Zapata Computing Holdings Inc. Common Stock (ZPTA) 8-K Entry into a Material Definitive Agreement

Exhibit 99.3

Zapata AI Business Combination with Andretti Acquisition Corp.

Investor Conference Call Script

September 6, 2023

Operator

Good day, and welcome to the Zapata AI Business Combination with Andretti Acquisition Corp. Announcement Conference Call. All participants will be in listen-only mode for the entirety of the call.

Before we begin, we remind you that certain comments made during this call may constitute forward-looking statements which are subject to significant risks and uncertainties that could cause the combined companies actual results to differ materially from expectations or historical performance. Please review the disclosure on forward looking statements included in Andretti Acquisition Corp.’s filings with the SEC for a discussion on these risks and uncertainties.

A recorded replay of this call and related materials will be available on the Zapata investor page. Please be advised that statements are current only as of the date of this call and, while Andretti may choose to update these statements in the future, it is under no obligation to do so unless required by applicable law or regulation. The comments made during this call are copyrighted by Andretti.

I will now turn the call over to Mr. Bill Sandbrook, Co-Chief Executive Officer and Chairman of Andretti Acquisition Corp. Please go ahead.

Bill Sandbrook, Co-Chief Executive Officer & Chairman of the Board, Andretti Acquisition Corp.

Thank you for joining us today.

The team at Andretti Acquisition Corp. is incredibly excited about our proposed business combination announced with Zapata AI. I will touch on why we see this as a compelling acquisition, at a foundational time, before turning to Michael to say a few words on his and Andretti Autosport’s experience with Zapata’s technology in action.

We see four main pillars for this transaction:

1. Generative AI is quickly expanding at a seemingly exponential rate across multiple verticals.

2. Zapata is a first mover in Generative AI

3. Zapata delivers products and solutions that are tailor-made for the specific needs and costs of their business customers

4. The transaction is aligned with Andretti Acquisition Corp’s goal of pushing emerging technologies into mobility, which stems from Michael Andretti’s visionary leadership of leveraging technology within motorsports.

1

To briefly elaborate.

You have likely heard the buzz around Generative AI – it seems to be “the” focus of Big Tech, with some eye-popping estimates around its future Total Addressable Market, or TAM, size. It has been a topic of significant discussion during the past few earnings seasons, with Big Tech noting the positive impact it is having on their business outlooks. As a fun fact, Alphabet, the parent company of Google, mentioned AI an eye-catching 70 times¹ on its July earnings call. The Zapata team will elaborate more on market size in a few minutes, but we think it is very important to note that generative AI’s applicability spans across many, many verticals and industries, including areas like automotive, pharmaceuticals, and finance, just to name a few.

Building on this, Zapata is a well-known and respected first-mover in the Generative AI space, with some of the brightest scientists, engineers, and developers in the business, and an existing revenue base from some impressive Fortune 1000 customers and the U.S. Government. They’ve been working on Generative AI since 2018, with a strong patent portfolio. You’ll hear more on this from the Zapata team too.

Further, the Zapata team is tackling the unique challenges posed by enterprises by deploying an Industrial AI tool. Businesses have very different needs from those of consumers. By combining, among other differentiating features, their background in quantum computing with their work in Generative AI, they can deliver products and solutions that are tailor-made for the specific needs and costs of their business customers. This very much fits in with our mission of identifying a company that displays technological leadership – a company with a business model that addresses or creates a market need that other companies have not.

Lastly, Zapata is putting their tech into practice right in the “wheelhouse” of the Andretti brand – mobility, specifically motorsports — through their partnership with Andretti Autosport in the NTT INDYCAR Series. Pushing the limits of technology is key to success in motorsports, which has long been a part of the Andretti vision. As one such example, look no further than the formation of Andretti Technologies in 2015 to build electric powertrains for Formula E while the all-electric series was still in its infancy.

For this reason, one of our goals with Andretti Acquisition Corp from day one has been to identify a partner that is pushing emerging technologies into mobility and, to the extent possible, motorsports.

Zapata could not check this box any more firmly.

The Zapata and Andretti Autosport teams entered into a commercial relationship in early 2022 to seek out ways to enhance results in their INDYCAR Program through Zapata’s advanced analytics.

The partnership is strong and continues to build today, and who better to say a few words on it than Michael Andretti himself, so Michael, over to you.

¹  https://www.pymnts.com/google/2023/google-reports-7-out-of-10-generative-ai-unicorns-are-cloud-customers/

2

Michael Andretti, Co-Chief Executive Officer & Director, Andretti Acquisition Corp.

Thank you, Bill. We are thrilled to announce this acquisition today. As Bill noted, I originally became aware of Zapata and their capabilities by having them as a technology partner with Andretti Autosport in the NTT INDYCAR SERIES.

The relationship with Zapata on the Andretti Autosport side started in 2021. In February 2022, we announced a multi-year partnership with Zapata, including brand sponsorship and a multi-million-dollar agreement to use their enterprise software platform, Orquestra.

In motorsports, every hundredth of a second counts, and every strategic decision counts. Races can be won or lost due to pit strategy and the timing of yellow flags. By deploying Orquestra in our racing operations, we believe we will be able to realize a real-time performance edge on race day. As an expert in their technology, I will let Christopher Savoie, CEO and Co-founder of Zapata, give more color on how they are helping us with their technology in a few minutes.

Put simply, with Zapata, we are racing toward a winning future with generative AI. We are excited to bring the company and awareness of their technology to the public markets at what is an incredibly exciting time for generative AI.

So with that said, let me turn the mic over to Christopher. Christopher?

Christopher Savoie, Chief Executive Officer & Co-Founder, Zapata AI

Thank you, Michael, Bill, and the team at Andretti Acquisition Corp., and hello to everyone listening in.

Zapata AI is an enterprise software company building Industrial Generative AI.

Industrial generative AI is similar to consumer generative AI tools like ChatGPT that generate text and images. But it’s tailored to enterprise use cases, taking the generative models behind these popular tools, and applying them to critical, industrial-scale applications involving both language and other forms of data. I’ll get into some examples in just a minute.

Zapata has been working on generative AI since long before applications like ChatGPT even existed. Our first generative AI patent was filed in 2018.

3

Since then, we have built two core offerings:

• The first is Zapata AI Prose: a set of large language model-based generative AI solutions similar to widely used generic chat applications but customized to an enterprise’s industry and its unique problems. Prose can help companies speed up time-consuming language tasks like filing for regulatory approvals or patents, filling in customs forms, or creating documents or reports.

• We’ve also built a complementary solution called Zapata AI Sense that can handle complex mathematical models – something that is ubiquitously important for industry but somewhat under-appreciated due to the recent generative AI boom. Sense can help enterprise leaders make smarter decisions by enriching their business analytics with realistic, generated data to fill in the existing gaps in their data — including for variables that are not otherwise measurable. We’ll touch on the importance of this shortly with an example from our work with Andretti.

Our generative AI solution set is “industrial”, meaning it is meant for use in a business context to address specific challenges or improve efficiencies. Again, this is not really the case with chatbot-like applications today. The stakes are much higher when doing things for business, which I will elaborate on in a bit as well.

All of this is provided over our full-stack enterprise software platform — Orquestra — which allows us to train and deliver our complex models over various cloud solutions, including Azure Cloud, AWS Cloud, and others.

One of Orquestra’s most impactful benefits is its flexibility, which allows our customers to experiment with different models on different types of computer hardware — and NOT be locked into any one specific cloud provider, which they tell us is very important to them.

To date, we have worked with customers across various industries, including automotive, oil and gas, chemicals, and finance, to name a few. Our customers have included the likes of BASF, BP, the global bank BBVA, and DARPA, which is the innovation arm of the Department of Defense, and, of course, Andretti Autosport.

Before getting into Zapata’s enterprise offering, I want to describe the basics of generative AI for anybody who has yet to sample a product like ChatGPT.

Generative AI technology is a huge step forward compared to “traditional” AI. At a simple level, traditional AI answers a specific question or performs a relatively simple, or straightforward discriminative task.

Generative AI, however, uses machine learning to generate what is, in effect, a new, original output or product. It will synthesize various inputs to distill and produce an original, creative output that is not just a regurgitation of the data it was trained on.

For example, in a matter of seconds, it can draft a multi-paragraph answer to explain a concept with varying degrees of specificity depending on how the question was asked. Compare this to a search engine like Google simply pointing you to various sources, from which you would then have to synthesize and distill the information yourself.

4

This works well enough—most of the time. But as we all know, what seems like a simple inquiry can sometimes consume more time than any of us would like.

While society has seen rapid growth in and awareness of generative AI among the general public due to applications like ChatGPT, it is essential to note that for Zapata, this is not an “overnight thing.” Since our first generative AI patent filing in 2018, Zapata has been, in effect, in a dead heat with its competitors in terms of the number of quantum-inspired AI patent applications on file. A 2021 insight report from the European Patent Office with a breakdown for most active applicants related to quantum computing and artificial intelligence /machine learning showed Zapata had nearly twice as many international patent application filings as Meta or Google that year.

We have over 100 global patents and patent applications, covering various algorithms, use cases, and supporting software and hardware. We believe there isn’t any startup or other company out there that’s a pure play in this category, and none are in the process of going public aside from Zapata.

One of the most important takeaways I want to leave you with is this: the applicability of industrial generative AI technology is truly vast.

I say this because companies across the globe — and sectors — recognize this trend and are racing to find the “killer app” for industrial generative AI. Executive decision-makers are eager to know how to leverage this technology to improve their businesses.

This is evidenced by a Gartner Poll from this past May that revealed that 45% of executives have noted that the deployment of ChatGPT merely six months earlier had prompted an increase in AI investment. More than 70% are actually in generative AI “exploration mode” already.

We are experiencing this momentum every day at Zapata. As awareness and the popularity of Generative AI have drastically increased in the past few months, there has simultaneously been a significant increase in interest from — and conversations with — current and potential future customers.

All that said, there are still several significant challenges with scaling up and commercializing generative AI “as is” today. From where we stand, it appears that Big Tech has, to this date, been focused on one-size-fits-all generative AI models that are trained on the entire internet and on general data. This means tremendous applicability in a consumer setting, but we don’t think it can really work in most business settings — at least not the way this technology is structured today:

• For one, the potential for errors and inaccuracies is too high, which is unacceptable in many business contexts. For example, recent research found that GPT-4, the more powerful model behind the paid version of ChatGPT, has seen its mathematical accuracy drop from 97.6% in March to 2.4% in June².

²  https://www.zdnet.com/article/gpt-4-is-getting-significantly-dumber-over-time-according-to-a-study/

5

• Secondly, the costs can be enormous when you factor in how much compute time and resources are required to run and train large generative models. For example, a model like GPT-3 costs around $4.6 million to train on the cheapest available compute³, and the more powerful GPT-4 likely costs significantly more than that.

• There are also legitimate issues around privacy and security, monitoring, ethics, and so on. Samsung, for instance, has banned staff from using ChatGPT after a leak of sensitive data⁴.

Said differently, we believe businesses do not need massive, costly, inefficient generative AI models trained on the internet to do very specific, customized applications. They do not want general-purpose models with unreliable outputs for their domain-specific problems. They also do not — by and large — want to be locked into a single vendor’s compute and cloud choices.

Rather, we believe enterprises want to keep their own data and their own models… they want to run these models on their own clouds and with their own security measures. They don’t want to worry that a Big Tech generative model trained on their private data and IP will expose this sensitive information or have it used against them by competitors. They want complete control over their generative AI applications, data, and models.

This is where Zapata’s value proposition really comes into play. Put simply, we have developed a suite of custom industrial generative AI solutions that can harness the power of language and numerical models for critical, sensitive industrial-grade applications. Our solutions are fine-tuned for our customers’ domain-specific problems, optimized for cost, benchmarked to the highest level of accuracy and business impact possible, and run securely in our customers’ own environments.

So, with that said, let’s dive a little deeper into our technology to demonstrate its advantages for generative AI.

Our technology is derived from math-inspired quantum physics. And if you’re thinking, “You mean, like electrons and photons and things like that?” — you would be correct. The hard part is turning that discipline of physics into useful technology. Fortunately, our work in this area has many transferable and positive implications for generative AI.

Being experts at quantum math —another one of Zapata’s differentiators given the robust staff of PhDs and physicists we employ — allows us to enhance key, desirable qualities of generative models. Namely, quantum statistics can enhance generative models’ ability to generalize — or extrapolate missing information and generate new, high-quality information — as well as their ability to generate a more varied range of solutions. This is called “expressibility”.

³  https://lambdalabs.com/blog/demystifying-gpt-3

⁴  https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak#xj4y7vzkg

6

We recently demonstrated the advantages of quantum math for generative AI in research published with Foxconn, Insilico Medicine, and the University of Toronto — with significant implications for drug discovery. Our research showed that generative models enhanced with quantum components generated more desirable drug-like molecules than those generated by traditional generative models.

Our technology is also efficient. Large language models, as the name suggests, are huge. This means they can use many computer processing resources — typically powered by Graphics Processing Unit chips, or GPUs as they are commonly called. These specialized chips are required to train and run models with the size and complexity of LLMs. And with GPUs comes high costs and large carbon footprints.

We believe we have the largest language model compressed with quantum-inspired algorithms, and this ultimately means that Zapata will be able to deliver high-quality and significantly more cost effective and environmentally friendly products and solutions.

Another very important factor is speed. A good example of an application that we can speed up is what’s called a Monte Carlo simulation. This is a type of model used to predict the probability of a variety of outcomes when the potential for random variables is present. It can be used across a variety of fields — examples include in finance to assess risks associated with an investment; and in project planning to arrive at informed views on the probability of completing a project within a certain timeframe.

At Zapata, we have demonstrated an example whereby we ran very complex scenarios on both our technology and via a traditional Monte Carlo simulation. As part of our testing, we determined that our approach was 8400X faster — with the Zapata method arriving at a solution in three seconds versus the seven hours it took using the Monte Carlo model. We have found that the quantum inspired method is thousands of times faster and it is also more accurate.

We understand this can be a challenging topic for many to fully understand and process, so let’s turn to some real-world examples of where this technology can be used.

As touched on earlier, the origin of our relationship with Andretti Acquisition Corp is the strong partnership Zapata has had with Andretti Autosport for the past two seasons. There is no better or more exciting way to provide an example of real-world applicability than this unique and, frankly, fun case study, so here it goes.

As any motorsports enthusiast knows, race cars are incredibly technologically advanced, and data is critical to performance. Races are often won or lost — sometimes by just milliseconds — based on strategy, and strategy relies on data.

An INDYCAR is outfitted with many sensors that gather real-time data, including factors critical to performance. However, not everything important to performance can be measured in real-time with sensors while a car moves around a racetrack at speeds exceeding 200 miles per hour.

7

One such example of helpful information that cannot be measured via sensors is the slip angle of a car, which describes where the car is slipping while the car is lapping the track. This is important to understand how fast the vehicle will go around a given track and certain aspects such as tires may perform as a result.

When the vehicle turns left, the car wants to go to the right because of the centrifugal force, which is extremely pronounced at high speeds. And other factors like downforce and the shock absorbers also alter the slip angle of the tires. The exact degrees of these various forces and how they evolve over a tire cycle impact tire performance, and thus impact lap time over that period.

Knowing exactly how the tires are performing or how the slip angle is affecting the speed with sensors in real-time is impossible, but being able to accurately predict the slip angle with generative modeling could inform the team when the optimal time is to make a pit stop to change the tires, for example — and if there are adjustments to the car’s set-up that can be made to potentially improve tire performance and lap speed.

With Zapata’s industrial generative AI offering, we are working to generate virtual sensors that gather this critical data that is otherwise unattainable in real-time, race-day conditions.

Andretti Autosport’s team has collected terabytes of data over twenty years, and using generative models, the Zapata platform is working to accurately model the performance of the vehicle.

In fact, when reviewing the data, the Andretti and Zapata teams have found that there is hardly any difference between the Zapata-predicted data – also known as synthetic data — and the actual performance data from the cars – with a less than 1% difference between the synthetic data from the generative model and the real data gathered. As you might imagine, the potential competitive edge allowed by having more accurate predictive models is most certainly game-changing.

Zapata’s technology can also be used to suggest new solutions to industrial optimization problems, particularly when making complex industrial processes more efficient. Our proprietary technique, which we call Generator Enhanced Optimization, or GEO, uses generative models to learn the distributions of possible solutions and then propose better solutions.

These generative models can use traditional computing — or what we call “classical computing” — quantum computing, or quantum-inspired architectures, thus providing a way to solve problems today with traditional and quantum-inspired computing and take advantage of the potential benefits of quantum approaches as quantum hardware matures.

A helpful example of this in practice can be found in manufacturing plant optimization and the work we’ve done with BMW and MIT.

8

Automotive manufacturing is incredibly complex, with hundreds of parts from various suppliers, skilled labor trained to perform specific functions, and various union and labor laws that dictate when and how much employees can work. Given these multiple inputs that all need to be synced together, inefficiencies and lost productivity can quickly become a severe problem. Because of this, finding the most efficient worker schedules to achieve production targets while minimizing idle hours is imperative for plant managers.

By partnering with BMW and MIT and deploying our GEO framework on our Orquestra platform across multiple BMW plants, we found that in 71% of cases, our algorithms could tie or outperform their existing state-of-the-art optimization algorithms, demonstrating that we can help them optimize their scheduling.

As touched on earlier, we already work with — or have worked with — several very large and well-known companies across the automotive, chemicals, and finance industries, to name a few. To elaborate on the market and our go-to-market strategy, I’d like to now turn the call over to Mick Emmett, VP, Marketing & Communications. Please go ahead, Mick.

Mick Emmett, VP, Marketing & Communications, Zapata AI

Thank you, Christopher.

When evaluating the market, it is imperative, we believe, to understand the potential benefit from generative AI in estimated dollar terms. The total addressable market for generative AI use cases and their adjacencies is expected to be $1.3 trillion by 2032, which includes a potential serviceable obtainable market of $280 billion in generative AI software and $86 billion in generative AI IT services. Even if off by an order of magnitude, this still represents a HUGE Total Addressable Market – or TAM, and serviceable obtainable market – or SOM.

So, within that context, how will Zapata AI grow its business?

We have two primary sales channels — a direct channel, where we approach companies with C-level relationships – and through a partner ecosystem.

Today, we have a global salesforce in the U.S., Europe, and Asia, but we cannot be in the market speaking to every company – that would be impossible. As such, we have partnered with companies such as Microsoft Azure, IBM, and Nvidia, to name a few, as well as a top-5 global consultancy company to amplify our reach.

In our view, the business model and how we generate revenues is straightforward. We sell our product as a bundled subscription of software and the scientific and engineering expertise and support necessary to build applications. If we can demonstrate the value of our technology, we convert these engagements into multi-year, multi-million-dollar contracts through a subscription model.

We believe Zapata has credible industry and academic backing. You will see us at many academic and industrial conferences, discussing how we build our brand through various customer success stories. That said, given how hot the generative AI niche is, we must be prepared for this to be a competitive environment.

9

Turning to the Zapata team: we are proud of the people driving Zapata forward. Even before our proposed business combination came into being, we had — and currently have — a public-ready board comprised of operators and people with rich histories in the enterprise software industry.

Some notable individuals include:

• Jeff Huber, founding CEO of GRAIL and former SVP of Google Ads, Apps and Maps

• Clark Golestani, the former Global CIO at Merck

• Rhonda Germany Ballintyn, former Chief Marketing Officer and Chief Strategy Officer of Honeywell

• Dana Jones, CEO of RealPage and the former CEO at Sparta Systems, a life sciences-focused next gen Software as a Service company that sold to Honeywell a few years ago

To conclude, we believe Zapata is a pure-play company within the burgeoning and transformational industrial generative AI technology space.

Through the proposed business combination with Andretti Acquisition Corp., we can become what we believe would be the first publicly traded, pure-play, industrial generative AI company. In a large and rapidly growing total addressable market, we have proprietary, industrial generative AI techniques and algorithms that we believe are at the leading edge of these new frontiers. Based on our data, we have determined that they are capable of demonstrating a 10X to 1000X improvement in modeling performance, and Zapata brings its proprietary, full-stack software platform to deliver these solutions.

We believe we have substantial near-term enterprise revenue opportunities with large language models and other models in AI simulation and optimization and a pioneering, founder-led, and visionary management team and board with a track record of execution.

With that said, I would now like to turn the call over to Matt Brown, President & CFO of Andretti Acquisition Corp., to provide a brief high-level overview of the transaction. Matt, please go ahead.

Matt Brown, President & Chief Financial Officer, Andretti Acquisition Corp.

Thank you, Mick.

The transaction we are announcing today values Zapata at an implied pre-money equity value of $200 million, with existing Zapata shareholders set to roll over 100% of their equity into the combined entity. Andretti Acquisition Corp.’s sponsors and certain investors that own or have the right to receive founder shares will own a combined 5.8 million shares or an implied value of approximately $58 million. Andretti Acquisition Corp’s current public shareholders will own the remaining outstanding shares with the amount depending on the level of redemptions.

10

Andretti Acquisition Corp today has approximately $84 million in cash in trust at a $10.66 per share price, or approximately 7.9 million shares. Putting this all together, the pro forma equity value of the combined company is expected to be between $281 million and $365 million depending on the level of redemptions. We expect to incur approximately $12 million in transaction-related expenses.

Lastly, we expect the transaction to close in the first quarter of 2024, with various factors outside of our control, such as the timing of the SEC review and approval process.

All of us at Andretti Acquisition Corp. and Zapata are excited to bring this transaction to market, and we would like to thank you for your time today.

We hope that you follow us along our journey in the coming months. Have a great day.

Operator

Thank you, you may now disconnect.

11