Did you miss yesterday’s The AI Boost to S&P 500 Profitability?
David and I discussed it further with some good counterpoints to the notion that tech margins are set to decline:
- Compute demand growth could well keep surprising, boosting revenues faster than costs, including depreciation. So far, every demand forecast has been easily exceeded (see below).
- Agentic AI (agents to agents) and robotics could evolve so rapidly and surprise everybody.
- Compute supply could be curbed by physical bottlenecks (e.g. power, memory, manpower) but strong demand would then push prices/revenues up.
- Some Chinese LLMs have recently raised prices due to a demand/supply imbalance.
- US politics are a wild card with control of Congress up in the air, but strong demand could then push prices/revenues up.
- How productivity evolves is another wild card with uncertainty on the labor market, at least in the early years.
- Hyperscalers will eventually slow physical expansion and re-focus on the asset-light model. Timing unknown.
So interesting!
OpenAI expects to burn $280bn by 2030 AI start-up has projected deeply negative cash flows as it invests in infrastructure and faces price pressures
(…) The company expects negative free cash flow of $278bn over five years from 2026 to 2030 as it invests aggressively to expand its access to computing power, according to a recent presentation seen by the FT.
OpenAI expects its expenses will far outrun its revenues, which are projected to grow tenfold in the same period, from $36bn this year to $350bn in 2030.
In total the company anticipates booking $840bn in revenue between now and the end of 2030. (…)
The $852bn AI lab recently entered talks for a major new funding round, the FT reported this week. Backers approached the company to discuss investing at a $1.2tn valuation. OpenAI is pushing for a higher valuation, according to a person close to the company.
OpenAI’s ability to meet its vast funding needs is critical to a network of financial arrangements and hardware deals the company has built up as it seeks to lock in scarce computing power. Large tech groups from chipmaker Nvidia to Oracle and SoftBank’s data centre business depend heavily on contracts with OpenAI for their future revenues.
OpenAI forecasts spending about $856bn on computing power and infrastructure by the end of 2030, by far its largest single expense.
The massive projected cash burn testifies to the pressure on OpenAI’s business. Having raised $122bn in March, it is on track to exhaust that cash in 2028, the presentation suggests.
To support its growth, OpenAI needs to pour hundreds of billions of dollars into data centres and computing power to train and run its models. It has also slashed prices as it seeks to win business from US rival Anthropic and see off the threat from cheaper, “open-weight” models from China.
There are signs its heavy spending may have a pay-off, with new model releases leading to an increase in OpenAI’s annualised revenue of about 20 per cent in July.
The group forecast even deeper negative free cash flow of $305bn in a prior projection in May. (…)
- Investors warn Anthropic could struggle to sustain revenues post-IPO OpenAI’s resurgence, cheap rivals and safety fears test faith in durability of business model
Investors are questioning whether Anthropic can sustain its extraordinary growth as competition, price-sensitive customers and the potential for AI to destroy humanity cloud its blockbuster IPO.
The company’s annualised revenue soared to $65bn in July and its backers forecast it will exceed $120bn by year-end, making Anthropic the fastest-growing business of all time.
But considerable uncertainty remains over whether it can sustain that pace, reflected in the wide range of IPO valuations estimated by investors familiar with the group’s performance. (…)
Among the concerns for prospective investors is the resurgence of Anthropic’s chief rival, OpenAI. The ChatGPT maker has attracted a growing share of customer spending since releasing GPT 5.6 in July, overtaking Anthropic in weekly spend for the first time in more than two and a half years, according to data from model platform OpenRouter.
OpenAI now claims to have the best available model in Astra, released earlier this month. (…)
Both companies face a looming threat from “open” models, AI software whose parameters are publicly available and can be customised. These cheaper models, primarily produced today by Chinese companies such as DeepSeek and Moonshot and tech giants such as Meta, are approaching performance levels of more sophisticated “closed” frontier models from OpenAI and Anthropic and gaining market share.
“This is the first time we’re seeing real price competition between the labs,” said Eric Glyman, co-founder and co-chief executive of corporate payments company Ramp. Ramp and many of its customers switched between rival models depending on the task, using routers that make it easier to change providers and harder for labs to lock in customers, he added.
The company had slashed AI spend by 40 per cent as a result. (…)
Executives including Microsoft boss Satya Nadella have long argued that AI models will become “commoditised” as competition drives down prices and users opt for the best-value model.
Anthropic has been better at retaining customers than rivals. Aleh Tsyvinski, an economics professor at Yale who has analysed data from OpenRouter, found 22.5 per cent of Anthropic users are still using its models 12 months after first use, compared with around 13.2 per cent for OpenAI. (…)
Another fear is that advances in AI trigger a social crisis. Jacob Coxon, a departing Anthropic researcher, said this month that the “people building AI earnestly believe that it could kill us all by the end of the decade”. Anthropic chief Dario Amodei has since called for AI labs to slow the development of cutting-edge systems to ensure they are safe. (…)
There may be more mundane hurdles to growth:
- Texas Orders Data Center Permit Halt Pending Audit
- Brookfield’s Flatt Says AI ‘Slowing Down Anyway’ as Builders Can’t Keep Up
Brookfield Corp. Chief Executive Officer Bruce Flatt said the artificial-intelligence race is already slowing because developers can’t build enough infrastructure to keep pace with what AI companies are demanding. (…)
“We as an industry can’t build enough. We can’t even build a fraction of what everyone thinks they need,” he said at Brookfield’s annual investor day on Thursday. “So when they say they’re going to slow it down, it’s slowing down anyway because there is not enough compute to be able to deliver what everyone thinks they need.”
Brookfield has estimated that AI infrastructure will require more than $7 trillion in capital investment over the next decade.
Flatt said the slowdown is good because it “will bring more discipline into the system.”
But no real sign of a slowdown just yet:
The dominant message from compute suppliers over the past few weeks is remarkably uniform: demand for AI compute is still accelerating and now consistently exceeds supply — every major supplier that has spoken recently (NVIDIA, Broadcom, AMD, Micron, TSMC) reported that supply, not orders, is the binding constraint on growth.
During its recent earnings updates, Nvidia leadership explicitly stated that the company is heavily supply-constrained, noting that its entire supply chain is “running flat out” to match unprecedented demand. Management hinted that annual revenues, currently forecast up 70% (to about $673 billion!), could actually double if not for these extreme limits on what their suppliers can produce, implying potentially $100 billion+ of unmet demand in 2027. CFO Colette Kress said “incredibly, we are seeing demand acceleration, even at our scale”. Nvidia is boosting prices 17% in 2026.
On September 17, Jensen Huang said Nvidia is seeing broad-based investment across nearly every country it operates in.
Broadcom CEO Hock Tan recently forecast revenues doubling in 2027 and again in 2028 (to $230B!), stating that “our demand actually exceeds this outlook” for 2027, with fab capacity and advanced packaging, not orders, as the limiting factor.
Daniel Roberts, co-CEO of cloud provider IREN, argued that AI compute demand is structural and potentially insatiable. He noted that “every unit of supply you bring online actually feeds a multiple of that in additional demand” due to the massive processing needs of next-generation agentic AI systems.
Micron CEO Sanjay Mehrotra said data-center customers want roughly 50% more supply than Micron can commit to, that tight conditions should persist beyond calendar 2027, and that the company has “no line of sight” to memory supply catching up with demand. Management expects 2027 to be even tighter than 2026 since demand growth is outpacing industry-wide supply growth, with meaningful new supply not ramping until ~2028.
TSMC has reportedly told customers it will raise prices up to 10% from January 2027, with an extra 10–15% premium on AI-capacity orders above original forecasts, a classic signal of allocation under shortage.
Totally fascinating!
Maybe not so much to central banks dealing with excessive inflation: they face 3 realities and one major unknown: compute demand remains very, very strong, LLMs and hyperscalers are in a critical race for dominance sacrificing costs for speed and footprint, complex geopolitics will endure while the promises of potentially offsetting productivity remain uncertain, at least in their timing.
Chicago Fed President Austan Goolsbee noted on Monday that central banks aren’t supposed to raise rates in response to supply shocks unless they are persistent. “These days, when you see a large supply shock,” such as Covid-19, tariffs or oil, “it is usually more accurate to assume it will be persistent than to assume it will quickly go away,” he said. (WSJ)
Kevin Warsh: “In my meetings these last few weeks — in Jackson Hole, in Asheville at the G-20 meeting, which the U.S. hosted, and at a central bank conference in Basel — it was evident that most advanced economies are facing price pressures,” he said.
Amazingly, not a single FOMC participant thinks the economy can slow on its own:
(KKR)
Is there value in Treasuries?
Saudi Arabia Spent Months Trying to Bypass Hormuz. Now There’s No Way Around It. Threats to the Red Sea have shut the kingdom’s workaround
(…) Saudi Aramco, the country’s state-controlled oil giant, is loading more oil onto tankers in the Persian Gulf and then taking its chances sending them through the Strait of Hormuz. The company has told several of its customers in Asia—its most important oil market—to pick up shipments just outside the strait off the coast of Oman, people familiar with the matter said.
The risky attempts to run the gantlet through the Strait show how the U.S. and its Gulf allies are struggling to keep critical oil routes clear when Iran and its allies can severely disrupt them with missiles and drones. The efforts are bringing the Saudis head-to-head with Iran, which is attacking ships in the waterway in hopes of shutting it down. (…)
The U.S. military has managed to suppress Iranian attacks enough to get millions of barrels of oil through Hormuz each day. But attacks by the Iran-backed Houthi militants and from Iraq are cutting into the impact of that progress by shutting off other routes for supplies. (…)
The most recent scramble was triggered by drone attacks from Iraq earlier this month that shut the East-West pipeline.
Without it, Saudi Arabia is increasingly falling back on an elaborate system of loading crude at its Persian Gulf terminal of Ras Tanura, shuttling it through Hormuz, transferring it to ships off the coast of Oman for distribution to its customers on global markets, then going back for more.
In the past two weeks, around 2.4 million barrels of Saudi crude and condensate a day were again flowing through the Strait of Hormuz, according to ship tracker Kpler, a level last seen in early July before the Houthi militants announced a maritime blockade of Saudi Arabia and began attacking Saudi shipping in the Red Sea.
The four-week average daily Saudi crude oil loadings from inside the Persian Gulf rose to more than 2 million barrels last week, compared with virtually zero for most of June and fewer than 1 million barrels a day in July, according to data provider Vortexa.
Meanwhile, loadings from Red Sea terminals have dropped to zero since the attacks on the East-West pipeline earlier in September. Saudi Arabia had already cut loadings from its Yanbu port in August after the Houthi militants declared a blockade of Saudi Arabian ports and ships in late July. (…)
The attacks on the Red Sea workaround are complicating hopes that Gulf states can simply build their way out of dependence on Hormuz. U.S. Treasury Secretary Scott Bessent said this month that new pipelines would make the strait a “worthless piece of water” within two years as oil moves overland instead.
But on Sunday, Saad Sherida Al-Kaabi, the energy minister of Qatar, which before the war exported its copious amounts of liquefied natural gas via Hormuz, pushed back on the idea.
“This is completely wrong,” he said at the Qatar Economic Forum in New York. “This waterway carries trade in all products, not only oil and gas, to and from all its adjacent countries.”
From Windward:
An opening?
Iran says it has sent conditions to Washington through Qatari mediators for restarting negotiations aimed at ending the U.S.-Iran war. Iranian security chief Mohsen Rezaei said Tehran is now waiting for President Donald Trump’s response.
Rezaei said Iran’s conditions include an end to fighting on all fronts, the release of frozen Iranian assets and an end to the U.S. naval blockade. Qatar and Pakistan have been working to re-establish negotiations between the two sides. (…)
Former Pentagon official David Des Roches told Al Jazeera that Washington may initially see Iran’s terms as effectively demanding U.S. “surrender” rather than offering a workable compromise. However, he noted that Iran appears to have dropped an earlier demand for war reparations, which he viewed as movement from its previous position. (…) (@Benzinga)
- Trump presses Volodymyr Zelenskyy to stop hitting Russian refineries US president tells Ukrainian leader in phone call it is all about ‘diesel, diesel, diesel’, according to officials
(…) The call between Trump and Zelenskyy came hours after Ukraine hit Moscow with what Russian authorities said was the largest drone attack on the capital, setting a major oil refinery ablaze and killing at least two people and injuring 20 others.
Moscow mayor Sergei Sobyanin said air defence shot down 450 drones as they approached the capital. Asked to comment on Trump’s call with Zelenskyy, a White House official directed the FT to the president’s Truth Social posts in recent days.
On Monday he wrote that Russia had “lost control of its Diesel Oil Industry”, adding: “This ridiculous and never ending War with Ukraine must be ended.”
Russia has continued its deadly missile and drone strikes on Ukrainian targets, hitting energy infrastructure, petrol stations, commercial warehouses and other businesses.
US diesel prices hit a fresh record of $6.51 a gallon on Monday, up more than 70 per cent since the Middle East conflict broke out in February. (…)
Trump claimed last Monday that Ukraine and Russia had agreed to refrain from striking each other’s energy facilities while attributing the global rise in diesel prices to the war rather than his conflict with Iran. But Ukrainian officials told the FT then that the US president had got ahead of himself and that no ceasefire had been agreed. (…)
It is not clear whether the Trump administration has made similar requests to the Russian side to halt attacks on Ukraine’s energy infrastructure.
Russia has not commented on Trump’s efforts to secure an energy ceasefire. Vladimir Putin, Russia’s president, is unlikely to agree before the end of this coming winter, when his forces are planning a devastating aerial bombardment of Ukraine’s energy infrastructure, according to people briefed on his thinking. (…)
Zelenskyy said in a statement after the call that it had been an “important conversation” and that his meeting with Trump on the sidelines of the UN General Assembly this week “could change a lot”. His office said the meeting was expected to take place on Tuesday.
“There is diplomatic momentum,” he said without elaborating. (…)
Trump Looks to Sideline Canadian Potash With Belarusian Supply Deal
President Trump is signaling a potential agreement to purchase lower-priced potash from Belarus, a move that could challenge Canadian exporters.
In a Monday post on Truth Social, Trump said he is working on a “massive deal” aimed at lowering fertilizer costs for American farmers and ranchers. (…)
Belarus, a close ally of Russia, is another major global producer of the mineral. In his post, Trump said that the pricing for Belarusian potash would be “substantially less than we are currently paying to Canada.” (…)
FYI, here’s what “massive” really means:
The US imported 13 million tons of potash in 2025, about 11.5 million from Canada.
The EurAsia Daily:
Vladimir Orlovsky, Chairman of the State Customs Committee, said that ” now 85% of all Belarusian exports and imports for Belarusian enterprises are in the eastern direction. First of all, China and other Asian countries. Of course, and the Russian Federation”, — he said.
China buys most of Belarus potash exports.
Belarus is landlocked. To move its product across the ocean, Belarus must rely on longer, highly expensive overland shipping routes through Russia, wiping out any initial savings on the raw material. Lithuania’s rail system ended its potash transportation contract with state-run Belaruskali OAO in 2022.
Canada’s National Post:
Canada’s location allows it to deliver potash at prices about 35 per cent lower than “landlocked Belarus, so without a corresponding European move at Baltic ports, the U.S. sanctions relief announced by President Trump will do little to restore Belarus’ advantage,” Bloomberg Intelligence analyst Alexis Maxwell said.
Trump has sought to rebuild ties with Belarus and Lukashenko, a close ally of Russian President Vladimir Putin. The U.S. lifted more sanctions on Belarusian potash in March following the release of political prisoners, even while European Union restrictions remained in force, signalling a divergence of approaches toward Lukashenko amid the Russia-Ukraine conflict. The U.S. is pushing Ukraine to ease restrictions on imports of potash fertilizers from Belarus and has urged Kyiv to encourage other European allies to do the same. (…)
New US tariffs coming for Lithuania?
The Hidden Agenda Behind the AI Panic A mathematical model doesn’t rebel. Anthropomorphizing it helps obscure human errors and accountability.
By Gabriele Mazzini, a fellow at MIT Connection Science. He led the team that drafted the European Commission’s AI Act proposal.
(…) Software doesn’t rebel. A mathematical model possesses neither intent, malice nor the will to defy its creators, let alone extinguish our species. AI is a human artifact, engineered for profit.
When an agentic model in an evaluation sandbox connects to an unauthorized server or executes an exploit, it hasn’t staged a coup. It has tried to meet the human-defined objectives set out before it through a path its designers failed to constrain. (…)
That powerful experimental models were able to discover novel vulnerabilities and breach external systems isn’t a sign of a dangerous superintelligence but of human error or negligence. There is no sentient actor lurking in the weights to be reasoned with, feared or pacified. There are only human software engineers, product managers and corporate boards deciding which guardrails are worth the latency cost and which permissions can be skipped in the race to market.
Policymakers and voters need to resist AI exceptionalism. In any other discipline—from civil engineering to pharmaceuticals—courts and regulators treat a system failure as evidence of bad product design and inadequate safety testing. If an aircraft crashes, we focus on finding the engineering defect, correcting it, and enforcing established liability standards for the damage created.
By leaning on an anthropomorphic narrative, Silicon Valley attempts to repackage its specific human choices that led to experimental, powerful models behaving unexpectedly during tests as an existential peril. Elevating the issue to a cosmic scale leaves the public paralyzed and takes ordinary product accountability off the table. (…)
Governments need to recognize regulatory capture when it stares them in the face. Tech leaders’ strategy looks transparent: Alarm Washington and Brussels into creating a regime in which only trillion-dollar incumbents with fully staffed compliance and safety departments can legally operate. By sitting at the policymakers’ tables before anyone else, these companies can help draft rules digging an impassable moat protecting them from open-source developers and upstart competitors, domestic or international. The real danger is in further concentrating the tech industry into the hands of only a few companies with deep pockets. (…)
Europe has been here before. In an effort to lead global regulation and react to fears borne from ChatGPT, Europe rushed its landmark AI Act into law in 2024. Already the world’s most restrictive rulebook, the framework quickly proved too broad and complex to enforce. Stalled by implementation delays and concerns about European competitiveness, the EU postponed the law’s full rollout, leaving regulations uncertain.
AI should be regulated—risks exist and should be taken seriously. But governments need to act based on available evidence and verified facts, not corporate PR panic, the views of industry insiders, or the desire for quick political wins.
The greatest danger facing society isn’t that software will awaken and overthrow its human masters. It is that we will allow the creators of the software to abdicate human responsibility for the systems they choose to build and help them pull up the ladder to market access behind them.
Television Networks Suspend Some Coverage of Trump After Bans Decision comes after White House removed CNN from rotating group of television journalists who cover the president
Major television networks agreed on Monday to suspend a rotation of broadcast journalists who cover President Trump after the White House removed CNN from a planned assignment covering the president’s trip to New York.
“Effective today, the TV pool will not be covering events designated as pool coverage of the President,” Bryan Boughton, the Washington bureau chief at Fox News, who currently serves as TV pool chair, wrote in an email to pool members. “This follows the White House’s position preventing CNN from fulfilling its assigned pool duties.”
The White House press pool is a rotation of reporters, videographers, photographers and technicians who cover the president when he travels or when he conducts events in spaces that are too small for the entire press corps. Monday’s decision suspends—for now—the television press pool, which distributes video footage and reporting about the president to TV networks and other broadcast outlets.
“The public has a vital interest in receiving accurate, independent information about its government,” White House television pool members Fox News, ABC News, CBS News, CNN and NBC News said in a statement. “No administration should restrict a news organization because it objects to its reporting.”
Print and photo journalists haven’t suspended their pool coverage of the president, and conservative television news outlet Real America’s Voice was scheduled to travel with the president on Monday to New York ahead of the United Nations General Assembly.
The decision marks an escalation of tensions between the press and the Trump administration after the president announced last week that he would ban CNN, MS NOW and Politico from the White House. (…)
Trump signaled he was prepared to appeal any ruling in the outlets’ favor. “Virtually every story about me, or anything having to do with me, is negative, wrong and, in many cases, DANGEROUS for our Country! No matter how GREAT my Achievements, they trivialize and demean,” he wrote. Later Monday, he told reporters he wasn’t worried about the decision by networks not to carry some of his events live. (…)