PAUL MCBETH: Anthropic asks investors to trust it won’t do evil

Paul McBeth is the editor of The Bottom Line and Curious News, and previously worked at BusinessDesk for 15 years. He has a Claude Pro plan and uses free versions of Grok, ChatGPT, Gemini and Kimi.

Anthropic is selling more than shares in its upcoming initial public offering.

It’s selling a vision of what a responsible steward of artificial intelligence can achieve.

And even if it’s half-right on the range of societal ills that the rapidly developing technology can cure, that will still put the Claude developer in a plum position to enjoy some very real economic benefits.

The firm’s S-1 filing with the US Securities and Exchange Commission already shows the five-year-old company has put some serious runs on the board. In the six months ended June 30, Anthropic’s revenue soared to US$16.29 billion from US$1.16 billion a year earlier, and its operating loss had narrowed to US$693 million from US$2.14 billion.

Throw in a cash balance and short-term investments of US$131.7 billion at the end of June, and Anthropic has some heft behind it heading into what could very well be an even bigger IPO than SpaceX’s listing in the middle of the year.

To put that in perspective, when Google went public in 2004, it was coming off an almost trebling of annual revenue to US$961.8 million in 2003, while Facebook’s 2012 IPO came after its annual revenue had almost doubled to US$3.71 billion.

It makes me wonder

Even allowing for inflation, Anthropic’s sales figures are stratospheric in comparison – and the company still has another six months of the year to play out. With growth like that, it’s easy to see why the doyens of Wall Street just can’t keep their mitts off the ever-changing AI theme.

So, what’s the catch?

Well, the price will be the starting point and we’re still to see where Anthropic will land on that, with the IPO filing refraining from settling on just how much money the AI giant wants to raise and how much it will be valued at.

The capital raising has been speculated to be as much as US$100 billion at a valuation of somewhere between US$1.5 trillion and US$2 trillion, and if you use SpaceX’s revenue multiples as a guide, you land somewhere in the realm of US$1.55 trillion-to-US$1.81 trillion.

So far, so normal in the heady land of AI.

And while the Anthropic prospectus acknowledges the potentially catastrophic and existential risk posed by AI, it backs itself as an organisation founded on the guiding principles to protect humanity from losing control of the emerging technology.

The voices of those who stand looking

After all, its co-founders Dario and Daniela Amodei, Tom Brown, Jack Clark, Jared Kaplan, Sam McCandlish and Chris Olah were willing to walk away from OpenAI amid growing discomfort with what they saw as an increasingly lax approach to the risks posed by its models.

Anthropic’s prospectus talks up the culture of safety, backed by governance arrangements under which the Long-Term Benefit Trust, which doesn’t have a financial interest in the firm, can appoint the majority of the board to make sure the company sticks to its public benefit mission.

There’s an element of ‘trust us, we know what we’re doing’ in the pitch, and it’s reminiscent of Google co-founders Larry Page and Sergey Brin talking up one of their guiding principles in their 2004 IPO: Don’t be evil.

That’s a much harder proposition for a publicly listed company to adhere to, and one that – on a much smaller scale – Californian outdoor clothing brand Patagonia seemed to get in 2022 when it chose to pour 98% of the company’s stock into a non-profit organisation to use its US$100 million or so of annual profits to battle climate change.

Owner Yvon Chouinard weighed up a number of options at the time, including an IPO, but ended up coming to the conclusion that public companies with good intentions still faced too much pressure to create short-term gain at the expense of long-term vitality and responsibility.

Google may have got caught up in business as usual for a publicly listed company. Its monopolistic tendencies got a rap over the knuckles when the courts stopped short of a full-blown Standard Oil moment for Big Tech, and its YouTube streaming platform has faced many of the same pitfalls as other social media providers in its aggressive hustle for never-ending engagement.

There’s still time to change the road you’re on

Putting those to one side, the search engine giant still provides a template for Anthropic in that it’s largely stuck to the 2004 pledge by Page and Brin to donate 1% of the group’s profits, with more than US$6 billion of funds being donated through the foundation set up alongside the IPO.

Anthropic seems to want to tap into that ethos, setting aside 5% of the company’s equity for a fund devoted to public good and will tip an extra 2% of equity for every 50% increase in the share price. Plus, the co-founders have each made voluntary commitments to pledge 80% of their equity in Anthropic to philanthropic causes.

That might sound a little earnest for the capitalists among us. Especially given the oodles of cash Anthropic needs to keep running at its current pace.

Because one of the secrets to the AI firm’s success was the co-founders’ recognition that scale matters. It turns out that you can write the most beautiful code to help a machine learn, but if you throw enough computing power at something, it will get there faster, even if it’s partly by brute force.

That’s becoming very expensive.

Sure, Anthropic has cut its cost of compute and infrastructure as a percentage of revenue to 55% in the June six-month period, from 201% a year earlier, but the company has largely leased that capability from others – including direct rivals such as SpaceX.

Our shadows taller than our souls

Those leases are hungry mouths to feed, with Anthropic signing up to almost US$660 billion over the coming decade in arrangements where it’s on the hook to pay, irrespective of whether it takes up the processing power. There are good reasons why UK neocloud firm Nscale is pushing ahead with its own US$35 billion IPO on the New York Stock Exchange.

Unsurprisingly, Anthropic is building its own data centre capabilities, with the first site open this year and more in the works.

And that will be a tough road to navigate for Anthropic, with data centres facing delays as local communities around the world start agitating against higher power prices and hits to water supplies.

The question being whether Anthropic will have to cede ground on its ideals if they run up against the economic reality of building data centres at scale.

That’s one area where New Zealand could potentially shine.

Sure, we’ve already got our naysayers pushing back against the hyperscalers over fears that the introduction of data centres will strain the national grid and lines companies, compound the dry-year risk to hydro dams, and divert water into industrial operations.

But that discounts the simple fact that our power companies are embarking on a major generation development programme and parts of the country have just the right mix of renewable electricity and water that data centres need.

Done right, the nation could effectively export its electrons to the world through the mammoth AI infrastructure build, in much the same way that we do with the aluminium produced at the similarly controversial Tiwai Point smelter.  

And for those of a more sensitive disposition, who better to partner with than a company that touts itself as a benign custodian of AI – and recognises that chasing a short-term buck could very well imperil the world?

Image from Solen Feyissa on Unsplash.