Everything About AI
NextFin News - George Colony, the founder and chief executive of Forrester Research, says artificial-intelligence investment will keep flowing even as a fresh tech selloff rattles the market, arguing that the AI boom is not ending but entering a new phase in which custom chips challenge Nvidia's dominance and economic value migrates toward software. His remarks on Nov. 5 came as the S&P 500 fell 3.2% and the Nasdaq Composite dropped 3.8%, with the Philadelphia Semiconductor Index down 5.2% and Nvidia itself losing 5.4%.
The selloff was broad across the AI supply chain, and it was concentrated exactly where the AI capital spending has flowed. The S&P 500 closed at 7,433.85, down 244.87 points, while the tech-heavy Nasdaq Composite shed 962.37 points to 24,595.74. The semiconductor index, the clearest proxy for the AI hardware trade, fell 512.39 points, or 5.2%, to 9,293.92. Nvidia closed at $185.52, down $10.55 or 5.4%, and AMD fell 7.5% to $418.52.
Colony's central claim cuts against the panic. The boom is not over; it is changing shape. In his telling, three shifts define the new phase: custom silicon will erode Nvidia's pricing power, value will move up the stack toward software, and the transition will not be smooth. Companies that fail to adapt to AI computing, he warns, risk being left behind.
That framing turns the selloff from a verdict on AI into a rotation within it. If Colony is right, the investors getting hurt are not those who bet on AI, but those who bet on a single architecture, a single vendor, or a single layer of the stack. The question for the market is whether this is a cyclical air pocket in a secular boom or the first crack in the boom itself.
The Selloff Is a Capex Reckoning, Not an AI Rejection
The first-order read of the tape is simple: the market is repricing the return on AI capital. For more than two years, hyperscalers have funneled hundreds of billions into data centers, GPUs, and power infrastructure with the promise that AI revenue would eventually catch up to the spending. The selloff is the market asking for proof.
The evidence for unease is in the concentration of the damage. A 5.2% drop in semiconductors against a 3.2% drop in the broad index shows the pain is not macro; it is specific to the AI hardware complex. When chips fall roughly twice as hard as the market, the market is not doubting growth. It is doubting margins and payback periods.
But Colony's point is that capex skepticism has appeared in every major technology cycle. During the buildout of the internet, fiber-optic networks, and mobile data, the infrastructure layer was overbuilt first, then rationalized, then became the profitable backbone of the next wave. The companies that owned the customer relationship and the software layer captured the durable returns. The pure infrastructure providers fought over commoditized capacity.
This is the mechanism behind Colony's "new phase" language. The hardware buildout is the first act. It is capital-intensive, visible, and easy to finance, which is exactly why it gets overbuilt. The second act is software: the applications, agents, and workflows that turn the installed compute into revenue. That shift is slower, less visible, and harder to underwrite, which is why the market hates it.
The scale of the spending under scrutiny is enormous. The five largest U.S. cloud and AI infrastructure providers have collectively committed to spending between $660 billion and $900 billion on capital expenditure in 2026, nearly doubling 2025 levels, according to research houses tracking the hyperscalers. J.P. Morgan's global research team projects global AI capex to reach $5.5 trillion by 2030. Against numbers like those, a single-digit percentage cut in guidance is enough to wipe hundreds of billions from equity valuations. The market is not wrong to demand accountability.
Custom Chips and the Erosion of Nvidia's Pricing Power
Colony's most specific prediction is that custom chips will challenge Nvidia's dominance. This is not speculation about the distant future. It is already happening in the purchasing patterns of the largest AI spenders.
Google has deployed TPUs across its data centers for years. Amazon has built Trainium and Inferentia to reduce its dependence on external suppliers. Microsoft has developed its own Azure Maia AI accelerator and Graviton processors. These are not hobby projects. They are procurement strategies. Every custom chip a hyperscaler deploys is a GPU it does not have to buy at Nvidia's margin.
The mechanism is straightforward. Nvidia's valuation rests on two pillars: a technological lead that lets it charge premium prices, and a software moat, CUDA, that makes switching costly. Custom silicon attacks both. A hyperscaler that designs its own chip internalizes the margin and, because it controls the workload, can optimize the software stack for its own models rather than Nvidia's general-purpose architecture.
Colony has previously dated the break in Nvidia's stranglehold. In a Forrester analysis of the AI-driven technology cycle, he wrote:
The NVIDIA stranglehold will take another 12 to 15 months to break.
He still expected the broader hardware segment to grow 8% to 10% per year over the following five years. That is a crucial distinction: Nvidia's share can fall even as the market grows. A smaller slice of a much larger pie is still a large slice, but it is not the monopoly pricing that the current valuation assumes.
Nvidia is not standing still. The company has responded by investing in the ecosystem that runs on its platforms, including a $3.5 billion investment in MediaTek announced in August 2026, and by announcing partnerships with Apollo, BlackRock, Blackstone, Brookfield, Goldman Sachs and KKR to establish financing platforms intended to mobilize more than $500 billion of third-party capital for AI infrastructure over time. Those moves are designed to keep demand flowing even as customers build alternatives. They also shift risk: if compute becomes a financed asset class, Nvidia has an incentive to keep its chips deployed, not let them sit idle.
The Value Migration to Software
The second leg of Colony's thesis is that value migrates toward software. This is the historical pattern of every technology wave he has tracked. Colony has written that his career has spanned six major tech changes, from the minicomputer to the cloud, and that each "brought a wave of new providers and destroyed swaths of legacy companies." He calls AI the Seventh Wave.
The logic is economic. Hardware is a cost center that depreciates. Software is a revenue engine that compounds. Once the compute is installed, the marginal cost of running another AI workload approaches zero, and the value accrues to whoever owns the workflow, the data, and the customer.
Colony has put a number on one slice of this. He expects the technology services sector, which has been under pressure since 2023, to grow 5% to 6% per year from 2026 to 2030, driven by AI, up from a 3.6% estimate for 2025. Companies such as Cognizant and Capgemini will be hired to migrate enterprises away from legacy vendors and implement AI computing systems. That is the plumbing of the value migration: before software can capture value, someone has to rebuild the systems that the software replaces.
He has also argued that in five years, 70% of the revenue created by AI will be in private models rather than public ones, for one reason: trust. Enterprises will not run their most valuable workflows on shared public models. That favors the vendors who can deploy AI inside corporate firewalls, again a software and services story, not a chip story.
Cyclical or Structural: The Call
This is where Colony's thesis has to be stress-tested against the market's fear. Is the selloff cyclical, a mean-reverting air pocket in a secular boom, or structural, the first crack in a bubble that will not reflate on its own?
The evidence favors a cyclical read of the selloff layered on top of a structural shift in the industry. The cyclical leg is the capex digestion. Hyperscaler spending has raced ahead of monetization, and every quarter in which revenue growth lags spending growth will produce exactly this kind of volatility. That is mean-reverting: as capacity gets absorbed and AI features start charging, the ratio of revenue to capex improves.
The structural leg is the reorganization of the stack. Custom silicon, private models, and AI-native applications are not fads that reverse when sentiment turns. They are cost advantages and capability advantages that, once built, do not get un-built. A hyperscaler that has spent years optimizing its models for its own chips will not switch back to buying GPUs at a premium because the stock market had a bad week.
The risk is in the transition between the two. Colony himself warns that the transition will not be smooth. A cyclical downturn can kill companies with structural ideas if they run out of cash before the cycle turns. That is the danger zone for the AI startups and neo-clouds that financed their buildouts with debt. And the system is levered: prime-brokerage data from Goldman Sachs showed hedge-fund gross leverage globally at 285.2% of capital in November, up about 12 percentage points for the year and close to an all-time record. When positioning is that stretched, a repricing cascades faster than fundamentals would justify.
The Counter-Thesis: What If the Capex Never Pays Back?
The strongest case against Colony is the one the market is pricing right now: that the AI capex cycle is not a normal infrastructure overbuild but a genuine bubble, and that the returns will never materialize at the scale required.
The bears have a concrete argument. AI infrastructure spending is running at a pace that requires trillions of dollars of incremental revenue to justify. If AI turns out to be a productivity enhancer rather than a revenue generator, if it cuts costs but does not create new willingness to pay, then the installed base of GPUs becomes stranded capacity and the custom-chip buildout becomes a write-down rather than a margin improvement. In that scenario, Colony's "new phase" is not a rotation within a boom but the top of the boom.
This view is not fringe. It is the view embedded in a 3.8% single-day drop in the Nasdaq and a 5.2% drop in semiconductors. It is the view that has produced repeated warnings from technology executives themselves about irrationality in AI valuations. And it is reinforced by the leverage in the system: when hedge funds trade with near-record leverage, a repricing of AI multiples can cascade faster than fundamentals would justify.
The answer to the bear case rests on the adoption curve, not the spending curve. Cost-cutting AI is still AI revenue. If enterprises are willing to pay for agents that handle customer service, code generation, financial reconciliation, and logistics, the revenue base is enormous even if it arrives slower than the capex did. The question is timing, not direction. Colony's bet is that the direction holds and the timing gap is what the market is currently punishing.
The falsifying signal is specific: if hyperscaler AI revenue growth fails to exceed 25% year over year for two consecutive quarters while capital expenditure remains above 30% of revenue, the capex-recovery thesis is wrong and the selloff is structural, not cyclical. That is the metric that separates a digestion period from a dead end. Watch the quarterly cloud-earnings releases from the largest AI spenders. The ratio of AI revenue to AI capex is the single number that decides this argument.
What Comes Next
The practical implication of Colony's thesis is a rotation map, not a market call. If value is moving from hardware to software and from general-purpose GPUs to custom silicon, then the winners and losers are identifiable even if the index direction is not.
The exposed are the pure-play hardware vendors whose valuations assume Nvidia-like pricing power without Nvidia's moat, and the leveraged neo-clouds that financed data centers on the assumption of perpetually tight capacity. The beneficiaries are the hyperscalers that own both the custom silicon and the customer relationship, the software vendors that can embed AI into workflows enterprises already pay for, and the technology services firms that will be hired to execute the migration.
The forward look splits by horizon. In the short term, sentiment and liquidity dominate: with positioning crowded and leverage elevated, further volatility is likely, and every AI earnings print will be judged against an impossibly high bar. In the medium term, fundamentals take over: the companies that can show AI revenue growing faster than AI capex will be rewarded, and those that cannot will be punished regardless of the narrative. In the long term, the structural shift holds if enterprise adoption of private models and AI-native applications continues, and Colony's 70% private-model revenue forecast is the specific claim to track.
Three scenarios frame the next six months. The base case is a choppy consolidation in which hardware multiples compress while software leaders grind higher, exactly the rotation Colony describes. The upside case is an AI revenue surprise from a hyperscaler that proves monetization is accelerating, which would reflate the entire complex. The downside case is a capital-spending guide-down from a major spender, which would confirm the bear thesis and extend the selloff into a broader de-rating.
The signal to watch is not the index level. It is the AI revenue-to-capex ratio in the next round of cloud earnings. Everything else is noise.
The selloff is not a verdict on whether AI matters. It is a verdict on who captures the value, and Colony is betting the market is still looking at the wrong layer of the stack.
Explore more exclusive insights at nextfin.ai.
