ARK • Disrupt

Anthropic—the AI development company behind the top-tier Claude models—is reportedly seeking a valuation as high as $40 billion in a new fundraising round.¹ In the wake of OpenAI's ongoing efforts to raise $6-7 billion at a valuation of ~$150 billion, Anthropic's talks with investors are still in early stages. These reports are leading to questions about the justification for these lofty valuations.²

ARK believes the justification lies in AI’s potential to transform the software industry. AI makes knowledge workers more productive.³ Based on the radical productivity gains that we expect AI to deliver, businesses could increase spending on software 47% at an annual rate during the next six years, from $1.1 trillion this year to $13 trillion in 2030.⁴

The math is simple: knowledge workforce wages outside of China are likely to total $30 trillion by 2030. Software vendors typically aim to capture 20 - 40% of the economic value that they deliver to enterprises, and we expect that AI software will be able to increase 2030 knowledge workforce productivity by between 2.5x and 6.5x. If AI were to capture 10% of that economic value, the result could look like an end-market ranging between $7 and $19 trillion.⁵

While seemingly good for software companies, the wrinkle is that AI could empower knowledge workers to write their own software quickly and inexpensively.⁶ Why would businesses pay for one-size-fits-all off-the-shelf software when they can code custom-tailored solutions quickly and more cost effectively? A multi-trillion-dollar question!

Historically, ~40% of public cloud compute spending has centered on software-as-a-service (SaaS), as opposed to platform-as-a-service (PaaS) like Palantir’s AIP and infrastructure-as-a-service (IaaS) like Amazon Web Services (AWS).⁷ SaaS providers often are customers of PaaS and IaaS providers, so a dollar spent on SaaS could result in ~$0.25 of downstream spend for PaaS and IaaS, but SaaS providers capture the bulk of the revenue.

AI is likely to transform the tech stack. If SaaS were to give way to in-house AI solutions, then open-source and commercial AI platforms like Llama 3, Anthropic, and OpenAI probably would power them.

The scenarios below dimension how the demand for enterprise AI software could cascade through the value chain. Were it to roughly retain its current share of spend, SaaS could generate $6.5 trillion in revenue and ~$40 trillion in enterprise value⁸ delivering 70+% annual growth on both measures, by 2030.⁹

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

In our view, however, AI will diminish the role of software-as-a-service in the value chain. Instead of maintaining the lion’s share of spend, SaaS could give way not only to home-grown solutions built on foundation models, but also to PaaS or IaaS. Among the hints that the software layer is commoditizing, Chamath Palihapitiya of Social Capital has offered to seed companies delivering software at one-tenth the price of traditional SaaS.¹⁰ Meanwhile, Klarna has announced that it is abandoning Salesforce and Workday, in part because AI will allow the company to develop its own custom solutions.¹¹ If these trends prove durable, the future state of software could look very different than it does today.

In the scenario below, while total software demand tracks the same trajectory as in the model above, software-as-a-service gives way to in-house software solutions enabled by PaaS and IaaS. Instead of $6.5 trillion in revenue, SaaS could generate little more than $1 trillion, growing 32% instead of 72% at an annual rate, and dropping from 38%of non-hardware AI spend to 9%.

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

The lower share of spend on SaaS would result in a higher share for AI foundation models and a much larger share for underlying AI compute hardware. Spend that skips the SaaS layer of the value chain results in lower gross margin capture, freeing up more resources for the underlying compute infrastructure. Moreover, given the potential of AI to benefit businesses with proprietary data, direct investments in on-premise AI compute hardware could proliferate.

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

Perhaps the true outcome lies somewhere in between the two scenarios, but if we are correct directionally that businesses will create much of their own software, the technology value chain should transform significantly, as shown below. Compute hardware manufacturers, AI foundation model companies, PaaS, and IaaS will accrue more value relative to SaaS than otherwise would have been the case.

Source: ARK Investment Management LLC, 2024. This ARK analysis draws on a range of external data sources, as of 9/27/24, which may be provided upon request. Forecasts are inherently limited and cannot be relied upon. For informational purposes only and should not be considered investment advice or a recommendation to buy, sell, or hold any particular security.

In that context, the seemingly lofty valuations of AI foundation model companies today could prove a good value in hindsight. Even if the reported valuation rumors for Anthropic and OpenAI are correct, along with that for xAI they would approach ~$220 billion in current market capitalization while the portion of the value chain that they are aiming to capture could scale to $15-20 trillion by 2030. Lofty valuations—yes—but a fraction of the ultimate opportunity they are pursuing.

At Meta Connect 2024, AI was the cornerstone of Meta's strategic vision.¹² Mark Zuckerberg highlighted how AI is enhancing user experiences across its current platforms and its new hardware devices. From automated content creation in many languages to real-time personal assistants integrated into smart glasses, AI is permeating every platform at Meta these days.

Central to Meta's consumer offerings is Meta AI, built on Llama 3, Meta’s open-source foundation model. With nearly 500 million monthly and 185 million weekly active users, Meta AI is poised to transform the way people interact with technology on a day-to-day basis.

Meta also introduced Llama 3.2, a versatile multimodal AI model capable of processing images, graphs, and text. The release includes small- and medium-sized vision large language models (LLMs) (11B and 90B parameters, respectively), along with lightweight and text-only models (1B and 3B, respectively) that can run on edge and mobile devices. The smaller models are available in both pre-trained and instruction-tuned versions. Llama 3.2 will power real-time video analysis, translation features, and object recognition for the updated Ray-Ban Meta smart glasses, transforming them into highly capable, intelligent wearables that offer live language translation and enhanced situational awareness.

The highlight of the conference was the prototype Orion AR glasses, which Mark Zuckerberg described as a "time machine." Weighing under 100 grams, Orion features a holographic display, eye and hand tracking, voice activation, and a neural interface operated through a wrist device. Though not yet ready for broad-based distribution, Orion showcased Meta's vision for the future of augmented reality, with AI integrated deeply into the fabric of daily computing and interaction.

With dedicated AI hardware, highly proprietary data sources, and an extensive user base, Meta is well-positioned to lead the next era of AI, combining technological innovation, extensive data, and widespread distribution to shape the future of computing and augmented reality.

Last week, Zipline livestreamed its new drone delivery platform, the P2 Zip, for American consumers.¹³ Zipline’s innovative design includes a droid that emerges from a larger drone and tethers down to deliver packages quietly and efficiently to homes. Zipline says it can meet or beat current delivery economics for every partner with whom it works today.

ARK previously estimated that a drone delivery platform should be able to deliver packages for less than a dollar when it hits full scale.¹⁴ Today, however, drone delivery simply needs to undercut the $15-16 fees on average per delivery for app-based services. Unbelievably, the exorbitant fees today can double the cost of a meal.¹⁵

Partnered with Walmart as well as fast-casual restaurant chains and healthcare providers, Zipline plans to start deliveries in Dallas later this year. P2 Zips have an all-weather range of 10 miles and can carry up to 8 lbs., nearly enough for a day of meals for the average US household.¹⁶ ARK believes that autonomous delivery could reshape shopping habits, possibly turning weekly grocery trips into daily autonomous deliveries.

All in, ARK estimates that the total addressable market (TAM) for food and parcel delivery could be $1-2 trillion globally by 2030.¹⁷ With over 1 million deliveries and counting, Zipline is positioned to become the market leader.¹⁸