Everything About AI
NextFin News - Goldman Sachs expects Anthropic's revenue to reach $685 billion by 2030, a forecast that asks investors to underwrite a roughly 15-fold jump from the AI lab's current annualized run rate and anchors one of the most aggressive valuation cases ever built for a company still preparing to list.
The projection, shared by the bank with investors ahead of Anthropic's initial public offering, sits far above the company's own disclosed trajectory. Anthropic's revenue run rate crossed $47 billion in May 2026, up from $9 billion at the end of 2025 and just $386 million in full-year 2024. The gap between the two numbers - $47 billion today versus $685 billion in five years - is the entire argument for the offering, and the entire reason to doubt it.
Anthropic filed its registration statement for a Nasdaq listing under the ticker ANT on October 26, 2026, with Goldman Sachs and Morgan Stanley leading a 22-bank syndicate. The filing opened the company's books at a moment when the AI trade had just absorbed a punishing week, and the numbers inside are now the baseline against which every bull and bear case will be measured.
The Forecast and the Gap It Must Cross
The bank's 2030 number is not a stray data point. It is the terminal value of a growth curve that Anthropic has already been tracing at an almost vertical angle. Revenue grew more than tenfold in each of the three years through early 2026, according to the company. The annualized run rate moved from $14 billion in February to $47 billion in May - an addition of roughly $33 billion in about ninety days, or about $2.5 billion of new annualized revenue every week.
Even that pace, however, does not get to $685 billion. To reach the bank's 2030 figure from a $47 billion May run rate, Anthropic would need to compound at roughly 80 percent annually for the next four and a half years. From the $16.29 billion of revenue the company actually recorded in the first half of 2026 - up 1,300 percent from $1.16 billion a year earlier - the required compound rate is higher still. This is growth that assumes no meaningful slowdown, no pricing compression, and no ceiling on the amount of compute the company can secure and sell.
The forecast also lands in a wider field of extrapolations. Anthropic has internally projected 2028 revenue of roughly $190 billion to $200 billion, according to people familiar with the company's financials. If that midpoint holds, the bank's 2030 number implies revenue roughly tripling again in just two years - from about $195 billion to $685 billion, a compound rate of about 87 percent. Every one of these figures is a bet that enterprise demand for frontier models continues to outrun the cost of supplying them.
There is a simpler way to state the stakes. At $685 billion of 2030 revenue, Anthropic would be larger by annual sales than almost any company that exists today. The forecast is not merely a growth estimate. It is a claim about the size of the AI economy itself.
The Engine: Who Is Buying, and Why the Retention Number Matters
The strongest evidence for the forecast is not the growth rate. It is the composition of the growth. Anthropic's customer base paying at least $100,000 on a trailing twelve-month basis grew from more than 950 in September 2025 to more than 11,900 by September 2026 - more than a twelvefold expansion in twelve months. Net revenue retention stood at approximately 900 percent as of the end of that period, a figure that implies existing customers are spending nine times more year over year rather than churning.
"We believe in America & we believe in helping to defend our country…[we] will continue to work to figure out a solution with [the Defense Department]," CEO Dario Amodei said at a Morgan Stanley technology conference, addressing the company's standoff with the Pentagon over safeguards on the use of its models, according to investor Brad Gerstner.
The retention figure is the linchpin of the bull case. A 900 percent net revenue retention rate means the installed base is not just staying - it is compounding from within. That is what converts a coding-tool adoption story into an enterprise-infrastructure story. Claude Code, the company's coding assistant, has been the sharpest edge of that adoption, and the early revenue concentration in developer tooling is now spreading across enterprise workflows.
But retention at that level also carries a warning. A customer spending nine times more this year than last is not yet a steady-state relationship. It is an adoption curve in its steepest phase. The question for 2030 is not whether customers will expand - they already are. It is whether expansion can persist once the easiest use cases are automated and the marginal dollar of AI spend has to prove a harder return.
The Cost of Staying at the Frontier
The second half of the Anthropic equation is the bill for the compute that makes the revenue possible. The company disclosed named compute contracts with Google, Amazon, Microsoft, Broadcom and others that add up to roughly $637 billion, one of the largest capital buildouts ever disclosed by a software company. The commitments include about $161 billion with Broadcom, $114.5 billion with Alphabet's Google, $110 billion with Amazon, and $44.9 billion with Microsoft. The company has also signaled it could spend as much as $750 billion on computing capacity in the coming years, including deals still being negotiated.
Those commitments are both the moat and the trap. They guarantee Anthropic access to the scarce resource - advanced chips and data-center capacity - that competitors are fighting over. They also mean that a large share of every revenue dollar is pre-committed before a single token is sold. If demand slows, the company cannot simply cut its way back to margin. The capacity payments keep coming.
The loss line tells the same story in accounting form. Anthropic reported a net loss of $80.7 billion in the first half of 2026. Roughly $72.4 billion of that was a non-cash charge reflecting changes in the fair value of convertible notes, derivative liabilities and SAFEs as the company's valuation rose, rather than cash spent on operations. Strip that out and the operating picture looks different: the company recorded an operating loss of $693 million for the first half, and $974 million of operating income in the second quarter of 2026 alone - its first profitable operating quarter in the six quarters shown in the filing.
That first operating profit is the detail the bank's forecast depends on. The path to $685 billion of revenue only creates shareholder value if the incremental margin on those dollars is not consumed by the next round of model training, the next cluster buildout, and the next talent war. Growth at any cost is not a strategy. It is a bet that cost will eventually stop mattering.
Cyclical Supercycle or Structural Regime? The Call That Decides the Stock
Here is the judgment the market must make, and the one that will decide whether the forecast is ambitious or absurd. Is the AI buildout a cyclical capex supercycle - the kind that overbuilds, clears inventory, and mean-reverts - or is it a structural regime shift in how computing is bought and sold?
The structural case is strong. Anthropic is not selling a product with a replacement cycle. It is selling a general-purpose input to economic activity - reasoning, coding, analysis - that embeds itself in workflows and expands as capability improves. The 900 percent retention rate is the fingerprint of a structural shift: customers are not buying once and re-evaluating. They are integrating the model into operations and scaling usage automatically. That is how electricity, databases, and cloud computing behaved in their own adoption decades. Once embedded, they did not shrink back.
The cyclical counter-case is equally familiar to anyone who has lived through a capex cycle. Every infrastructure boom - fiber optics in the 1990s, solar panels in the 2010s, cloud capacity in the 2010s - begins with demand that looks infinite, because at the start it is. Then capacity arrives, prices fall, and the survivors are the ones with the lowest cost per unit, not the ones with the best story. Anthropic's roughly $640 billion compute commitment is a bet that it will be the low-cost survivor. It is also a bet that the boom does not produce enough competing capacity to crush pricing before the commitment is paid down.
The honest answer is that both forces are present, and they operate on different time horizons. In the short run, the cyclical leg dominates: capacity is scarce, pricing is firm, and every lab with a model can grow. In the long run, the structural leg dominates: if AI becomes a permanent layer of the economy, the total addressable market expands regardless of who wins the pricing war. The bank's $685 billion number prices the structural outcome while the company still lives in the cyclical one. That gap is where the risk - and the opportunity - sits.
The Adversarial Case: What Would Break the Forecast
The strongest argument against the forecast does not question Anthropic's execution. It questions the multiple the market will pay when the company is no longer growing at three digits. A revenue target of $685 billion by 2030 is only half of a valuation. The other half is the price-to-sales ratio investors assign at that point. If Anthropic reaches $685 billion but the market prices it at 3 times sales rather than 8, the equity outcome is less than half of what the headline implies.
There is also the competition question. OpenAI, Alphabet, and a field of cheaper open-weight models are all racing down the same cost curve. If frontier capability commoditizes - if the difference between the best model and the cheapest model narrows to a level customers are willing to accept - pricing power migrates to the lowest-cost producer. Anthropic's compute commitments protect its supply. They do not protect its margin if a rival can deliver comparable output for less.
Customer concentration is the third fault line. Nearly half of Anthropic's sales last year were routed through Amazon and Google - the same two companies that are its largest suppliers, its distribution channel, and its direct competitors in frontier AI. That relationship works while all parties benefit from a growing pie. It becomes a risk the moment the pie's growth slows and the partners' own models start competing for the same enterprise dollars.
The falsifying signal is specific and observable. If Anthropic's net revenue retention falls below 300 percent for two consecutive quarters while revenue growth decelerates below 50 percent year over year, the structural-retention thesis is broken and the 2030 forecast should be treated as a ceiling rather than a base case. Retention is the early-warning sensor: it turns before revenue does.
What to Watch: Scenarios Across Time Horizons
The base case is that Anthropic grows into the forecast unevenly. Revenue compounds at 50 to 60 percent annually through 2028, driven by enterprise embedding and coding-tool adoption, then slows to 30 to 40 percent as the base widens. Operating margin expands gradually as the non-cash financing charges roll off and utilization of committed capacity improves. In this scenario, the company reaches public-market leadership in frontier AI but the stock underperforms the revenue story because the entry multiple already assumed perfection.
The upside case requires the structural thesis to hold fully. Enterprise AI spend proves to be a permanent, compounding layer of IT budgets rather than a discretionary experiment. Anthropic maintains pricing power, its retention stays above 500 percent, and the 2030 revenue target becomes achievable. In that world, the company becomes one of the defining businesses of the decade, and today's valuation looks like the last cheap entry point.
The downside case is a capex-cycle reckoning. Capacity arrives faster than demand, pricing compresses, and the roughly $640 billion commitment becomes a fixed-cost anchor. Revenue still grows - perhaps to $300 billion or $400 billion by 2030 - but margins disappoint and the multiple contracts. The company survives and remains strategically important. Shareholders who paid for the $685 billion outcome do not get it back.
Across all three scenarios, the near-term catalysts are the same: quarterly revenue prints against the $190 billion to $200 billion 2028 internal target, the trajectory of net revenue retention, and any update to the compute commitments. The IPO pricing itself will set the first hurdle - a debut above the $965 billion May private valuation prices in significant execution; a debut below it leaves room for the story to be proven wrong without immediate capital destruction.
The takeaway: Goldman's $685 billion forecast is not a prediction that Anthropic will execute flawlessly. It is a bet that AI demand is a structural regime rather than a cyclical boom - and every investor in this offering is making the same bet, whether they call it due diligence or faith.
Explore more exclusive insights at nextfin.ai.
