What Anthropic’s IPO means for IT buyers
Computerworld ·

Anthropic will make its stock market debut Thursday, seeking a valuation of around $1.3 trillion. While the initial public offering (IPO) is set to attract huge interest from investors, there are implications for Anthropic’s enterprise customers too. The IPO will be “a huge inflection point” in terms of transparency around Anthropic’s operations, said Arun Chandrasekaran , distinguished VP analyst at Gartner. “At the same time, it might also force a different engagement from Anthropic as a company, particularly around longer-term contracts, hybrid pricing models, and so on,” he said. Anthropic had more than 11,900 enterprise customers spending $100,000 or more annually as of the end of September, according to its recent S-1 filing , with over 1,900 spending $1 million or more, and around 250 customers spending $10 million or more. Many of its largest customers are not on long term contracts and have the ability to reduce or, in some cases, cease spending at any point, Anthropic said, though a growing number of enterprise customers are signing annual or multi-year agreements with minimum spending commitments. As a publicly traded company, Anthropic will be under greater pressure to show continued revenue growth and improving margins in its quarterly earnings. This could change how it does business with enterprise customers, say analysts. “Enterprise buyers … can expect a stronger push toward committed-spend and multi-year contracts, because investors reward predictable revenue and visible backlog more than usage that can swing from month to month,” said Nick Patience , VP and practice lead for AI Platforms at Futurum Group. Chandrasekaran agreed, noting that one- to three-year contracts are standard for software as a service vendors, and frontier AI labs such as Anthropic may seek to replicate this. Patrick Moorhead , founder and chief analyst at Moor Insights & Strategy, also expects a shift towards longer contracts, alongside an emphasis on “larger prepaid volumes and fewer one-off discounts.” This push to secure longer deals and larger spending commitments could hand customers leverage in negotiations, analysts believe. “Right now, Anthropic wants … the committed revenue,” said Brian Hopkins , Forrester vice president and principal analyst, “so we think this is the time to ask for price protection, rollover of unused credits, the right to shift spend to new models as they ship, and decent exit terms.” Patience added that buyers should use the opportunity to “lock in price protection, capacity guarantees and clear terms on model deprecation.” And, Chandrasekaran suggested, to entice customers to longer deals, Anthropic may also offer benefits such as early access to new features, and opportunities to work with its forward-deployed engineers . However, Patience said, buyers shouldn’t overestimate their potential leverage, such as around negotiating lower prices. With demand for its services currently outstripping supply, Anthropic has “little reason” to cut its prices, he said. Competition from the likes of OpenAI and Google — alongside availability of increasingly capable open-weight models — is likely to have a greater effect on pricing than going public, he said. Hopkins, meanwhile, expects changes to Anthropic’s pricing structure rather than to its list prices, with additional payment tiers and increased bundling of agents, coding tools, and enterprise features. What should IT buyers monitor in Anthropic’s quarterly reports? For Hopkins, a key upside to the IPO for customers is transparency. Aside from the company’s recent S-1 filing, most of the information about the company’s finances to date has come from press leaks, he said, and the quarterly reports will provide customers with figures for revenue, margins, customer concentration, and infrastructure commitments. “I’d make those quarterly filings part of vendor risk management, same as for any strategic supplier,” said Hopkins. Patience agreed. Anthropic’s customers should read quarterly reports as closely as investors do, as “the numbers that move the stock also predict what happens to pricing and service,” he said. Chandrasekaran observed that quarterly earnings reports may also shed light on Anthropic’s expansion beyond AI models, as it invests in various areas such as functionality for Claude users, developer tools, and professional services. This could provide IT buyers with “better visibility into which bets are working and are not,” he said. Patience noted some of the factors that enterprise customers should track in earnings reports: Compute cost as share of revenue. This fell to 48% in the second quarter before rising to 58% in the third, as Anthropic paid premium rates for short-term capacity. “If that number keeps climbing, pressure follows on prices, rate limits, or which models are offered at which tiers,” he said. Customer concentration. Anthropic’s S-1 filing showed that a small number of customers account for a large share of its revenue. “Any change there shows how dependent Anthropic is on a few big buyers and how much weight smaller enterprises carry in its plans,” he said. Long-term compute commitments to cloud and chip partners. “Most of these can’t be cancelled, so they show whether Anthropic will have the capacity to serve customers reliably and what fixed costs it has to cover,” said Patience. Even as businesses increase their spend on AI, there’s considerable uncertainty about whether customer demand will increase fast enough to underpin the immense sums being invested in compute infrastructure. Hopkins suggested customers pay attention to Anthropic’s financial position, but to not panic at the potential of an AI bubble bursting. Forrester’s “ AI Bubble Model ” tracks the companies funding and supplying the AI build-out, such as Microsoft, CoreWeave, Oracle, and the big chip suppliers. This indicates that they are financially healthy, he said, “which is why we don’t think a burst is imminent.” However, Hopkins said that AI labs such as Anthropic and OpenAI face different challenges due to their huge infrastructure commitments that rely on continued growth. Anthropic’s S-1 showed commitments of at least $121.3 billion to Google Cloud through 2033, and at least $110 billion to AWS through 2036. Around $161 billion in rent is due over five years on leases for Broadcom-built chips. “Those are bets that revenue keeps growing fast,” said Hopkins. If growth were to slow, the risk to customers isn’t that Anthropic disappears overnight — “Amazon and Google have a lot riding on it,” he said — but rather that terms, such as prices, rate limits, and capacity, could change. As well as the company’s finances, Patience said customers should pay attention to Anthropic’s corporate governance structure. Anthropic is a public benefit corporation, which means it is not legally bound to maximize shareholder value over other aims, such as promoting AI safety, while its Long-Term Benefit Trust has the authority to select and remove board members. The governance structure means that “public shareholders won’t control the company the way they would at a conventional listed firm,” said Patience. “Buyers should watch whether that structure holds up under quarterly pressure, because it shapes how stable Anthropic’s usage policies and model roadmap will be.” ‘Optionality is your negotiating leverage’ As Anthropic becomes a publicly traded company, analysts highlighted the importance of flexibility in contracts and the ability to avoid technical lock-in. Hopkins recommends clients be prepared to switch models, and avoid signing any commitment they can’t easily exit. “So if Anthropic tries to boost revenue by monkeying with contract terms, customers can threaten to walk and mean it,” he said. Patience added that Anthropic’s IPO is a “good prompt to check that applications aren’t locked to one model provider.” In addition, Chandrasekaran pointed out, “the good news is there are a lot of abstractions, like model routers and AI gateways and similar technologies, that enable customers to … reduce their dependency.” However, he also warned that it becomes harder to avoid lock-in as customers adopt a broader range of tools from a provider. With technologies such as agent harnesses, “the propensity to be locked in is far higher than at the model layer,” he said. Moorhead suggested that customers avoid signing long single-vendor terms “without price protection, model version and deprecation guarantees, and committed capacity and rate limits.” “Get clear data handling and exit terms,” he said. “And keep an abstraction layer in your stack so you can move workloads between models. That optionality is your negotiating leverage.”