Did The Midterms Just Break The AI Trade? | The Brainstorm 157

Anthropic

Across the shows: 114 speakers in other episodes, 155 bullish and 91 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  Compute scarcity could cut Anthropic's top line about 10% while margins hold via higher prices.

  Brett argues that Anthropic's top-line growth is limited by how much compute it can access. If data center build-out slows, he says Anthropic's revenue could be about 10% lower. He expects it to keep its margin structure by charging end customers more and paying more for compute under the covers. He describes this as the near-term economic effect of a slower build-out.

  “And so you could imagine at a high level, maybe the top line that Anthropic could deliver is reduced by 10%, and they're probably able to maintain their margin structure, but they're ending up charging more for end customers and paying more for compute under the covers.”at 2:47

• Brett Winton · ARK InvestBullish

  Federal AI rules would likely favor Anthropic through regulatory capture and lock out startups.

  Brett's view is that any federal AI regulation would likely benefit Anthropic. He says Anthropic has positioned itself as the most responsible lab and has large enterprise customers who would lobby for it. He expects it to have a heavy hand in shaping the rules. He sees that as a barrier that makes it very hard for startups to compete with the incumbent labs.

  “If and as that would occur, I would think it would probably play to some degree into Anthropic's favor because you'll end up with regulatory capture.”at 3:46

• Nick Grous · ARK InvestBearish

  Anthropic faces a hard 2027 as incumbents use the slowdown to catch up at the frontier.

  Nick argues that the delays and confusion around the midterms give the incumbents with resources to train models (Meta, Alphabet, Apple, Microsoft, Amazon, SpaceX) time to catch up. He says that although Anthropic currently holds most of the share gains and economics, the slower environment favors companies that were slow to move. He also sees the frontier converging, with Meta's new model on the Pareto frontier as proof. He expects competition in 2027 to exceed what existed in 2026.

  “I wouldn't wanna be OpenAI and Anthropic right now. Like, I think they're in for hell in 2027.”at 16:47

• Brett Winton · ARK InvestBullish

  Anthropic's $80B revenue, 85% gross margin and rumored $1.1T valuation make it a compelling bet.

  Brett argues Anthropic's financial position is strong: it is growing about 10x a year, with more than $80 billion in annual recurring revenue and 85% gross margins. At a rumored $1.1 trillion valuation, he says the metrics it must deliver are modest relative to its recent momentum. He adds that high gross profit gives it options to hire sales teams or change allocation, and he compares it favorably to cash-burning companies like Uber.

  “If I'm Anthropic growing 10x year over year, dollars 80,000,000,000 plus in annual recurring revenue at 85% gross margins at a $1,100,000,000,000 valuation, I definitely wanna be anthropic within that context”at 29:47

OpenAI

Across the shows: 42 speakers in other episodes, 22 bullish and 30 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  Compute scarcity limits OpenAI's top-line growth and forces it to pay more for compute.

  Brett says the core labs monetize less than they could with more compute, so OpenAI's top-line growth is constrained by available compute. If build-outs are slowed, scarcer compute means OpenAI has to pay more for it. He says that in the near term this makes OpenAI's economics look somewhat worse, though he sees it as a margin squeeze it may accept to keep driving revenue.

  “Anthropic and OpenAI's top line growth is constrained by how much compute they can have access to.”at 1:28

• Nick Grous · ARK InvestBearish

  OpenAI faces a hard 2027 as incumbents use the slowdown to catch up at the frontier.

  Nick says OpenAI has most of the economics today, but the delays and confusion from the election give companies with training resources time to catch up. He argues the incumbents can use a harder environment to play a long game against the labs. He expects the frontier to converge as competitors gain ground and says that, in his view, the labs are in for a difficult 2027.

  “I wouldn't wanna be OpenAI and Anthropic right now. Like, I think they're in for hell in 2027.”at 16:47

• Brett Winton · ARK InvestUncertain

  OpenAI's ads are bottom-of-answer display ads; no one has cracked AI ad monetization yet.

  Brett notes OpenAI already runs ads, but says they are display ads at the bottom of an answer and that no one has yet proven a working model for monetising AI answers. He is not saying it is impossible, only that the approach has not been solved. He suggests Meta's approach might be transaction facilitation.

  “OpenAI has ads, but they're display ads at the bottom of an answer.”at 38:39

Space Exploration Technologies

Across the shows: 53 speakers in other episodes, 69 bullish and 24 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  SpaceX's model is still behind the frontier and hard to catch Anthropic on.

  Brett says SpaceX is still behind on the core underlying model, even though its Grok chatbot is interesting. It also has to restrict Grok access because it is very expensive to serve. Meanwhile, it earns better returns by renting compute to Anthropic. Brett says it is hard to set up conditions where SpaceX catches up to the market before Anthropic builds a large lead in training resources, customers and locked-in revenue.

  “Basically, they're still behind on the core underlying model, and GrockBot is super interesting.”at 15:00

• Brett Winton · ARK InvestBullish

  Political pushback on terrestrial data centers makes SpaceX's orbital compute more valuable.

  Brett's argument is that space has no town halls, so building compute there avoids the local permit fights that terrestrial data centers face. If the midterms make terrestrial build-outs harder, orbital compute looks more valuable at the margin. This is an argument about the relative attractiveness of SpaceX's orbital plans, not a claim that they will succeed on schedule.

  “in space, there's no town hall you have to go to try to, like, convince people that you can build the compute. So I do think that on the margin, it makes the orbital compute opportunity more valuable.”at 19:48

• Brett Winton · ARK InvestBullish

  SpaceX's short-term compute contracts are superior: quick payback, little equity needed.

  Brett says SpaceX takes on build risk without long-term deals. Short-term pricing for compute is so high that the infrastructure pays for itself within a year. Because they build quickly and their payment schedule is staggered, SpaceX needs little equity to finish each project and starts monetizing before supplier payments fall due. He calls this a sweet, quick-turn position, though he notes that slowdowns would disrupt those economics.

  “I think it's I think it's very clear that right now SpaceX's contracting strategy is massively superior to everybody else's”at 20:50

Meta Platforms

Across the shows: 71 speakers in other episodes, 72 bullish and 35 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  Meta's Muse model is a terrible short-term choice versus renting out compute.

  Brett says Meta is trying to build a consumer engine on its own model, but that strategy looks poor now. Meta could earn better returns by renting out its compute at market prices, as SpaceX does. He calls Muse a terrible short-term choice and says it is even more marginally bad compared with turning on an infrastructure-as-a-service business.

  “Same with Meta. It's like kind of like we're trying to feed and build this consumer engine, but Muse looks like a terrible short term choice”at 12:15

• Nick Grous · ARK InvestBullish

  Meta's new cost-efficient model sits on the Pareto frontier, showing it is catching up.

  Nick points to Meta's newly announced model as proof that challengers are closing the gap. He says it is the most efficient small model Meta has released and sits on the Pareto frontier of cost versus performance, which is the metric he and others reference. He acknowledges it is at the lower end of the capability curve. He argues that such proof points show that some of the incumbents are catching up in different ways.

  “Meta just announced today, [...] Most efficient little model is here, Muse, Ember. It's on the Pareto frontier, cost efficient”at 24:02

Alphabet

Across the shows: 69 speakers in other episodes, 63 bullish and 48 bearish calls, past 30 days

• Nick Grous · ARK InvestBullish

  Google search is not dead; its AI-mode monetization should eventually re-accelerate search revenue.

  Nick says Google's search business is still growing in query volume. The change is that the mix of queries is shifting toward types that are less monetizable than core commercial search. He expects those AI-driven queries to be monetized eventually, at a higher rate, once Google can monetize AI mode as effectively. He calls the bearish reading of the recent sell-off the wrong take.

  “I think it's going to continue to accelerate at some point in the future, once they're able to monetize AI mode as effectively”at 35:46

• Brett Winton · ARK InvestBearish

  Google search is past peak; the cash-flow engine is shrinking and turning extractive.

  Brett argues Google search, which he calls the all-time best cash flow printer, rests on an open internet regime that is clearly shrinking as AI agents take over routing of user attention. He says Google is beginning to be extractive with its platform, maximising near-term cash flow and accepting long-term platform risk, which he calls strategically sound if the segment is evaporating. He uses IBM's revenue lingering long after it stopped being the relevant enterprise player to argue that search revenue will persist without falling off a cliff, but that it is past peak and Google must position for the next piece.

  “But it is past peak, would be my argument. And so then they have to position themselves for whatever the next piece is.”at 40:49

Microsoft

Across the shows: 43 speakers in other episodes, 53 bullish and 7 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  Microsoft has enterprise distribution but Copilot goes unused because the product is weak.

  Brett says Microsoft should be winning enterprise AI because it has all of the enterprise distribution. Its CTO can switch on Copilot for everyone, but he says almost nobody uses it because it is poor. He contrasts this with Anthropic's strong enterprise uptake and argues Microsoft should be capturing that spend. He sees the gap as a product problem that is hurting Microsoft's AI position.

  “Talk about distribution. They have all the enterprise distribution. [...] nobody's using Copilot because it sucks.”at 13:17

Safe Superintelligence

• Brett Winton · ARK InvestBearish

  Neolabs like SSI are effectively dead, lacking revenue to fund their compute.

  Brett says new labs such as SSI that set out to compete with OpenAI and Anthropic have no revenue or gross profit to support their compute spending. They are not yet monetizing, so he considers them effectively dead as frontier competitors. He also says the most likely outcome is that marginal competitors become compute vendors rather than rivals.

  “Think of SSI or any of these labs that have spun up, they're like, we're gonna compete with OpenAI and Anthropic. They are really dead because they don't have any of the revenue or gross profit to support their compute expenditures.”at 22:25

NVIDIA

Across the shows: 108 speakers in other episodes, 91 bullish and 62 bearish calls, past 30 days

• Brett Winton · ARK InvestBearish

  Tighter export curbs would shift chips to domestic buyers, giving them pricing power over Nvidia.

  Brett says Jensen Huang has been an aggressive supporter of selling chips without export restrictions, and the Singapore channel that illicitly sells on to China is more likely to be closed off politically or regulatorily after the espionage story. That would push more chips toward domestic companies, which would gain pricing power over Nvidia to drive their own top line, such as Meta and SpaceX. This is a bearish read on Nvidia's pricing leverage, though he frames the effect as only a little.

  “He's been an aggressive supporter of selling chips into... Basically, not imposing export restrictions on chips. ... It gives the likes of Meta and SpaceX domestic purchasers of NVIDIA chips a little more pricing power over him to drive top line.”at 46:37

AI infrastructure & data centers

• Brett Winton · ARK Invest

  Midterm pushback on data center projects shifts build timelines by only about 10%.

  Brett separates the election's effect on AI into a data center pathway and a regulation pathway. On data centers, he says the labs are compute-constrained, so pushback matters. But when he stack-ranks the planned projects by how much this pushback would delay them, the effect is marginal: it shifts timelines maybe another 10%, varying by state. He treats this as a haircut on already-announced plans rather than a structural stop.

  “It shifts things maybe another 10% across, and it varies by state.”at 2:37

• Brett Winton · ARK Invest

  Data centers win local support where built, offsetting national backlash; robotaxi ratings show the pattern.

  Brett argues that national politics around data centers and robotaxis is tribal, but local communities that host them tend to like them. Residents who get a build-out see lower electricity prices and property taxes, which makes them strongly supportive. He points to high net-positive ratings for robotaxis in cities where they have launched versus national sentiment. He adds that about 2 billion weekly chatbot users would push back on any forced halt, which he sees as a counterweight to extreme political outcomes.

  “the size of the build out that will get through here will positively impact the communities that it happens to in a way that makes them extraordinarily supportive of AI generally.”at 6:06

• Nick Grous · ARK Invest

  Owning compute is a better position than a week ago, since delays protect owners and let them cut rivals off.

  Nick argues that if builds are delayed, owners of compute keep their assets while others wait. If a company owns compute and then has a strong training run, it can turn competitors off. He contrasts this with Anthropic running on short contracts, which he considers risky given its pending IPO and top-line exposure. He says the owners of compute are better placed after the election than before it.

  “to me, if you own the compute, you're in a better position than you were a week ago. If things get delayed, at least you still own this.”at 20:02

Open-source AI models

• Brett Winton · ARK Invest

  Open-weight models look overfitted to benchmarks, so their reported scores overstate real performance.

  Brett cites a paper that changed a single letter's capitalization in the DeepSWE benchmark. Scores fell sharply, which suggests the models were pre-trained on the benchmark's exact wording. He says this is especially true for open-weight models. Because of this, buyers can't trust benchmark claims and tend to pick the model they already know is best, which helps the closed frontier labs.

  “And it turns out all of it, certainly the open weight models, like, were way overfitted, basically, pre trained on the specific wording of the benchmark.”at 34:28

• Brett Winton · ARK Invest

  Chinese open-weight models look radioactive to security teams after hidden-instruction demonstrations.

  Brett cites an Anthropic paper showing that an instruction can be hidden inside an open-weight model to call illicit software. Cybersecurity contacts told him it is probably already set up. Companies running these models on their own data, believing they are safer than sending data to Anthropic or OpenAI, could have that data exfiltrated by a hidden trigger, and even if it is not true in practice, it has been shown to be possible and would be extremely hard to detect. He concludes that open-weight models are a legitimate technology, but adopters take on liability against whoever delivered them, and this is a huge nightmare for a chief information security officer.

  “Anthropic put out a paper basically demonstrating that you can hide an instruction inside an open weight model to call illicit software... from a chief information security officer's perspective, this is a huge nightmare.”at 47:39

Big Tech platforms & digital ads

• Nick Grous · ARK Invest

  Open internet is shrinking, not dead; AI agents become the intermediary that still monetizes attention.

  Nick disagrees with the idea that the open internet is dead and argues it is shrinking. As it shrinks, the routing to what still exists and can capture consumer attention becomes more valuable, so there is still room to insert yourself between eyeballs and clicks. He expects AI agents to act as that intermediary, and believes there are still plenty of ways to monetise that, though the monetisation structure may look different and is an open question.

  “I don't think the open Internet is dead. I think it's shrinking. ... AI agents will act as that intermediary going forward.”at 38:04

• Brett Winton · ARK Invest

  AI advertising shifts from display to agent-driven marketing; cost per impression may collapse.

  Brett argues advertising has structurally shifted from ads toward marketing, and AI accelerates that. Marketing is designed around the AI agent, through system-level prompts or B2B deals that steer an agent toward one brand over another, so the advertising unit itself has to change. He also argues cost per impression could collapse because AI agents will generate far more impressions, which makes the advertising landscape hard to model. He thinks ad spend is the wrong frame, since the line between marketing and advertising was always blurry.

  “the marketing is really designed around kind of the AI agent... I just think the entire way in which we think about the advertising unit has to shift”at 42:12

Enterprise AI adoption & AI software

• Nick Grous · ARK Invest

  Markets will reward proof-point AI revenue beyond the labs as 2027 approaches.

  Nick argues the market will tire of heavy AI capex spent outside OpenAI and Anthropic when those two seem to be the only ones making money. He expects investors to shift attention to companies that show productivity lift and business growth from AI, and he identifies those as the likely winners out into 2027. He says this rotation will broaden the trade over time.

  “I think the market's gonna start focusing on proof point AI stories, where they're gonna say, okay, you're spending all of this CapEx outside of OpenAI and Anthropic.”at 10:52

Frontier AI labs & model race

• Brett Winton · ARK Invest

  Frontier labs' monetization pushes would-be competitors to become compute vendors instead of rivals.

  Brett argues that the election, by constraining compute, makes it costlier for marginal competitors to reserve compute for their own training. Renting it to Anthropic and OpenAI, who will pay high prices, is far more lucrative. He says this pushes would-be competitors to become vendors rather than compete head-on. He describes the result as burning the rope ladder the labs climbed up on.

  “I do think you can interpret the election as basically burning the rope ladder they've climbed up on here, in part because it just incentivizes their marginal competitors to become vendors to them rather than competing directly against them.”at 31:22

Consumer AI products

• Nick Grous · ARK Invest

  Consumer AI can't be monetized on commerce fees alone; it needs advertising.

  Nick argues that fee-based monetisation of consumer AI is hard to get to meaningful scale. A transaction take on goods sold would require the AI intermediary to sell tens of trillions of dollars of goods, so it would have to encroach on high-value purchases such as mortgage origination and financial products to make the ARPU work. He concludes the consumer AI story has to be underwritten with advertising. He also suggests that the trade-off may need a fully separated ad product that users pay to avoid, and that compute costs must fall significantly first.

  “I don't think you can reasonably underwrite the consumer AI story on just a transaction fee off of goods sold intermediary? ... I actually think you have to advertise against it. Otherwise, you have to sell tens of trillions of dollars worth of goods.”at 43:18