Pricing isn't a moat, running with eyes open, Fermi
Anchors are uplifting, breaking 2 hour marathon barrier, Adobe
Bank knows best
Fermi1, the ‘massive data center agglomeration meets nuclear power meets 5,000+ acres in Texas’ company, has fired its founder/CEO after raising $682 million in an IPO that went out at a $19 billion valuation. The company has failed to sign a client in the seven months since IPO. Bloomberg reports:
“Going from zero to a 17-gigawatt AI hyper campus in one leap before they had a single tenant or a dollar of project finance — that was a lot in hindsight,” said Timm Schneider, an energy analyst and founder of the Schneider Capital Group.
Fermi’s board has removed the CEO/co-founder, who is the company’s largest shareholder, and the CFO has departed the company.2
With the capital and public market valuation in hand, the company was missing several advantages that early stage startups that are searching for product market fit normally have. Chief among those is the ability to iterate on the offering and how it is presented to customers and investors.
What businesses in Fermi’s position normally have is a signed tenant. An anchor ready to plunk down as soon as possible. Instead, they looked more like a startup than an infrastructure provider.
In Fermi’s case, iteration wasn’t possible at startup speed. What they did have was land. Fermi had an agreement with the Texas Tech university system to use a massive parcel in the Texas panhandle. According to industry experts, the scope of the project is huge:
Spanning ~5,800 acres, the HyperGrid™ campus is a bold step to secure U.S. leadership in AI and energy innovation. This ambitious project integrates nuclear, natural gas, solar, wind, battery storage, and grid power to deliver up to 11 GW—enough to power ~10 million homes and 18 million sq ft of AI and data center infrastructure.
The value proposition was pretty well established going into the venture: make it easy for huge data centers to come in, build and get all the power they need. Fermi also had political connections built in, with former Texas governor Rick Perry as a (still at the company) co-founder. Perry had also been US Secretary of Energy which certainly helps with founder/market fit.
Fermi’s story skips a few typical steps in iteration.
Normally, if a business in its starting phases isn’t successful, you can change the story. Here, they didn’t. When you have such massive infrastructure pieces as the building blocks of your story, it makes it much harder to do.
If you can’t change the story, try changing the product. Fermi was being built on the premise that off-grid power fed by massive reactors and other energy sources was better than on-grid. This leads to issues with bank financing, as grid power projects are more palatable to banks. It doesn’t seem like there was any openness to modifying the planning.3
Unstated but probably real: banks don’t like projects that don’t have anchor tenants attached to them. Probably both with and without grid access.
Normally, startups keep iterating on plans and product, but we seem to have just blown past that phase. So if you can’t change the story or the product, try changing the team. This is the current phase. Bloomberg’s reporting on the current positioning:
The board had considered removing Neugebauer for a while, and his departure has prompted “positive feedback” from investors, potential tenants, vendors and other partners in the project.
I mean. This could be true. But it could be true for multiple reasons, before the board pats itself on the back. Finally you got rid of that bozo we didn’t like—or—we’re so glad you realized this whole thing wasn’t working.
If that doesn’t work either, maybe try to sell the business? We aren’t quite there yet, according to a company statement: “the Company firmly believes a sale is not in the best interest of its continued momentum.”
Keep up the momentum! Tenants bring with them the banks.
Run with your eyes open
Two runners last week broke the 2 hour mark for the first time in an official marathon race and both of them were wearing Adidas shoes. Sporting accomplishment aside, how Adidas ended up on top of the podium was a smart mix of watching competitors and running their own race.
But first, we need to go back to 2017 and 2019, when Nike had big fanfare and media races, first unsuccessfully and then successfully, to break the 2-hour mark. Nike’s races used a bag of tricks not allowed in official races like lasers, pacers, and hydration delivered to the runner via bicycle. What these events also had was the sort of marketing and promotion that drew interest from outside of the running community before the races were run.
There was a unique curiosity surrounding attempts outside of official races because almost as intriguing as the athletic performance were the decisions of how to best optimize the race for the runner to give them the best chance.4 Manufactured conditions only excite the imagination so much. So, the world waited while shoes and runners and nutrition5 improved until the point where the under 2-hour time could legitimately be broken in a normal race under normal conditions.
A big part of the 2019’s successful attempt by Nike was the use of a shoe, the AlphaFly prototype that wasn’t legal for use in competitions. Those shoes had 3 carbon fiber plates in them whereas competition rules only allow for having a single carbon fiber plate in the shoe. The reaction to these events by Adidas is instructive. According to a profile in Dezeen, a UK magazine:
Unlike most other competing brands Adidas did not adopt full-length carbon plates. Instead, it developed “energy rods” – five carbon-infused rods placed and curved to mirror the bones of the foot. The brand claimed this system allowed for a “more natural gait” than a large carbon plate.
Nike had success with the full carbon plates and kept going in that direction. Adidas took a different direction and also emphasized the importance of foam in the shoes.
For Adidas, it was nine years of iterating and tinkering on a product with a well established market, marathon running shoes, from the time they first saw their rivals at Nike organize a well funded attempt at the time barrier. The key turn was the breakthrough in foam, rather than the carbon plate that led to Nike’s early lead. From Dezeen:
This was largely achieved through the use of a newly created foam, named Lightstrike Pro Evo. According to the brand, this foam is 50 per cent lighter than [what was] previously used in the trainers. This was combined with integrated carbon elements to achieve the support needed.
Research is indicating that the foam ended up being the bigger driver of the speed increase than the carbon plate.6
So in the end, winning was a mix of watching the leader and learning carefully from what they were seeing, combined with betting on the right variable to innovate on to win.
It turned out to be good business. Adidas’ stock price has moved up about 10% and stayed there since the race. This has added over $2 billion in market cap to the shoe maker.7 A nice foam bounce?
Pricing isn’t a moat
Adobe has announced that they are going to be charging customers for AI agents only when they work. Charging based on the results that the AI agents get, rather than token usage, is better aligned with how customers think about what they get.
There are a lot of advantages to this approach. It signals pretty heavily to the customer that, hey, we are going to get to know your problems as a business, as a user, and we’re gonna address them directly. We’re gonna make sure that we’re always providing them value. There are a lot of positive aspects to that, and it also addresses head-on the concern about wasteful tokens. We’re gonna spend a lot of money on tokens. What’s really the value here? Can we connect it to a business case or something that drives outcomes for us as a business? And so Adobe, with this type of approach, addresses all of the customer concerns. It brings them more into the mode of task completion and getting whole jobs done.
What is still missing is the hard work to define more objective measures of how well the task is completed. For Adobe, if you think about their domain, whether it’s creation of creative assets or adapting creative assets, it’s not just whether something was done, but how well it was actually executed. Adobe (and others) will want to close the gap between “done” and “done well” if they want to price this way.
This is a significant gap for Adobe given the nature of their creative product suite. Visual quality and production value matter to their customers and are harder to define than workflows in other industries, like customer service, where either a case gets resolved or it doesn’t.
There’s another risk: thinking that pricing does more for you than it does. The Information writes in a story with background from Adobe president Anil Chakravarthy:
Like many software executives, he argues that prices for AI models will continue to fall over time, and customers will focus on measuring the value of their purchases.
That’s a bet that could pay off: If running AI models gets cheaper, software firms that charge based on results, rather than usage, would be in a good position.
And here’s where there’s just too much of a leap. There is a difference between things that are defensible and really sticky for customers versus what are just pricing formulations that can be pretty easily copied. Yes, it’s true that if there are more models, and they become more competitive, then pricing will drop. It is true that if you are charging on results, it’s different from charging directly on usage.
But let’s play it out. If Adobe’s competitors are offering the same types of jobs being completed within their software and using either their own proprietary models or wrapping around other models, they can compete on price just as easily if they’re outcome-based. Since the outcome is defined in terms of who the customer is and what their actions are, it can be replicated by a competitor to achieve the same outcome at the same or lower price.
If you look at another space, Sierra does outcome-based pricing for customer service, measuring outcomes like a successfully closed ticket with positive feedback from the customer. That’s not that hard to replicate if you are Fin (Intercom), also competing with outcome-based pricing in the AI customer service space.
Just as there can be price competition in usage-based pricing, it’s just as easy for there to be competition in outcome-based pricing as well. And this is where it gets potentially misleading for Adobe. They should be focused on an amazing customer experience and establishing their marketing and distribution channels for such products, and building the thing that’s great that people really want to use.
Pricing models don’t provide any moat.
Like these essays? I write a trio of essays on wide ranging topics delivered four times a week. Subscribe and forward it to your friends. Have feedback, thoughts to share or comments? Comment or reply.
Fermi was presumably named after Enrico Fermi, the ‘father of the nuclear age’, rather than Fermi’s paradox which is that aliens are highly likely to exist and yet there isn’t conclusive evidence of them.
Ex-CEO has sued the company for wrongful termination.
From Bloomberg: “The idea of giving data centers their own, dedicated power supply not dependent on the grid may sound tempting, but former US Department of Energy official Jigar Shah said banks don’t want to finance it. The grid, drawing power from many sources, is more reliable than a handful of expensive, on-site plants, he said.”
Would there be special pavement? Music? In ear coaching?
Jason Gay at the Wall Street Journal had a piece on the importance of sugar in the breaking of the milestone.
Adidas had campaigns ready to go to promote its winning shoe should the record be broken and they went out right away. Preparation was important but they didn’t jump the starter’s pistol.