Most photonics pitches sound as if someone discovered light last Tuesday.

Please stop.

Buyers do not pay because photons are elegant. They pay when a product cuts power, moves data faster, reduces heat, fits into a package, shortens a test cycle, or helps an AI system do something useful at a price the buyer can defend.

TL;DR: Photonics startups are startups that use light for compute, data movement, sensing, switching, packaging or chip-to-chip links. In AI compute, the founder opening is not vague "faster AI." It is one paid bottleneck: moving data between chips, cutting power waste, reducing delay, helping co-packaged optics, testing optical engines, proving thermal fit, or serving an edge AI buyer who cannot afford giant GPU infrastructure. Bootstrapped founders should sell one narrow proof before trying to build a full optical compute stack.

I am Violetta Bonenkamp, founder of Mean CEO, CADChain, and F/MS Startup Game. CADChain lives close to engineering files, IP rights, manufacturing data, machine learning and deep tech funding. That makes me deeply allergic to physics-as-marketing.

Photonics is serious.

The problem is that too many founders explain it like theatre.

If you want customers, explain where the buyer loses money without you.

1 · Definition

What Photonics Means In AI Compute

Photonics means using light particles, photons, to transmit, process, switch, sense or measure information.

In AI compute, photonics can show up in several places:

Founder checklist
Founder checks worth seeing together
  • Optical links between racks.
  • Optical links between chips.
  • Co-packaged optics near switch silicon.
  • Silicon photonics chips.
  • Optical circuit switches.
  • Lasers for data movement.
  • Photonic processors.
  • Optical input and output for AI accelerators.
  • Sensing systems for robotics, factories and labs.
  • Test tools for optical and electronic packages.

The important split for founders is this:

There is optical data movement, and there is optical computing.

Optical data movement uses light to move information between systems or chips.

Optical computing tries to use light for actual compute work.

Those are different startup paths, with different buyers, proof needs, costs and timelines.

Do not blur them because the pitch sounds shinier.

The market already knows data movement is painful. The Semiconductor Engineering OFC 2026 article on optical links in AI data centers argues that optical interconnects are moving from telecom into AI clusters as copper links run into power and distance limits.

That is the buyer signal.

Light matters when copper, heat, power or physical distance becomes too expensive.

2 · Market signal

Why AI Compute Needs Photonics

AI systems are hungry for data movement.

Training and inference move data between processors, memory, switches, storage and network gear. Every movement costs power, time, cooling and money.

That creates the photonics opening.

Light can help when the buyer needs:

Founder checklist
Founder checks worth seeing together
  • Faster data movement across distance.
  • Less power per bit moved.
  • Lower heat from electrical signaling.
  • Better links between AI processors.
  • Cleaner rack-to-rack communication.
  • Better package-level data paths.
  • More useful edge AI hardware.
  • Less dependence on the biggest GPU clusters.

The Nature industry article on photonics for AI data centers describes photonic methods such as co-packaged optics, optical circuit switches and silicon photonics across package, rack and network layers. The article uses larger-market language, but founders should read it with one question in mind:

Where does this become a purchase order?

The answer is rarely "replace all electronics."

The answer is usually one smaller job:

  • Replace a painful copper link.
  • Reduce a power bill.
  • Fit optics into a package.
  • Prove a thermal path.
  • Test an optical engine.
  • Connect accelerators with less delay.
  • Build a sensing module for a buyer who already has a budget.

That is where a startup can start.

3 · Risk filter

The Trap: Selling Light Instead Of Buyer Proof

Photonics founders often make one expensive mistake.

They sell the beauty of the science instead of the pain of the buyer.

The buyer does not care that your waveguide is elegant.

The buyer cares whether:

  • The AI system runs cheaper.
  • The device fits the package.
  • The test data passes.
  • The data path is faster.
  • The part can be sourced.
  • The product can survive heat.
  • The supply chain is believable.
  • The customer can buy without betting the company.

This is why advanced packaging startups for AI chips frames the adjacent buyer decision. Photonics often becomes commercially interesting when it meets packaging, heat, testing and data movement. Light has to live inside real hardware.

If the package fails, the photonics story fails.

4 · Europe lens

Europe Has A Photonics Opening

Europe has strong universities, photonics clusters, semiconductor programs, industrial buyers and deep tech capital pressure.

The European Chips Act includes chips, packaging, design and supply resilience in a broader semiconductor plan. The APECS pilot line also matters for founders because co-packaged optics and chiplet work often touch package choices, test paths and partner access.

Then there are funding signals.

EU-Startups reported CamGraPhIC’s EUR211 million round for graphene photonics linked to AI data bottlenecks. EU-Startups also covered Scintil Photonics raising EUR50 million for silicon photonics work tied to AI factories and co-packaged optics. CNBC’s April 2026 report on AI chip rivals also shows that European AI chip companies are fighting for serious capital while US peers remain far ahead.

Do not read those headlines as permission to chase hype.

Read them as proof that buyers and investors are watching the compute stack below the model layer.

The smarter founder move is to ask:

Which photonics wedge can I sell before I need a giant round?

5 · Decision filter

The Photonics Startup Table

Use this before writing an investor memo or grant plan.

Risk map
The Photonics Startup Table
Optical link review
Buyer

AI data center supplier, networking team

First paid proof

Link loss, power and distance memo

Trap to avoid

Talking about light without system cost

Co-packaged optics support
Buyer

Switch or accelerator team

First paid proof

Package fit and heat risk review

Trap to avoid

Ignoring packaging until too late

Silicon photonics test service
Buyer

Photonics chip team, lab, spinout

First paid proof

Test report a buyer can act on

Trap to avoid

Running tests that do not change a decision

Laser source supply map
Buyer

Hardware startup, module maker

First paid proof

Supplier shortlist and lead-time view

Trap to avoid

Becoming a parts list with no buyer action

Edge AI sensing module
Buyer

Factory, robotics, medical device team

First paid proof

Working demo tied to one buyer job

Trap to avoid

Chasing data center glamour

Optical circuit switching tool
Buyer

Data center operator, AI infrastructure team

First paid proof

Traffic and switching study

Trap to avoid

Selling a full system before one path is paid

Thermal and package check
Buyer

Photonics package team

First paid proof

Heat and fit memo

Trap to avoid

Treating optics as separate from heat

CAD and IP file trail
Buyer

Engineering team, supplier network

First paid proof

File access record and rights proof

Trap to avoid

Sharing sensitive design files casually

Public funding pack
Buyer

Photonics founder, university spinout

First paid proof

Grant and buyer proof plan

Trap to avoid

Letting grant work replace sales

Buyer education kit
Buyer

Industrial buyer, public buyer

First paid proof

Plain-language purchasing brief

Trap to avoid

Teaching forever without closing paid work

This is the point:

Do not start with "photonics will change AI."

Start with "this buyer pays to reduce this bottleneck."

6 · Key idea

Startup Ideas That Can Start Small

Here are founder wedges that feel more believable than "we are building a light-based Nvidia."

Optical link decision service

Sell a fixed-scope review for teams deciding whether copper, pluggable optics, co-packaged optics or another link path fits their AI hardware plan.

The output can include:

  • Distance limits.
  • Power per link.
  • Heat risk.
  • Supplier options.
  • Test plan.
  • Purchase risk.

This can begin as service work and later become a tool or data product.

Co-packaged optics readiness review

Co-packaged optics means moving optical engines closer to switch or compute silicon.

That creates packaging, test and heat questions.

The EDN article on co-packaged optics in 2026 frames CPO as a data center architecture path for AI and high-performance computing, while also noting that broad commercial rollout is still a practical challenge.

A founder can sell:

  • Package readiness check.
  • Thermal memo.
  • Optical engine partner shortlist.
  • Reliability test plan.
  • Buyer evidence folder.

Photonics test and measurement kit

Many photonics startups will fail between lab result and buyer trust.

Sell test plans, fixtures, data cleanup and outside-lab coordination for one category:

  • Optical engines.
  • Waveguides.
  • Laser arrays.
  • Transceivers.
  • Package-level links.
  • Sensing modules.

This is not glamorous.

That is why it can be a business.

Industrial sensing with photonics

Photonics is not only for data centers.

Sensors can serve factories, healthcare devices, agriculture, robotics, defense, climate monitoring and lab equipment. For bootstrappers, a sensing wedge may be nearer to revenue than optical compute.

This connects with physical AI for manufacturing and field operations and AI quality inspection in European factories. A factory buyer may not care about photonic elegance, but it may pay for fewer defects, better inspection, safer measurements or faster field checks.

IP and design-file protection for photonics teams

Photonics teams share CAD files, package drawings, mask layouts, supplier notes, test data and process details.

That is sensitive company property.

CADChain is relevant here because engineering files need access trails, rights proof and careful sharing. Photonics founders should not wait until a supplier relationship goes wrong before asking who touched which file.

A narrow startup can sell:

  • File rights review.
  • Supplier access map.
  • Design version record.
  • Evidence folder for buyer due diligence.
  • Secure sharing workflow.

Yes, it sounds boring.

So does losing your design file.

7 · Key idea

The 30-Day Photonics Founder Test

Use this before applying for a grant, hiring a hardware team, or naming your company after a beam of light.

Day 1 to 3:

  • Pick one buyer group.
  • Pick one bottleneck.
  • Write the buyer cost in plain language.
  • Decide whether the first product is service, data, test, software or hardware.

Day 4 to 7:

  • Talk to 10 buyers, suppliers, labs or chip teams.
  • Ask what delays them.
  • Ask what they already paid for.
  • Ask what proof would change a purchase decision.
  • Ask what they distrust about photonics vendors.

Day 8 to 14:

  • Build a paid mini-offer.
  • Keep it narrow.
  • Promise one useful output: a link review, heat memo, supplier map, test plan, IP record or buyer brief.

Day 15 to 21:

  • Sell it.
  • Charge real money.
  • If nobody pays, change the buyer or the bottleneck before touching hardware spend.

Day 22 to 30:

  • Deliver.
  • Write down what repeated.
  • Remove anything that felt custom but did not affect the buyer decision.
  • Turn the repeated part into your first product method.

That is how bootstrappers enter deep tech without pretending invoices are optional.

8 · Key idea

What To Measure Before You Sell

Photonics founders need proof that non-technical buyers can understand.

Useful measures include:

  • Power per data link.
  • Heat created at the package.
  • Distance before the signal weakens.
  • Error rate.
  • Test time.
  • Supplier lead time.
  • Package fit.
  • Cost per module.
  • Buyer payback period.
  • Field reliability.

Do not bury the buyer in a lab report.

Give them a decision.

The right output is not "our photonic platform is amazing."

The right output is closer to:

"This optical link path can cut link power, but it adds package and supplier risk. Here is the test we need before purchase."

That sentence helps a buyer.

That sentence can sell.

9 · Capital lens

Funding Without Grant Addiction

Photonics often needs public money because the science is hard, the testing is expensive, and the path from lab to buyer can be long.

Public money can help.

It can also make founders write beautiful documents while the market stays silent.

The F/MS deep tech guide for women founders is useful here because deep tech needs technical confidence, peer support and funding discipline. The F/MS Startup Game exists because first-time founders learn better by doing than by admiring templates.

For photonics founders, the rule is simple:

Use grants to buy proof.

Do not use grants to avoid sales.

The companion article on public-private funding for European deep tech goes deeper on this. The short version is that public money should move a photonics company toward buyers, partners and test evidence. If it moves you toward more paperwork and less selling, the grant is becoming a comfortable trap.

10 · Buyer lens

The Buyer Script

If you are selling photonics into AI compute, try a buyer script like this:

"We help AI hardware teams reduce one data movement bottleneck. We compare the electrical path against an optical path, test power and heat assumptions, and deliver a purchase memo your engineering and finance teams can both understand."

That is plain.

That is paid-problem language.

Now compare it with this:

"We are building the light-based compute layer for the coming era of AI."

The second one may sound grand.

The first one can get a meeting.

11 · Red flags

Mistakes Photonics Startups Should Avoid

The first mistake is selling magic speed.

Speed alone is not enough. The buyer needs cost, heat, package fit, supplier access and proof.

The second mistake is confusing data movement with optical compute.

One may be nearer to purchase. The other may need more capital, testing and patience. Name the path honestly.

The third mistake is ignoring packaging.

Optics do not float above hardware. They need to fit with chips, memory, boards, heat, test plans and supply chains.

The fourth mistake is chasing only hyperscalers.

Large data center buyers get attention, but smaller industrial, edge AI, sensing and lab buyers may give a founder faster proof.

The fifth mistake is hiding behind grant language.

If the buyer would not pay for a small proof, the grant may be covering a demand problem.

12 · Decision filter

Where Photonics Fits In The AI Compute Stack

Photonics connects to several founder decisions:

That cluster matters because AI compute is not one thing.

It is chips, memory, optics, package work, software, power, cooling, supplier access, data rights and customer economics.

Photonics becomes a startup opportunity when it solves one part of that stack with buyer proof.

13 · Verdict

The Bottom Line

Photonics startups should stop pitching light as destiny.

Sell the bottleneck.

If light helps move data faster, prove the power and heat case.

If optical compute helps one workload, prove the buyer job.

If sensing is the nearer business, stop pretending the data center is the only prize.

Europe has enough science, enough industrial pain and enough policy pressure to make photonics worth serious founder attention. But deep tech does not forgive vague thinking.

Pick one buyer.

Pick one bottleneck.

Sell one proof.

Then earn the right to build the bigger machine.

14 · Reader questions

FAQ

What are photonics startups?

Photonics startups are companies that use light to move, process, switch, sense or measure information. In AI compute, that can mean optical links, silicon photonics chips, co-packaged optics, optical circuit switches, photonic processors, laser sources, test tools or sensing modules. The best startup path depends on the buyer and the bottleneck, not on the prettiest physics.

Why does AI compute need photonics?

AI compute moves huge amounts of data between processors, memory, switches and racks. Electrical links can become costly when distance, heat and power rise. Photonics can help move data with less power waste and lower delay across certain paths. The startup opening is proving where the optical path beats the electrical path for one buyer, not claiming light replaces every part of the stack.

Is optical computing the same as optical interconnect?

No. Optical interconnect uses light to move information between parts of a system. Optical computing tries to use light to perform compute work. Optical interconnect may be nearer to commercial use in AI data centers because data movement is already painful. Optical computing can still be powerful, but it may demand more capital, proof and patience.

Can a bootstrapped founder build a photonics startup?

Yes, if the first product is narrow. A bootstrapper should not begin by building a full optical compute system. Better first offers include optical link reviews, thermal checks, test plans, supplier maps, buyer education briefs, sensing modules, IP file trails or pilot-line readiness support. Start with paid proof, then decide whether the company should move toward software, services, test tools or hardware.

Who buys photonics products for AI compute?

Possible buyers include AI data center suppliers, switch vendors, accelerator teams, semiconductor startups, labs, photonics spinouts, edge AI companies, robotics teams, industrial inspection firms, medical device companies and packaging partners. The best buyer is not always the largest one. It is the one that already feels pain from data movement, heat, distance, sensing limits or test delays.

How should photonics startups price early work?

Price around buyer decisions. A short optical link review might be EUR2,000 to EUR5,000. A package and heat memo might be EUR5,000 to EUR15,000. A test coordination project can be higher, with outside lab costs separated. These numbers are not universal. They are a reminder that deep tech work should not be priced like cheap content production.

What proof do photonics startups need before fundraising?

They need proof that connects physics to a buyer decision. That may include measured power per link, heat data, error rates, supplier interest, package fit, test results, letters from buyers, paid pilots, or a clear purchase memo. Investor excitement is useful, but customer evidence is cleaner. A round should help a real business grow, not cover confusion.

How do photonics startups connect with advanced packaging?

Photonics often touches packaging because optical engines, lasers, chips and heat paths must fit close to electronic systems. Co-packaged optics is one clear example. Founders should understand package fit, thermal behavior, test plans and supplier access early. If the package cannot work, the photonics promise will not matter.

Are photonics startups relevant outside data centers?

Yes. Photonics can support sensing, inspection, robotics, medical devices, agriculture, labs, defense and climate monitoring. For many bootstrapped founders, these markets may be easier to enter than hyperscale data centers. A factory buyer may pay for a sensing module that reduces defects long before a data center buyer trusts a new optical compute path.

What should a founder do this week to test a photonics idea?

Pick one buyer and one bottleneck. Talk to 10 people close to that bottleneck. Ask what they have tried, what failed, what they paid for, and what proof would make them act. Then sell a small paid output, such as a link review, test plan, heat memo, supplier shortlist or IP file map. If nobody pays, adjust before spending on hardware.