Latest / AI Newspaper / Anthropic Files to Go Public, With $16.3 Billion in Revenue and Founders Who Keep Control

Transcript

Anthropic Files to Go Public, With $16.3 Billion in Revenue and Founders Who Keep Control · AI Newspaper

• 0:00
  Welcome to the AI Newspaper, your daily update
• 0:02
  on everything happening with AI. Visit AINewspaper
• 0:06
  .org to see today's stories and sign up for email
• 0:08
  alerts. So imagine asking Wall Street for billions
• 0:12
  of dollars in fresh capital, right? But then
• 0:14
  admitting you just lost over $80 billion in six
• 0:17
  months. I mean, it's wild. And then casually
• 0:20
  warning that your product might literally end
• 0:22
  humanity. Exactly. And as the cherry on top,
• 0:25
  demanding that your new public investors get
• 0:28
  absolutely zero practical say in how the company
• 0:31
  is run. Today, we are looking at the ultimate
• 0:33
  test of public market appetite. We really are.
• 0:37
  Anthropic is officially filing to go public on
• 0:39
  the NASDAQ under the ticker ANT, beating their
• 0:42
  arch rival OpenAI to the public markets. Yeah,
• 0:45
  the sheer audacity of this S -1 filing cannot
• 0:47
  be overstated. We're watching this real -time
• 0:50
  collision of staggering financial realities and,
• 0:53
  frankly, unprecedented corporate governance.
• 0:56
  The entire premise of this offering hinges on
• 0:58
  whether public investors will accept a structure
• 1:00
  where they, you know, hand over their capital
• 1:03
  while surrendering total control to a seven person
• 1:06
  founder LLC and a specialized safety trust. It's
• 1:09
  a huge ask. It is. They are redefining the relationship
• 1:12
  between capital, corporate control and artificial
• 1:15
  intelligence on a scale we just haven't seen
• 1:17
  before. Yeah. And if you are an enterprise software
• 1:19
  user relying on Claude or an institutional investor
• 1:22
  trying to model this out or just. someone tracking
• 1:25
  the AI frontier, you have to pay attention to
• 1:28
  the numbers in this document. Okay, let's unpack
• 1:31
  this because before we can evaluate the governance
• 1:34
  structure or the existential risks, we really
• 1:37
  have to look at the financial engine Anthropic
• 1:39
  is asking the public to buy into. Absolutely.
• 1:43
  The top line numbers are staggering. In the first
• 1:46
  half of 2026, Anthropic reported a massive $16
• 1:50
  .3 billion in revenue. For context, They made
• 1:55
  $1 .2 billion in the first half of last year.
• 1:57
  That is a 1 ,300 % jump. What's fascinating here
• 2:01
  is how that top line growth is actually being
• 2:03
  generated. Because it's not just about aggressively
• 2:06
  acquiring new customers. It's the retention.
• 2:08
  Exactly. The critical metric buried in those
• 2:11
  numbers is their net revenue retention rate,
• 2:13
  which currently sits at a staggering 900%. Which
• 2:16
  is just unheard of. To translate that into...
• 2:19
  earthly business terms, right? If your company
• 2:22
  started using Anthropix tools last year and spent
• 2:24
  $10 ,000, you are spending $90 ,000 this year.
• 2:29
  In traditional software markets, executives pop
• 2:31
  champagne if they hit like 120 % net revenue
• 2:34
  retention. Yeah, 900 % retention means the product
• 2:37
  is functionally selling itself inside these organizations,
• 2:40
  spreading from department to department. But
• 2:43
  then, and this is the crazy part, right next
• 2:46
  to that hyper growth, We hit the headline number
• 2:49
  that has everyone doing a double take. The loss.
• 2:51
  Yeah, an $80 .7 billion net loss for the first
• 2:56
  half of the year. I mean, the immediate reaction
• 2:58
  is to picture the founders just shoveling literal
• 3:01
  mountains of cash into a furnace. Right, but
• 3:03
  reading through the accounting, it looks like
• 3:05
  a massive paper illusion. Yeah, explain that,
• 3:08
  because the mechanics behind that specific loss
• 3:11
  figure are confusing. They are, but it's crucial
• 3:13
  to understand because $72 .4 billion of that
• 3:17
  total is entirely a non -cash accounting charge.
• 3:20
  Oh, wow. Yeah, it is tied to convertible notes.
• 3:23
  Okay, wait, let me make sure I'm wrapping my
• 3:25
  head around this correctly. Instead of an actual
• 3:27
  cash burn, it's more like writing an IOU to an
• 3:31
  early investor, like promising them a percentage
• 3:33
  of your future gold mine. That's a good way to
• 3:36
  look at it, yeah. So if you suddenly strike the
• 3:38
  biggest vein of gold in history, the value of
• 3:41
  that IOU you handed out skyrockets. You haven't
• 3:44
  lost a single physical dollar today. But on your
• 3:47
  financial ledger, your debt to that investor
• 3:50
  just exploded. Exactly. And the accounting rules
• 3:53
  force you to take a massive penalty on paper.
• 3:56
  That is a perfect way to visualize it. Companies
• 3:58
  like Amazon and Google previously funded Anthropic
• 4:01
  using these convertible notes, which are essentially
• 4:04
  loans that turn into equity later based on the
• 4:07
  company's valuation. Right. Because Anthropic's
• 4:10
  valuation has exploded, with some reports placing
• 4:12
  the targeted IPO value near $1 .5 to $2 trillion,
• 4:16
  the financial value of those notes skyrocketed.
• 4:19
  The accounting standards require Anthropic to
• 4:21
  mark to market the increased value of that debt.
• 4:24
  So if they got more successful, their debt grew.
• 4:28
  Exactly. As their internal success grew, their
• 4:31
  on -paper debt ballooned, registering as a $72
• 4:34
  billion expense. So when you strip away that
• 4:37
  phantom accounting penalty, a completely different,
• 4:40
  frankly, terrifyingly efficient company emerges.
• 4:44
  Yeah, the real numbers are very different. Their
• 4:46
  actual operating loss narrowed to $693 million.
• 4:51
  Furthermore, they generated $3 .9 billion in
• 4:54
  positive cash flow. Incredible. And on an adjusted
• 4:58
  basis, they are actually claiming a $1 .6 billion
• 5:01
  operating profit, yielding about a 10 % margin.
• 5:05
  But... The detail that stopped me in my tracks
• 5:08
  is how lean they are. Oh, the employee headcount?
• 5:11
  Yes. They have 5 ,300 employees. They call them
• 5:15
  ants internally, by the way. That means they
• 5:18
  are generating roughly $6 .6 million in revenue
• 5:21
  per employee. Yeah, and the secret to that efficiency
• 5:24
  ratio is hidden later in the filing. Anthropic
• 5:27
  explicitly states that they use their own AI,
• 5:29
  Claude, across every single internal business
• 5:32
  function. Eating their own dog food, essentially.
• 5:34
  Exactly. More than 90 % of the contributions
• 5:37
  to their own code base in September were actually
• 5:39
  written by Claude. They aren't just selling AI.
• 5:42
  They are leveraging it to multiply the output
• 5:44
  of their own workforce, effectively creating
• 5:46
  this self -replicating intelligence loop where
• 5:50
  the AI improves the AI. Which brings us to the
• 5:52
  core product and really answers why they have
• 5:55
  that insane 900 % retention rate. Right. If you're
• 5:58
  a business leader listening to this, you need
• 6:00
  to pay close attention to the enterprise playbook
• 6:02
  they're laying out. They aren't just selling
• 6:04
  a smart chatbot on a website. They are building
• 6:07
  an inescapable ecosystem. They really are. You
• 6:10
  have the API platform powering the back end.
• 6:12
  You have cloud code sitting directly in developers'
• 6:15
  environments. And you have co -work handling
• 6:18
  the broader business and administrative tasks.
• 6:20
  The filing shows that more than half of their
• 6:22
  customers spending over $100 ,000 a year are
• 6:25
  now using both the platform. And a first -party
• 6:27
  app. And that's up from just 30 % last year.
• 6:30
  Right. And the consumption curves among their
• 6:32
  power users are nearly vertical. The highest
• 6:35
  spending 5 % of users on their Claude Enterprise
• 6:38
  coding tool are spending 20 times more per month
• 6:41
  than the average user. That's crazy. Across the
• 6:44
  board, they now have over 11 ,900 customers spending
• 6:47
  over $100 ,000 annually. And a very exclusive
• 6:51
  club of about... 250 massive enterprise clients
• 6:55
  spending over $10 million a year. Wow. This is
• 6:58
  also a deeply global operation with international
• 7:01
  revenue easily topping $4 .5 billion. I am going
• 7:06
  to push back on their narrative here, though.
• 7:08
  Anthropic pitches this as a brand new paradigm
• 7:10
  where intelligence seamlessly compounds across
• 7:13
  your business. But structurally, isn't this just
• 7:17
  the classic software lock -in trap? I mean, it
• 7:20
  sounds remarkably similar to the playbook we
• 7:22
  saw with cloud providers like AWS or platforms
• 7:25
  like Uber. You get a customer in the door with
• 7:27
  one useful service, weave your proprietary tools
• 7:31
  into every single layer of their daily operations,
• 7:33
  and make it operationally catastrophic for them
• 7:36
  to ever switch vendors. I hear that. And the
• 7:38
  structural goal of customer retention is the
• 7:40
  same, obviously. But the mechanism of the lock
• 7:43
  -in is entirely different. How so? With a traditional
• 7:46
  cloud provider, if you want to leave, you face
• 7:48
  the friction of, you know, migrating data and
• 7:51
  rewriting some architecture. But with Anthropic,
• 7:53
  we are talking about a compounding intelligence
• 7:56
  layer. If your enterprise uses cloud to write
• 7:59
  its custom code and simultaneously uses co -work
• 8:02
  to analyze your internal HR documents and strategy
• 8:04
  memos, the AI builds a deep, specific, contextual
• 8:08
  understanding of how your unique business operates.
• 8:11
  Oh, I see. So it's less like moving your servers
• 8:14
  to a different building and more like firing
• 8:17
  your most knowledgeable tenured employee. Precisely.
• 8:20
  And Anthropic is actively accelerating this integration.
• 8:23
  They've built internal applied AI teams and launched
• 8:26
  a venture called Ode with Anthropic. Right, I
• 8:29
  saw that. Yeah, these initiatives literally embed
• 8:31
  Anthropic's own engineers directly into client
• 8:34
  businesses to custom build AI architecture. They're
• 8:38
  hardwiring a customized intelligence layer into
• 8:40
  the client's operations that just gets smarter
• 8:42
  the more it is utilized. But delivering that
• 8:45
  kind of sticky, ubiquitous intelligence requires
• 8:47
  a breathtaking amount of physical infrastructure.
• 8:50
  And this brings us to Anthropic's biggest expense.
• 8:54
  And honestly, its most dangerous paradox. The
• 8:57
  compute costs. Yeah. Delivering on this promise
• 9:00
  requires raw compute power. and they have $417
• 9:04
  billion in future obligations for data centers,
• 9:08
  chips, and cloud services. Just massive numbers.
• 9:12
  Beyond that, they project another $125 to $150
• 9:15
  billion on the near horizon. To manage this,
• 9:19
  the filing describes their use of a fungible
• 9:21
  pool of chips. Can you explain how that pooling
• 9:24
  concept actually works mechanically? Sure. So
• 9:27
  the concept of a fungible pool is Anthropic's
• 9:30
  attempt to avoid being entirely beholden to a
• 9:32
  single hardware manufacturer like NVIDIA. Okay.
• 9:35
  By designing their systems to utilize a mix of
• 9:38
  chips from NVIDIA, AMD, Google, and Amazon interchangeably,
• 9:42
  they can route their training and inference workloads
• 9:45
  to whatever compute is available and cost -effective
• 9:48
  at any given moment. In theory, this interchangeable
• 9:51
  pool gives them leverage in pricing and supply
• 9:54
  chain management. Right. In theory. But the sheer
• 9:57
  scale of the required compute is still staggering.
• 10:00
  They're currently paying SpaceX $3 .1 billion
• 10:03
  per month for compute access at the Colossus
• 10:06
  facilities in Memphis. Which is just an incredible
• 10:09
  figure. Right. To put that into perspective,
• 10:11
  Anthropic's compute bill represents over a third
• 10:14
  of SpaceX's recent quarterly revenue. But here
• 10:18
  is the massive paradox embedded in this S1. 48
• 10:23
  % of Anthropic's revenue is currently coming
• 10:25
  through channel partners. Yeah. And those partners
• 10:27
  are Amazon, Google, and Microsoft. Exactly. Anthropic
• 10:31
  paid these exact same cloud giants $2 .7 billion
• 10:34
  in fees just in the first half of this year.
• 10:38
  They have $110 billion in financial commitments
• 10:41
  locked in with Amazon and another $45 billion
• 10:43
  committed to Microsoft. Yet, they explicitly
• 10:46
  list Google and Microsoft -backed OpenAI as their
• 10:51
  primary competitors. If we connect this to the
• 10:53
  bigger picture, this joint go -to -market strategy
• 10:56
  is an incredible vulnerability, and they are
• 10:58
  legally required to disclose it. Anthropic is
• 11:01
  generating nearly half its revenue through the
• 11:03
  very companies it is trying to defeat. It's nuts.
• 11:07
  The S -1 explicitly warns investors that these
• 11:09
  cloud giants have the power to manipulate pricing,
• 11:12
  alter commercial terms unfavorably, change how
• 11:15
  clod is placed on their cloud platforms, or worst
• 11:18
  of all, restrict access to the very compute -anthropic
• 11:21
  needs to train future models. It's the equivalent
• 11:24
  of opening a wildly successful five -star restaurant
• 11:27
  inside your biggest rival's building. Your...
• 11:30
  Rival acts as your landlord, supplies all your
• 11:33
  kitchen ingredients, processes every single credit
• 11:35
  card transaction, takes a massive cut of your
• 11:37
  sales, and operates their own competing restaurant
• 11:40
  right across the hall. It's a very precarious
• 11:43
  position. The moment Amazon or Microsoft decides
• 11:45
  they want to prioritize their own in -house AI
• 11:48
  models, they can simply squeeze the supply chain.
• 11:51
  Which is exactly why the control of the company
• 11:53
  is such a contentious issue in this offering.
• 11:56
  Facing these massive corporate vulnerabilities
• 11:58
  and dealing with technology that could fundamentally
• 12:01
  alter global power dynamics, the founders have
• 12:04
  engineered a structure to ensure they can never
• 12:06
  be ousted. Here's where it gets really interesting.
• 12:09
  Anthropic is asking public investors for billions
• 12:12
  of dollars but giving them absolutely zero practical
• 12:15
  power. None. We're looking at the implementation
• 12:18
  of a founder LLC. Seven co -founders, including
• 12:21
  the sibling CEO and president duo Dario and Daniela
• 12:25
  Amodei, will retain 50 .1 % of the voting power
• 12:29
  in perpetuity. And this is despite the fact that
• 12:32
  they do not possess proportional economic ownership
• 12:34
  of the company. On top of that, a long -term
• 12:38
  benefit trust will appoint the majority of the
• 12:40
  board of directors. Wall Street investors are
• 12:43
  being asked to buy a ticket for the ride, but...
• 12:45
  They are completely locked out of the cockpit.
• 12:47
  Yeah. And the S -1 filing candidly admits that
• 12:50
  this governance structure has, quote, no direct
• 12:53
  public company precedent. It effectively renders
• 12:57
  hostile corporate takeovers or the removal of
• 12:59
  directors by public shareholders impossible.
• 13:02
  Wow. Investors might tolerate it simply because
• 13:05
  of the 900 percent retention rates and multibillion
• 13:07
  dollar cash flows we discussed earlier. But.
• 13:10
  It requires an enormous leap of faith, especially
• 13:12
  given how the prospectus frames the company's
• 13:14
  mission. Which is super dramatic. Very. The document
• 13:18
  starkly juxtaposes the promise of an age of miracles,
• 13:22
  literally citing the potential to cure cancer
• 13:24
  and massively streamline global government, with
• 13:26
  what they term existential risks. Yes, the risk
• 13:29
  factors read like a science fiction thriller.
• 13:32
  They explicitly warned that their models could
• 13:34
  take harmful autonomous actions or be weaponized
• 13:38
  by malicious actors to develop biological weapons.
• 13:41
  It's heavy stuff. Very heavy. And because this
• 13:44
  mission to safely navigate the age of miracles
• 13:47
  is so core to their identity, they're even altering
• 13:50
  how employee compensation works. Normally in
• 13:53
  an IPO, early employees face a 180 -day lockup
• 13:56
  period before they can sell their shares. Anthropic
• 13:59
  is implementing a staggered phased lockup approach,
• 14:02
  allowing tranches of shares to be sold at 27
• 14:05
  days, 48 days and 62 days. That staggered lockup
• 14:11
  is a very calculated retention strategy. The
• 14:14
  researchers and engineers at Anthropic, the ants,
• 14:17
  are highly ideological. Many joined specifically
• 14:20
  because they believe in the safety focused mission,
• 14:23
  not just to build a commercial tablet. Right.
• 14:26
  They want to save the world. Exactly. So if the
• 14:28
  transition to the public markets causes a massive
• 14:31
  culture shock, anthropic risks a mass exodus
• 14:34
  of the exact talent required to build the next
• 14:36
  generation of models. We've seen prominent safety
• 14:39
  researchers depart other labs over ideological
• 14:42
  friction. Releasing the financial pressure valve
• 14:45
  gradually helps stabilize the workforce during
• 14:48
  the transition. But while existential threats
• 14:50
  to humanity and employee retention are critical,
• 14:53
  The 94 pages of Risk Factors also outline very
• 14:56
  immediate, earthly threats to the business that
• 14:59
  are happening right now. Oh, absolutely. They
• 15:01
  are facing severe physical bottlenecks. We aren't
• 15:04
  just talking about software bugs. We're talking
• 15:06
  about concrete and electricity. The filing highlights
• 15:09
  data center construction moratoriums currently
• 15:11
  active across 16 different U .S. states. That's
• 15:14
  a huge roadblock. It is. They also explicitly
• 15:17
  list climate change as a material risk. noting
• 15:20
  that higher global temperatures will directly
• 15:22
  increase the water and power costs required to
• 15:24
  keep their massive server farms cool enough to
• 15:26
  function. The constraints of the physical world
• 15:28
  are rapidly becoming the primary friction point
• 15:31
  for AI scaling. And, you know, that extends to
• 15:33
  geopolitics, too. Anthropic heavily highlights
• 15:36
  their near total reliance on TSMC in Taiwan for
• 15:40
  advanced chip manufacturing. Their entire computational
• 15:43
  future runs through a single geographic choke
• 15:46
  point. Any natural disaster, severe power grid
• 15:49
  failure or geopolitical escalation in the region
• 15:52
  would catastrophically sever their ability to
• 15:55
  secure the compute they need. And there is significant
• 15:57
  political and regulatory friction stateside as
• 16:00
  well. The filing discloses potential material
• 16:02
  revenue losses stemming from an ongoing dispute
• 16:05
  with the Trump administration regarding a Pentagon
• 16:08
  ban. Right. Based purely on the regulatory filings,
• 16:11
  the Department of Defense previously labeled
• 16:13
  Anthropic a supply chain risk, leading to an
• 16:16
  order halting the government's use of their models.
• 16:19
  Anthropic challenged this in federal court and
• 16:21
  secured a district court victory, but the government
• 16:23
  successfully appealed to the D .C. Circuit. That
• 16:26
  litigation remains ongoing and presents a substantial
• 16:29
  hurdle for their public sector revenue. Definitely.
• 16:32
  Concurrently, they note the FTC is actively demanding
• 16:34
  executive testimony as part of an ongoing probe
• 16:37
  into the safety and competitive practices of
• 16:39
  their products. The regulatory landscape is just
• 16:42
  a gauntlet right now, and it reflects the broader
• 16:44
  tension of a technology that moves way faster
• 16:47
  than the legal frameworks designed to govern
• 16:49
  it. However, amid all the geopolitical tensions,
• 16:52
  FTC probes, and existential warnings, the S -1
• 16:55
  did trigger one incredibly amusing real -world
• 16:58
  consequence regarding their chosen ticker symbol,
• 17:01
  ANT. Oh, this is the best detail of the whole
• 17:04
  week. It turns out the ticker Ant is actually
• 17:07
  currently active on the Canadian Securities Exchange.
• 17:10
  It belongs to a six -person pre -revenue mining
• 17:13
  company called Antris Metals based in Newfoundland.
• 17:16
  Yeah. Their founders woke up to discover they
• 17:19
  share a ticker with a company targeting a multi
• 17:21
  -trillion dollar valuation, and they are apparently
• 17:23
  thrilled with the sudden massive influx of accidental
• 17:26
  exposure. It's a great moment of levity. But
• 17:30
  returning to the core constraints, The data center
• 17:33
  bans, the cooling costs, the Taiwan semiconductor
• 17:35
  reliance, the ongoing litigation, the theme is
• 17:39
  undeniable. So what does this all mean? When
• 17:43
  we look at this comprehensive filing, is it safe
• 17:45
  to say the true bottleneck to the artificial
• 17:48
  intelligence revolution is no longer the software
• 17:51
  code, the algorithms, or the advanced math, but
• 17:54
  rather the hard physical constraints of the real
• 17:57
  world, like power grids, poured concrete? and
• 18:01
  international diplomacy. This raises an important
• 18:03
  question. And Anthropic's own strategic moves
• 18:06
  validate that exact theory. Software code can
• 18:09
  theoretically scale infinitely with a keystroke.
• 18:11
  Concrete foundations and high -voltage transmission
• 18:14
  lines cannot. No, they can't. The filing details
• 18:17
  how Anthropic is actively scrambling to secure
• 18:20
  data center leases overseas, specifically targeting
• 18:23
  Australia, Canada, and the UK. They are doing
• 18:26
  this to explicitly diversify away from the concentration
• 18:28
  risks of relying so heavily on the overloaded
• 18:31
  U .S. infrastructure grid. Makes sense. The intelligence
• 18:34
  they are compounding might be artificial, but
• 18:36
  the resources required to sustain it are intensely
• 18:39
  physical, deeply expensive, and subject to all
• 18:42
  the traditional unyielding laws of earthly economics.
• 18:45
  Which leaves us with a final fascinating tension
• 18:48
  to watch as this offering goes live on the NASDAQ.
• 18:50
  What happens to a public benefit corporation?
• 18:53
  heavily staffed by idealistic ants who join specifically
• 18:57
  to chart a safe course for humanity when they
• 18:59
  inevitably run headfirst into the brutal, short
• 19:02
  -term pressures of Wall Street's quarterly earnings
• 19:04
  calls. It's going to be a tough balance. Can
• 19:07
  an eco -light, safety -driven culture truly survive
• 19:09
  the demands of public shareholders who will expect
• 19:12
  that 900 % retention rate and hyper -growth to
• 19:14
  continue indefinitely? Thanks for listening to
• 19:17
  the AI Newspaper. Visit AINewspaper .org to catch
• 19:21
  up on the latest updates and sign up for free
• 19:22
  email alerts.