Latest / Elon Musk Podcast / Alphabet's Risky Two Hundred Billion Anthropic Bet

Transcript

Alphabet's Risky Two Hundred Billion Anthropic Bet · Elon Musk Podcast

• 0:00
  Anthropic committed to spending $200 billion with Google Cloud
• 0:04
  for server capacity and infrastructure.
• 0:06
  You know, to put that in perspective, if you look at the
• 0:09
  total revenue backlog for Google Cloud in the latest reporting
• 0:11
  period, it nearly doubled to over $460 billion.
• 0:17
  So that single startup accounts for more than 40% of Alphabets
• 0:21
  entire contracted future cloud revenue.
• 0:24
  Which is just a wild concentration of risk.
• 0:26
  I mean, the obvious question hanging over those numbers is
• 0:29
  whether this huge infrastructure build out is a bulletproof
• 0:33
  strategy to own the artificial intelligence market or basically
• 0:37
  a House of Cards resting on the survival of one single startup.
• 0:40
  Yeah, it really forces you to look at the underlying mechanics
• 0:44
  here because you are essentially watching the world's most
• 0:47
  profitable advertising company quietly attempt to turn itself
• 0:51
  into a heavy industry, highly leveraged utility.
• 0:54
  And if you just glance at the recent performance data, the
• 0:56
  shift seems to be paying off, on paper at least.
• 0:59
  Google Clouds revenue jumped 63% to $20 billion.
• 1:04
  Yeah, and their operating income tripled to 6.6 billion.
• 1:07
  Those are exceptionally strong numbers for the reporting
• 1:09
  period. They are definitely, but the
• 1:11
  nature of the product there actually selling is
• 1:13
  fundamentally different now. I mean, Alphabet isn't just
• 1:16
  selling generalized cloud storage or basic website hosting
• 1:19
  anymore. Right, it's a completely
• 1:20
  different ball game. Exactly.
• 1:22
  They are locking in gigawatts of power and these huge allocations
• 1:27
  of their custom tensor processing units.
• 1:30
  The TP use. Right the TP use these are
• 1:33
  physical chips specifically tuned for Anthropics clod
• 1:36
  models. So breakdown the physical
• 1:39
  difference there, because you know, a server is a server in
• 1:41
  most people's minds. Well, a traditional data center
• 1:44
  is built for flexibility. You have racks of standard
• 1:47
  processors. If, say, an e-commerce client
• 1:51
  scales down their server usage, you can instantly reallocate
• 1:55
  that same compute power to a hospital network or a streaming
• 1:58
  service. Because it's generalized.
• 2:00
  Exactly. It is generalized math, but
• 2:02
  artificial intelligence requires specialized math.
• 2:06
  The physical architecture requires different power
• 2:08
  delivery, entirely different liquid cooling systems, and
• 2:11
  literally reinforce concrete floors just to handle the
• 2:15
  heavyweight of the server racks. So you are building physical
• 2:19
  infrastructure tuned to a very specific set of requirements,
• 2:23
  right? It is basically like a utility
• 2:25
  company building a bespoke multibillion dollar power plant
• 2:30
  exclusively for one giant factory.
• 2:32
  OK, I see. If the factory thrives, the
• 2:35
  utility prints money. If the factory stumbles, the
• 2:38
  utility is stuck with an empty plant that cannot easily be
• 2:40
  rewired to power residential neighborhood and that.
• 2:43
  Completely limits Google's flexibility.
• 2:45
  I mean their physical and infrastructure road map, their
• 2:47
  supply chain for securing these highly specific chips, and their
• 2:50
  long term power purchase agreements with local energy
• 2:53
  grids. They are now tightly tethered to
• 2:55
  1 customer's success. Yeah, custom AI accelerator and
• 2:58
  the specific data center architectures built around them
• 3:01
  are incredibly rigid. You are building the physical
• 3:03
  world around the specific needs of Anthropic.
• 3:06
  Which brings a very particular kind of risk into the equation.
• 3:10
  You have this highly capital intensive operation depending
• 3:13
  heavily on the commercial viability of a venture backed
• 3:16
  startup. Right.
• 3:17
  And the payment structure here has a certain, I don't know,
• 3:22
  irony to it. Alphabet is basically giving
• 3:25
  Anthropic the money to pay them back.
• 3:27
  Yeah, the mechanics of the deal create this totally closed loop.
• 3:30
  Right. So Alphabet committed to
• 3:32
  investing up to $40 billion in anthropic.
• 3:35
  Then Anthropic turns around and uses that money to commit to a
• 3:39
  $200 billion cloud spend with Alphabet.
• 3:43
  Which is just wild to think about.
• 3:45
  It is, and that guaranteed future spend in turn inflates
• 3:49
  Anthropic's private valuation to around $350 billion.
• 3:53
  Just think about the logic of that for a second.
• 3:54
  It's the corporate equivalent of giving your kid $50.00 to buy
• 3:58
  lemonade from your own lemonade stand and then bragging to the
• 4:01
  neighborhood about your incredible sales volume.
• 4:04
  Yeah, and you know talking up how much your kids business is
• 4:07
  worth now. The money you just moves from
• 4:08
  your left pocket to your right pocket.
• 4:09
  Then you get to record a transaction.
• 4:11
  Exactly. Which leads right into the
• 4:14
  accounting reality of Alphabets latest earnings report.
• 4:17
  And the headline numbers were huge.
• 4:19
  Alphabet reported $109.9 billion in total revenue, 62.6 billion
• 4:26
  in net income and earnings per share of $5.11.
• 4:30
  But if you actually dissect those numbers and look at the
• 4:32
  balance sheet, $37.7 billion of that net income was categorized
• 4:38
  as other income. So what does other income
• 4:40
  actually mean in this specific context?
• 4:43
  It is mostly unrealized mark to market paper gains on their
• 4:47
  investments. Like Anthropic?
• 4:48
  Yes, exactly like Anthropic. Mark to market is an accounting
• 4:52
  practice where you record the value of an asset based on its
• 4:55
  current market price. OK.
• 4:57
  Since Anthropic is a private company, its market price is
• 5:00
  determined by its latest funding rounds and revenue projections,
• 5:03
  projections that obviously went up because of the cloud contract
• 5:06
  Alphabet just gave them. So they.
• 5:07
  Are literally recording a profit based on the increased private
• 5:12
  valuation of the company they just funded.
• 5:14
  That's it. Alphabet gives Anthropic cash.
• 5:18
  Anthropic signs a contract to buy server time from Alphabet,
• 5:22
  right. That huge contract makes
• 5:24
  Anthropic look incredibly valuable on paper.
• 5:27
  Alphabet holds a stake in Anthropic.
• 5:29
  So Alphabet looks at their own balance sheet and says, well,
• 5:33
  our investment is now worth 10s of billions of dollars more and
• 5:37
  they record that increase as profit.
• 5:39
  Wow. And if you strip out those paper
• 5:41
  profits, the actual adjusted operating earnings per share is
• 5:45
  only $2.76. It changes the entire narrative
• 5:48
  of the earnings report. The headline numbers look like
• 5:50
  this massive operational beat signaling A booming business.
• 5:54
  Everyone cheers. Right, but a huge portion of the
• 5:56
  profit is just a circular accounting loop flattering the
• 5:59
  balance sheet. You fund the startup, the
• 6:01
  startup buys your servers. The startups revenue and
• 6:03
  valuation go up because of those compute resources and you record
• 6:06
  a non operating profit on your investment.
• 6:09
  And you know, it is perfectly legal and standard accounting
• 6:12
  practice, but it masks the actual cash generating power of
• 6:15
  their core operations. And while paper profits look
• 6:19
  great on an earnings report, you cannot use unrealized games to
• 6:24
  buy physical servers. No, you definitely cannot.
• 6:26
  You can't use paper gains to pour concrete for new data
• 6:30
  centers. You need actual cash for that,
• 6:32
  and the capital requirements right now are incredibly severe.
• 6:35
  They really are. Alphabet raised its total
• 6:38
  capital expenditure guidance for the current fiscal year to an
• 6:41
  astronomical 180 billion to $190 billion.
• 6:47
  That number is just hard to wrap your head around, right?
• 6:49
  And in just one recent reporting period, they spent 35 point $7
• 6:54
  billion on servers, networks and custom tech infrastructure.
• 6:59
  And this level of spending is actively compressing their free
• 7:01
  cash flow. It dropped 47%, down to $10.1
• 7:05
  billion. That squeeze on cash is also
• 7:08
  exacerbated by multibillion dollar acquisitions they have
• 7:10
  made recently. I mean, they purchased the
• 7:13
  cybersecurity firm Whiz and the clean energy developer
• 7:16
  Intersect. Yeah, and those require actual
• 7:18
  liquidity. You can't pay for those with
• 7:19
  paper gains. Exactly.
• 7:21
  So to cover that gap, Alphabote is turning to the debt markets.
• 7:25
  They recently issued $31.1 billion in senior unsecured
• 7:31
  notes. So just for anyone listening who
• 7:33
  might not be deep into bond markets, a senior unsecured note
• 7:36
  is essentially a corporate bond, right?
• 7:39
  It is a promise to pay the buyer back with interest.
• 7:42
  It is senior, meaning if the company goes bankrupt, these
• 7:45
  bond holders get paid before other creditors.
• 7:47
  But it's unsecured. Exactly.
• 7:49
  Meaning Alphabet did not put up any specific collateral like a
• 7:52
  building or patent to back alone.
• 7:55
  It is backed entirely by Alphabets general
• 7:57
  creditworthiness and their promise to pay.
• 7:59
  And that new issuance pushes their long term debt up to $77.5
• 8:05
  billion. And here is the craziest part.
• 8:08
  Some of these newly issued notes have maturity dates stretching
• 8:11
  out a full century. 100 years, yes, the financial logic there
• 8:15
  is highly questionable. Alphabet is fundamentally
• 8:18
  altering its historical structure as a cash generating
• 8:21
  engine. I mean, they operated for years
• 8:23
  with minimal debt and massive cash reserves.
• 8:25
  They were a company that generated so much cash from
• 8:27
  search advertising that they never really needed to borrow.
• 8:30
  Right, but now they're taking on decades long debt, and in some
• 8:34
  cases century long debt to finance custom AI hardware and
• 8:38
  servers that will depreciate and become entirely obsolete in
• 8:42
  three to five years. You're a matching 100 year
• 8:44
  liability against a three-year asset.
• 8:47
  Yeah, think about it like taking out a 30 year mortgage to buy a
• 8:50
  smartphone. That is a great way to put it.
• 8:52
  The physical asset you are purchasing with that borrowed
• 8:55
  money is going to be functionally useless in a few
• 8:58
  years due to the natural advancement of silicon chips.
• 9:01
  AI models require exponentially more compute power with every
• 9:05
  new generation. So the server's alphabet buys
• 9:07
  today with borrowed money will not even be capable of running
• 9:11
  the models and tropical developed three years from now.
• 9:14
  Exactly. They will have to buy entirely
• 9:16
  new servers requiring entirely new capital expenditures, while
• 9:20
  still paying interest on the century bonds they used to buy
• 9:23
  the first batch. And taking on massive amounts of
• 9:26
  debt is particularly risky right now because the global cost of
• 9:30
  capital is being squeezed by geopolitical conflicts.
• 9:34
  The macro environment is incredibly hostile to debt
• 9:37
  funded expansion. It really.
• 9:38
  Is, I mean, the global economy is currently dealing with an
• 9:41
  energy supply shock. The ongoing war in Iran and the
• 9:44
  blockade of the Strait of Hormuz have choked off 35% of the
• 9:49
  world's seaborne crude oil trade.
• 9:51
  Which is a huge chunk of global supply, and oil is trading well
• 9:55
  over $100 a barrel right now. Current projections point to it
• 9:59
  hitting $150.00 if the blockade persists.
• 10:02
  And energy shock of that magnitude changes the entire
• 10:05
  global economic math. It basically creates A
• 10:07
  stagflationary environment. Yeah, and stagflation is the
• 10:10
  absolute worst case scenario for central banks.
• 10:12
  It is a combination of stagnant economic growth and high
• 10:16
  inflation, right? When energy prices spike, the
• 10:18
  cost of manufacturing and transporting every single
• 10:21
  physical good on earth goes up. That drives up inflation across
• 10:24
  the board. And to fight that inflation,
• 10:26
  central banks are forced to keep interest rates high.
• 10:29
  Which ties directly back to Alphabet's new debt fueled AI
• 10:33
  strategy. Higher interest rates make
• 10:35
  servicing that $77.5 billion debt pile much more expensive.
• 10:40
  Exactly. When you rely on floating rate
• 10:42
  debt or you need to continuously issue new bonds to fund ongoing
• 10:47
  server purchases, a free interest rate environment just
• 10:50
  severely punishes your bottom line.
• 10:53
  And simultaneously, the energy shock makes powering their new
• 10:56
  GW scale data centers vastly more costly.
• 11:00
  Because they use so much power. Right, you have the cost of
• 11:02
  capital going up at the exact moment you need to borrow
• 11:04
  billions and you have the cost of electricity surging at the
• 11:08
  exact moment you are building energy hungry infrastructure.
• 11:11
  And we haven't even touched on the final squeeze here.
• 11:14
  Alphabet Score advertising business still accounts for 73%
• 11:17
  of their total revenue. Yeah, people forget that digital
• 11:20
  advertising is highly sensitive to the broader economic climate.
• 11:24
  If inflation and high energy costs force everyday consumers
• 11:27
  to pull back their spending, advertisers immediately pull
• 11:30
  back their budgets. It's a direct chain reaction.
• 11:33
  If a consumer is paying an extra $200 a month for gas and
• 11:37
  heating, they are not buying a new pair of sneakers.
• 11:39
  And if there are not buying sneakers, the shoe company has
• 11:43
  no reason to bid on Google search keywords or buy
• 11:46
  unskippable YouTube ads. Exactly.
• 11:49
  Advertisers cut their budgets when consumers stop buying.
• 11:52
  If the core search and YouTube ad revenues decelerate because
• 11:55
  of a global stagflationary environment, the free cash flow
• 11:58
  compression we discussed earlier gets even.
• 12:00
  Worse, so you are left with a highly leveraged balance sheet,
• 12:04
  massive ongoing capital expenditure requirements to buy
• 12:07
  servers that become obsolete every three years, surging
• 12:10
  electricity costs and a declining primary revenue
• 12:13
  stream. It's a precarious spot to be in.
• 12:15
  Alphabet is quietly transforming from a high margin digital
• 12:18
  advertising monopoly into a highly capital intensive debt
• 12:21
  carrying infrastructure company betting everything on an AI
• 12:25
  super cycle. The biggest question going
• 12:27
  forward isn't whether Alphabet has the engineering talent to
• 12:30
  build these massive data centers.
• 12:31
  They obviously do. The real question is what
• 12:34
  happens to that $460 billion backlog if the venture capital
• 12:39
  money funding the biggest clients ever stops flowing?
• 12:43
  If you're not subscribed yet, take a second and hit follow on
• 12:47
  whatever app you're using. It helps us keep making this.
• 12:49
  We appreciate you being here. Also, check out our YouTube
• 12:52
  channel for more business and tech updates.
• 12:53
  There's a link in the description.