ARK Invest just released 2026 big ideas: The great acceleration and AI.
Global markets are entering an unprecedented phase of technology funding.
- Data center systems investment could reach ~$1.4T by 2030 as inference costs collapse >99% and API demand surges.
- Nvidia leads large model throughput, AMD and Google rival it for small models, and custom ASICs should keep gaining share.
- Foundation models are becoming a consumer operating system, moving interactions to agents that plan, compare, and transact.
- Purchasing agents compress checkout to ~90 seconds and could mediate ~25% of online spend by 2030.
- AI search may reach 65% of queries by 2030 and push ad budgets toward AI interfaces.
- Companies recoup the cost of an AI subscription in about 0.5 day, then as those tools boost output, total software spending could grow to $3.4T to $13T.
Example, if a $30 per user tool saves 1 hour worth $60 on day 1, it has paid for itself. Over a month, the extra saved hours justify buying more AI apps.
- AI plus multiomics could cut drug development to ~8 years (instead of the usual ~12 to 15 years.) with ~4x lower costs and enable high value cures, including a ~$2.8T US TAM for ASCVD.
- Reusable rockets are driving launch costs toward <$100 per kg, expanding satellite connectivity, and opening a path to space AI compute (pages 80 to 82).
- Robotics, distributed energy, robotaxis, and autonomous logistics will lower transport, power, and delivery costs at global scale.
A household humanoid robot could add ~$62,000 to GDP each year by converting unpaid chores into market activity and freeing time for paid work, net of costs.
At 90M homes, impact approaches ~$6T, lifting US GDP growth toward 5%–6%.
At 90M homes, impact approaches ~$6T, lifting US GDP growth toward 5%–6%.
Past tech shifts raised long run GDP growth, and ARK expects another step up this decade with AI.
They argue new capital in robotaxis, AI data centers, and enterprise AI agents could add about 1.9 percentage points to annual real growth.
By 2030, ARK forecasts global real GDP growth near 7.3%, versus IMF at 3.1%.
They argue new capital in robotaxis, AI data centers, and enterprise AI agents could add about 1.9 percentage points to annual real growth.
By 2030, ARK forecasts global real GDP growth near 7.3%, versus IMF at 3.1%.
Quantum’s progress is too slow to be disruptive soon, per ARK
Even with faster scaling, cracking RSA-2048 and hitting compelling economics likely waits until the 2040s to 2060s.
Qubit counts and error rates would need to double and drop every 2 to 4 years, today’s pace is roughly every 4 years, so useful impact is decades out.
Even with faster scaling, cracking RSA-2048 and hitting compelling economics likely waits until the 2040s to 2060s.
Qubit counts and error rates would need to double and drop every 2 to 4 years, today’s pace is roughly every 4 years, so useful impact is decades out.
Inference price for strong models fell >99% from about $30 per 1M tokens in late 2024 to about $0.10 by mid 2025, with multiple releases pushing costs down each month.
As costs collapsed, usage exploded. Tokens served on OpenRouter rose roughly 25x from Dec 2024 to Jan 2026, showing surging developer and enterprise demand.
As costs collapsed, usage exploded. Tokens served on OpenRouter rose roughly 25x from Dec 2024 to Jan 2026, showing surging developer and enterprise demand.
Data center systems spending inflected after late 2022 i.e. after ChatGPT moment.
Annual investment reached ~$500B in 2025, nearly 2.5x the 2012–2023 average, and growth accelerated from a 5% CAGR to 29% CAGR.
ARK expects this curve to continue, with investment potentially tripling to ~$1.4T by 2030 as AI demand drives more buildout.
Annual investment reached ~$500B in 2025, nearly 2.5x the 2012–2023 average, and growth accelerated from a 5% CAGR to 29% CAGR.
ARK expects this curve to continue, with investment potentially tripling to ~$1.4T by 2030 as AI demand drives more buildout.
Hyperscalers are set to spend >$500B on capex in 2026, nearly 3x 2021, pushing tech and telecom capex as % of GDP to highs not seen since 1998.
Unlike the dot-com era, valuations are far lower, the big tech cohort sits well below the late 1990s P/E peaks, so spending is elevated while multiples are restrained.
Unlike the dot-com era, valuations are far lower, the big tech cohort sits well below the late 1990s P/E peaks, so spending is elevated while multiples are restrained.
Nvidia absolutely leads large model inference, GB200 shows the highest tokens per TCO dollar (total cost of ownership.), while AMD lags on big models.
For small models, AMD MI355X delivers more tokens per TCO dollar than Nvidia B200, so cost efficiency favors AMD in that segment.
Ship dates and specs matter. Nvidia H200 and B200 shipped in 2024 to 2025, GB200 in 2025, AMD MI300 in 2023, MI355 in 2025, MI455 in 2026, TPU v7 in 2025. Memory ranges 141GB to 432GB, power 700W to 1400W, and hourly costs span ~$1.13 to ~$2.21, with TPU v7 near ~$1.28.
For small models, AMD MI355X delivers more tokens per TCO dollar than Nvidia B200, so cost efficiency favors AMD in that segment.
Ship dates and specs matter. Nvidia H200 and B200 shipped in 2024 to 2025, GB200 in 2025, AMD MI300 in 2023, MI355 in 2025, MI455 in 2026, TPU v7 in 2025. Memory ranges 141GB to 432GB, power 700W to 1400W, and hourly costs span ~$1.13 to ~$2.21, with TPU v7 near ~$1.28.
AI workloads are pushing data center investment toward ~$1.4T by 2030, growing ~30% CAGR, with compute the largest slice and networking plus storage rising alongside.
Server mix shifting from traditional CPUs to accelerators, GPUs dominate near term then ASICs take growing share by 2030 as large buyers seek lower cost per inference.
Server mix shifting from traditional CPUs to accelerators, GPUs dominate near term then ASICs take growing share by 2030 as large buyers seek lower cost per inference.
AI agents are becoming the primary interface, so users rely less on individual apps and more on goal based prompts.
Adoption is much faster than the early internet, AI chatbot penetration among smartphone users reached ~18% by year 3 versus PCs at ~3% in their year 3.
This marks a shift from command, to web, to mobile, to an agentic era.
Adoption is much faster than the early internet, AI chatbot penetration among smartphone users reached ~18% by year 3 versus PCs at ~3% in their year 3.
This marks a shift from command, to web, to mobile, to an agentic era.
AI purchasing agents are compressing the funnel of Consumer Transactions by doing discovery, comparison, and decisioning upfront, so shoppers arrive at checkout already convinced.
Personalization becomeing the moat because 95% of the journey happens before the click.
Transaction time fell from ~60 minutes pre-internet to ~90 seconds in the agentic era, so purchase velocity and conversion rates rise while ad wastes shrink.
Personalization becomeing the moat because 95% of the journey happens before the click.
Transaction time fell from ~60 minutes pre-internet to ~90 seconds in the agentic era, so purchase velocity and conversion rates rise while ad wastes shrink.
Agent protocols are standardizing how AI agents are accessing data and executing purchases across retailers, so messy one off integrations are giving way to a clean agentic stack.
Anthropic’s MCP is enabling agents to pull real time context, and OpenAI’s ACP plus Google’s A2A and AP2 are enabling secure payments, fulfillment, and transaction execution.
Retailers are exposing back end systems through APIs, an agentic AI layer is brokering context and settlement, and consumer interfaces are shifting to voice, chat, and wearables, so digital commerce is becoming simpler and faster.
Anthropic’s MCP is enabling agents to pull real time context, and OpenAI’s ACP plus Google’s A2A and AP2 are enabling secure payments, fulfillment, and transaction execution.
Retailers are exposing back end systems through APIs, an agentic AI layer is brokering context and settlement, and consumer interfaces are shifting to voice, chat, and wearables, so digital commerce is becoming simpler and faster.
AI agents are scanning catalogs, videos, and reviews continuously, surfacing in spec items without keyword hunting.
They are comparing attributes across sellers, filtering counterfeits, clustering reviews by theme, and simulating price plus promotion combos based on a user’s budget and preferences.
They are completing checkout with saved payments and tracking, so AI facilitated online spend is rising from ~2% in 2025 to ~25% by 2030, or >$8T.
They are comparing attributes across sellers, filtering counterfeits, clustering reviews by theme, and simulating price plus promotion combos based on a user’s budget and preferences.
They are completing checkout with saved payments and tracking, so AI facilitated online spend is rising from ~2% in 2025 to ~25% by 2030, or >$8T.
AI search is rising from ~10% of global queries in 2025 to ~65% by 2030, so traditional search is losing share.
Advertisers are shifting budgets, AI search ad spend is growing at ~50% per year and is reaching roughly $200B by 2030 with overall search ads still expanding.
Monetization is following usage with an estimated ~2 year lag, so revenue mix is tilting toward AI results.
Advertisers are shifting budgets, AI search ad spend is growing at ~50% per year and is reaching roughly $200B by 2030 with overall search ads still expanding.
Monetization is following usage with an estimated ~2 year lag, so revenue mix is tilting toward AI results.
AI mediated consumer revenue is growing ~105% per year from ~$20B in 2025 and is reaching ~$900B in 2030.
Lead generation and advertising are contributing most of the growth, subscriptions are staying smaller.
AI agents are transforming digital commerce and are directing spend toward ad and lead flows that they are controlling.
Lead generation and advertising are contributing most of the growth, subscriptions are staying smaller.
AI agents are transforming digital commerce and are directing spend toward ad and lead flows that they are controlling.
Models are gaining reasoning, tool use, and context, so agents are reliably completing longer tasks, rising from 6 minutes to 31 minutes during 2025 at 80% success.
Workers are reporting ~50 minutes saved per day, which is valuing at ~$47 using a $56.5 hourly wage, so a $20 monthly subscription is paying back in ~0.5 day.
Workers are reporting ~50 minutes saved per day, which is valuing at ~$47 using a $56.5 hourly wage, so a $20 monthly subscription is paying back in ~0.5 day.
Costs for capable models are collapsing across domains, often falling 90% to 99% on major benchmarks.
Software development with strong coding models is dropping from ~$3.50 to ~$0.32 per 1M tokens between Apr 2025 and Dec 2025.
Newer reasoning models like DeepSeek V3.2 are pushing blended costs toward ~$0.32 per 1M tokens, so teams are scaling usage without blowing budgets.
Software development with strong coding models is dropping from ~$3.50 to ~$0.32 per 1M tokens between Apr 2025 and Dec 2025.
Newer reasoning models like DeepSeek V3.2 are pushing blended costs toward ~$0.32 per 1M tokens, so teams are scaling usage without blowing budgets.
US frontier models are maintaining a roughly 6 month lead in capability, while China is catching up fast and is dominating open weight releases.
TSMC is producing far more advanced chips each day, about 135B transistors, while China's SMIC is producing about 3.5B, so capacity is about 38x larger on TSMC’s side.
meaning US aligned firms are getting supply first, since TSMC is fabricating the newest nodes at volume, so hardware scarcity is hurting China more.
In practice, that 38x edge is compounding, because more compute is creating better models, better models are attracting more users and money, and more money is buying even more TSMC capacity.
So China will need much more access to cutting edge compute to close the remaining gap.
TSMC is producing far more advanced chips each day, about 135B transistors, while China's SMIC is producing about 3.5B, so capacity is about 38x larger on TSMC’s side.
meaning US aligned firms are getting supply first, since TSMC is fabricating the newest nodes at volume, so hardware scarcity is hurting China more.
In practice, that 38x edge is compounding, because more compute is creating better models, better models are attracting more users and money, and more money is buying even more TSMC capacity.
So China will need much more access to cutting edge compute to close the remaining gap.
OpenAI and Anthropic are scaling revenues to levels rivaling large software firms, with implied 2025 run rates near $20B and $9B.
AI native startups are exploding, companies like Cursor, Harvey, OpenEvidence, and Sierra are hitting $100M to $1B ARR within ~2 to 3 years, and some are growing >1,000% year over year, showing demand is accelerating across verticals.
AI native startups are exploding, companies like Cursor, Harvey, OpenEvidence, and Sierra are hitting $100M to $1B ARR within ~2 to 3 years, and some are growing >1,000% year over year, showing demand is accelerating across verticals.
Software’s share of spend is rising as AI is augmenting knowledge workers.
Under modest adoption, global software spend will grow to ~$3.4T at 19% CAGR, unlocking ~$22T in surplus while employment is growing 6.3%.
Under accelerated adoption, spend will reach ~$7T at 37% CAGR, unlocking ~$57T, with more automation and some hour reduction.
Under rapid mass adoption, spend will reach ~$13T at 56% CAGR, unlocking ~$117T, with 81% of tasks automating and average hours dropping 20%, while long term unemployment is not rising.
Under modest adoption, global software spend will grow to ~$3.4T at 19% CAGR, unlocking ~$22T in surplus while employment is growing 6.3%.
Under accelerated adoption, spend will reach ~$7T at 37% CAGR, unlocking ~$57T, with more automation and some hour reduction.
Under rapid mass adoption, spend will reach ~$13T at 56% CAGR, unlocking ~$117T, with 81% of tasks automating and average hours dropping 20%, while long term unemployment is not rising.
Generated by Thread Navigator
Press ⌘ + S to quick-export