Pitching a semiconductor startup to accelerators and early-stage investors

How evaluators judge chip, equipment, metrology, and semiconductor software startups, and what founders should prepare on proof, data access, fab adoption, business model, and team.

Larry Chao, PhD, Sovereign Pivot · Last updated October 7, 2026

Semiconductor startups are judged by a different standard from most of the companies they pitch alongside. Evaluators accept long timelines and heavy capital needs, and many investors actively want more deep tech in their portfolios. In return they expect physical proof, a credible way into conservative customers, and a team that has done the hard part before.

This guide is for founders building chips, equipment, metrology, process and yield software, EDA tools, advanced packaging, photonics, or quantum hardware. It covers how these pitches are evaluated and what tends to sink them, then sets out what to prepare. It draws on patterns observed across early-stage selection interviews and builds on the general pitch preparation guide.

How chip pitches are judged differently

A domain expert usually holds the deciding view. Most selection groups include only one or two people with fab, equipment, or chip-design experience. Generalists in the room defer to them openly, so a single expert's doubt can outweigh enthusiasm from everyone else. Pitch for that person's questions while keeping the story clear for the generalists.

Early enthusiasm fades under deeper diligence. A big technical idea often clears a first conversation on plausibility alone. Later meetings ask whether there is a credible path to a paying customer within the next six to twelve months. Companies that advance easily at first can stall at that second test, so prepare for the later question from the start.

Software in the semiconductor stack is held to the software standard. Process AI, yield analytics, and EDA tools face the same "why won't an incumbent build this" question as any software company. Here the incumbents include equipment makers and EDA vendors, along with the chip companies' own internal teams.

Talent bets need talent for the hard part. Evaluators will back a strong team ahead of revenue when that team can plausibly do the specific hard thing. In this sector that usually means getting a tool qualified, closing a fab partnership, or taping out.

Proof: what counts

The most common reason a chip pitch stalls is that the evidence does not reach the real world.

Measured results on real equipment count most. Data from an actual tool, actual wafers, or a fabricated device carries the most weight. A clear before-and-after on a meaningful metric does more than any slide of projected benefits, whether the metric is resolution and yield or power and throughput.
Simulation reads as illustration. A simulated result, a paper design, or a model trained only on synthetic data is received as a statement of intent. Label it honestly, and pair it with whatever physical result you have, however small.
Correlation with production is the bar for process software. If you model a process step, expect to be asked whether your model has been checked against a real tool. "No correlation with an actual tool" is treated as no path to production.
University and national-lab facilities are a credible first step. If you cannot get onto a production line yet, validation on a shared fabrication facility or research tool is accepted as a real milestone. Name the facility and the result.
Each generation of the product needs its own evidence. When the competitive claim rests on a future generation that has not been built, evaluators compare what you can build today with the incumbent, and that comparison is usually unflattering. Show the buildable version honestly.

The data question

For any company whose product depends on fab or customer process data, the first hard question concerns the source of that data. Evaluators with fab experience know that leading manufacturers treat process data as their most protected asset, and they rarely share it with a startup.

Prepare a specific answer covering:

The source. Name which customer or partner will provide data, under what agreement, and in what form.
The minimum viable data. State how little data the product needs to show value, and whether it can work inside the customer's environment without the data leaving it.
The bootstrapping path. Explain how the first data set is obtained before you have a reference customer. Partner facilities, smaller fabs, and design partnerships all work.

A pitch that assumes access to a top-tier foundry's data, or that leads with the leading edge, loses the room quickly. The more credible path starts where the data is easier to reach.

Getting into the fab

Evaluators who have worked in manufacturing will test whether you understand how conservative the customer is.

Every process step has neighbors with a veto. A new tool or method affects the steps immediately before and after it, and the engineers who own those steps can block adoption. Show that you know who they are and how you will win them over.
Engineers fear the yield crash. A process engineer who changes a recipe on a startup's advice and causes a yield excursion carries the consequences personally. Your pitch should explain why the customer can trust your output, and how a change can be tested safely before it reaches production.
Autonomous control of capital equipment is a hard sell. Recommendations a human approves are far easier to adopt than closed-loop systems that adjust tools on their own. If closed-loop control is the long-term goal, show the human-approved path that gets you there.
Start where the pain is high and the incumbents are thin. Trailing-edge and mature-node fabs, analog and mixed-signal lines, MEMS foundries, and packaging and test houses face the same yield problems as the leading edge, with fewer in-house teams to solve them. They are a more credible first market than the largest manufacturers.
Go through the equipment chain. Partnering with the equipment makers and material suppliers who already sell into the fab can shorten qualification considerably. Expect the partner to take a share, and plan for it.
Expect long cycles and say so. Multiple demonstrations before a pilot, and pilots before purchase orders, are normal. A co-development agreement or a paid engineering pilot is accepted as early traction.

The business model

Evaluators push hard on how technical improvement turns into money.

Translate the metric into dollars. A yield, throughput, or power improvement means little until you state what it is worth to the customer per wafer, per tool, or per year. Do the conversion on a slide.
Show the margin structure. Model the specification, the component costs, and the target gross margin from first principles before claiming the business works.
Be careful with share-of-savings pricing. Customers resist paying a percentage of improvements they believe their own engineers produced, and they tend to treat the resulting process knowledge as theirs.
Seat-based software stays small. Process and yield engineering teams are small, so per-seat licenses rarely add up to a venture-scale business on their own. Show how revenue scales beyond headcount.
Choose between license, component, and full system. Evaluators will ask whether you sell IP, a module that slots into someone else's tool, or a complete product. Capital need and margin differ across the three, as does the sales cycle, so pick one for the next stage and explain why.
State the capital and time to revenue plainly. Investors in this sector accept capital intensity, but they expect the founders to know the number and the milestones it buys.

Competition and the right to win

Name the competitors, then explain why you are entitled to win against them.

Incumbent tool and EDA vendors may already offer a version of your product, or can add one. Show the structural reason they will not, such as a conflict with their installed base, a measurement they cannot make, or a market too small for them to prioritize.
The chip companies' own teams often build internal versions of process and design software. Show why a customer would buy from you.
Well-funded startups may already be ahead in your niche. Evaluators often know them, and a large recent raise by a competitor will come up. Explain precisely where you differ.
AI-native services now offer custom software for manufacturers, which raises the bar for any tool that can be described as a workflow.

A first-principles case carries weight here. Show that your approach can meet the specification at a cost and margin the incumbents cannot match, and why.

The team

Production experience changes the conversation. A founder who has worked in high-volume manufacturing, qualified a tool, or shipped silicon answers the credibility question before it is asked. If no founder has that background, an advisor or early hire who does should be visible in the pitch.
Scientist-led teams need a commercial plan. Strong technical founders without a commercial lead get the same question every time. Name the person, or the hire and when it happens.
Everyone must be committed. Founders who are part-time, split across ventures, or waiting on a later start date draw sustained skepticism. Have a dated plan for each founder.
Coachability counts, even for experts. Engaging openly with a challenge to a technical assumption earns credit. Defending the assumption without examining it reads as a founder who will be hard to help.

For silicon companies heading to tapeout

Founders designing their own chips should expect the general points above, plus questions specific to the silicon path.

The tapeout plan. Process node, foundry access, a shuttle or full mask set, the cost, and the date.
The first design win. Name the target customer and socket, then the evaluation process that leads to a purchase decision.
Software and toolchain. For compute chips, the compiler, SDK, and integration effort are often harder than the silicon. Show the plan.
Benchmarks against shipping products. Compare against what customers can buy today, on their workloads, and avoid comparing against a competitor's roadmap.

This section reflects common investor practice more than the selection interviews this guide draws on, which included few companies at the tapeout stage.

Questions to rehearse

Technical

Where does your data come from, and will the customer share it?
Have you validated against a real tool, real wafers, or a fabricated device?
What exactly happens under the hood, step by step?
Where does the heat go, and what happens to performance under thermal load?
How does the system scale from one unit to production volume?
Why won't the incumbent tool, EDA, or chip vendor build the same capability?
Can you show from first principles that you can meet the specification at a winning cost?
What is the failure mode if your system is wrong, and who carries the risk?

Business

Who is the first named customer, and what exactly do they buy?
What is the improvement worth to them in dollars?
Do you sell a license, a component, or a full system?
Who is full-time, who owns the IP, and which hire is missing?
How much capital does the next milestone need, and when does revenue start?
If your risk is technical, why do you need this program or investor now?

Red flags to avoid

Leading with the largest foundries or the leading edge as the first market
Assuming access to protected process data
Presenting simulation or a paper design as proof
Proposing autonomous control of production equipment as the first product
Treating scale-up from one unit to many as straightforward
Pricing as a share of savings the customer believes it produced
A team with no manufacturing, tool, or chip-design experience, and no plan to add it
Founders who are part-time or committed elsewhere
Analogies in place of a mechanism
Defending a technical assumption without engaging with the challenge

Checklist

Physical result on real equipment, with the metric before and after
Simulated or projected results clearly labeled
Data source, minimum data need, and bootstrapping path named
First market chosen where data access and adoption are realistic
Adoption plan covering neighboring process owners and safe testing
Equipment or supplier partner identified, with expected revenue share
Improvement translated into dollars per wafer, tool, or year
Gross margin modeled from first principles
License, component, or system decision made and explained
Capital needed and time to first revenue stated
Competitors named, with the structural reason you can win
Production or silicon experience visible on the team
Commercial lead named or hire planned
Dated full-time plan for every founder