagues. Both have obvious reasons to retain telemetry. Mitigation is revenue share plus free analytics back to them, and moving before anyone frames data as strategic.
• 02An incumbent notices early. Sportradar and Genius already do exactly this for conventional sport, and Chainlink is expanding into prediction market settlement. The only defence is speed and signed relationships.
• 03The sport does not sustain. Audiences may not persist beyond novelty. This is why the AI-lab and hedge-fund data products matter, because they pay regardless of viewership and depend only on fights happening.
• 04Regulatory pressure on markets. Keep data licensing structurally separate and load-bearing, so that no jurisdiction's treatment of prediction markets can take the company down with it.