Startups binge on AI while big firms sip cautiously, study shows
(2025/10/03)
- Reference: 1759511161
- News link: https://www.theregister.co.uk/2025/10/03/startups_binge_on_ai/
- Source link:
The Andreessen Horowitz venture capital firm (aka A16z) crunched startup spending data and found young firms stuffing AI into everything, while bigger businesses remain far more restrained.
To see what startups are prioritizing, A16z worked with fintech firm Mercury, which provides financial services to many Silicon Valley companies, and its database of the spending habits of more than 200,000 commercial customers between June and August of this year. Armed with that data, the pair arrived at a ranking of the top 50 AI-native application layer companies. That list, the pair said, suggests startups are going hard on AI in a way that bigger companies aren't.
"While big companies are getting incremental benefits to existing team structures, at startups we're seeing truly AI-native companies emerge, built around the next generation of software," [1]wrote A16z partners Olivia Moore, Marc Andrusko, and Seema Amble.
[2]
Horizontal AI platforms and their vendors - those that focus on boosting productivity across an entire company instead of just for specific roles - are the most popular on the list, making up 60 percent of the 50 entries on the list. Unsurprisingly, the two most popular vendors on the list (OpenAI and Anthropic) are both horizontal and popular, but the third entry may surprise those who have been resistant to the [3]Next Big Thing in AI-assisted software development. It's vibe coding outfit Replit.
[4]
[5]
Yes, that's the same Replit, whose AI agent [6]was accused over the summer of deleting a customer's live production database despite code-freeze instructions, then fabricating data, issuing misleading reports, and attempting to conceal bugs in a test run. Despite those drawbacks, A16z believes vibe coding "is no mere consumer trend" based on its report. Along with Replit, vibe coding platforms Cursor, Lovable, and Emergent are all on the list too.
The report also concluded from looking at Mercury customer spending habits that most startups are using AI to assist human employees rather than replacing them. Among the vertical (i.e., role-specific) applications on the list, 12 focus on assisting human workers, while only five aim to serve as replacements for human workers.
[7]
Those five, however, suggest that startups are keen on automating very specific roles: Two offer agentic AI legal services (maybe [8]not [9]the [10]best [11]idea ), one offers an AI IT helpdesk, another pitches AI software engineers, and the fifth claims to automate go-to-market team functions.
In short, startups are looking to get rid of those expensive employees who ask for good money for their expertise, and A16z expects such employee replacements to increase in popularity.
"We expect to see more end-to-end agentic products – and even AI-native services businesses – emerge," the report predicts. Hiring an AI, the trio of A16z partners said, will help startups avoid "multi-year contracts with expensive providers like lawyers, accountants, etc."
Your vibes are powerless in the face of the bubble
To read A16z's report makes it seem as if there are two business worlds emerging with regard to their approach to artificial intelligence. There are established enterprises that seem [12]largely unwilling to dive headlong into the use of disruptive, overhyped technology that has [13]yet to prove its worth , and there are startups that have no history or established practices to upend by bringing in AI, so why not bring it in from the beginning?
That's all well and good if it works out, but the envisioned AI future is still far from a sure bet.
[14]AI coding hype overblown, Bain shrugs
[15]If you want a picture of the future, imagine humans checking AI didn't make a mistake – forever
[16]AI coding tools make developers slower but they think they're faster, study finds
[17]AI has had zero effect on jobs so far, says Yale study
As we've covered extensively, most big businesses that have poured money into AI [18]aren't seeing much return , and the next big wave of it - agentic AI that promises independent decision-making and worker replacement - is still immature, with Gartner predicting more than 40 percent of such projects will be [19]canceled by 2027 .
Then there's the fact that the entire AI market is [20]blowing up like a balloon so rapidly that it's impossible to overlook the possibility that it's going to collapse soon.
[21]
Take the OpenAI/Oracle deal, for example. OpenAI recently signed a deal with Oracle to [22]buy some $300 billion worth of compute capacity over the next five years. It doesn't have that money, and so will need investors to prop up its ambitions.
Oracle, meanwhile, may need to borrow [23]as much as $100 billion over the next four years to build the datacenters required for its OpenAI deal, according to analyst estimates.
"Growth before profit" was a pervasive theme of the dot-com crash; OpenAI has yet to turn a profit, nor has Anthropic - [24]lots of companies are losing money hand over fist trying to prop up the AI future they're pushing.
If A16z's reading of the state of AI spending is anything to go on, either giant enterprises are going to be left behind as startups innovate by going AI first, or the bubble is going to pop and a lot of small outfits are going to find themselves in trouble. We reached out to the authors to get their take, but didn't hear back. ®
Get our [25]Tech Resources
[1] https://a16z.com/the-ai-application-spending-report-where-startup-dollars-really-go/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://www.theregister.com/2025/07/25/opinion_column_vibe_coding/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/07/21/replit_saastr_vibe_coding_incident/
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/02/25/fine_sought_ai_filing_mistakes/
[9] https://www.theregister.com/2025/07/08/georgia_appeals_court_ai_caselaw/
[10] https://www.theregister.com/2023/06/22/lawyers_fake_cases/
[11] https://www.theregister.com/2025/02/14/attorneys_cite_cases_hallucinated_ai/
[12] https://www.theregister.com/2025/07/15/sec_risk_factors_ai/
[13] https://www.theregister.com/2025/05/06/ibm_ai_investments/
[14] https://www.theregister.com/2025/09/23/developers_genai_little_productivity_gains/
[15] https://www.theregister.com/2025/07/16/if_you_want_a_picture/
[16] https://www.theregister.com/2025/07/11/ai_code_tools_slow_down/
[17] https://www.theregister.com/2025/10/01/ai_isnt_taking_people_jobs/
[18] https://www.theregister.com/2025/09/17/return_on_investment_for_copilot/
[19] https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
[20] https://www.theregister.com/2025/10/03/ai_bubble_watch_markets/
[21] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[22] https://www.theregister.com/2025/09/11/openai_reportedly_on_the_hook/
[23] https://www.theregister.com/2025/09/29/oracle_ai_debt/
[24] https://www.wheresyoured.at/why-everybody-is-losing-money-on-ai/
[25] https://whitepapers.theregister.com/
To see what startups are prioritizing, A16z worked with fintech firm Mercury, which provides financial services to many Silicon Valley companies, and its database of the spending habits of more than 200,000 commercial customers between June and August of this year. Armed with that data, the pair arrived at a ranking of the top 50 AI-native application layer companies. That list, the pair said, suggests startups are going hard on AI in a way that bigger companies aren't.
"While big companies are getting incremental benefits to existing team structures, at startups we're seeing truly AI-native companies emerge, built around the next generation of software," [1]wrote A16z partners Olivia Moore, Marc Andrusko, and Seema Amble.
[2]
Horizontal AI platforms and their vendors - those that focus on boosting productivity across an entire company instead of just for specific roles - are the most popular on the list, making up 60 percent of the 50 entries on the list. Unsurprisingly, the two most popular vendors on the list (OpenAI and Anthropic) are both horizontal and popular, but the third entry may surprise those who have been resistant to the [3]Next Big Thing in AI-assisted software development. It's vibe coding outfit Replit.
[4]
[5]
Yes, that's the same Replit, whose AI agent [6]was accused over the summer of deleting a customer's live production database despite code-freeze instructions, then fabricating data, issuing misleading reports, and attempting to conceal bugs in a test run. Despite those drawbacks, A16z believes vibe coding "is no mere consumer trend" based on its report. Along with Replit, vibe coding platforms Cursor, Lovable, and Emergent are all on the list too.
The report also concluded from looking at Mercury customer spending habits that most startups are using AI to assist human employees rather than replacing them. Among the vertical (i.e., role-specific) applications on the list, 12 focus on assisting human workers, while only five aim to serve as replacements for human workers.
[7]
Those five, however, suggest that startups are keen on automating very specific roles: Two offer agentic AI legal services (maybe [8]not [9]the [10]best [11]idea ), one offers an AI IT helpdesk, another pitches AI software engineers, and the fifth claims to automate go-to-market team functions.
In short, startups are looking to get rid of those expensive employees who ask for good money for their expertise, and A16z expects such employee replacements to increase in popularity.
"We expect to see more end-to-end agentic products – and even AI-native services businesses – emerge," the report predicts. Hiring an AI, the trio of A16z partners said, will help startups avoid "multi-year contracts with expensive providers like lawyers, accountants, etc."
Your vibes are powerless in the face of the bubble
To read A16z's report makes it seem as if there are two business worlds emerging with regard to their approach to artificial intelligence. There are established enterprises that seem [12]largely unwilling to dive headlong into the use of disruptive, overhyped technology that has [13]yet to prove its worth , and there are startups that have no history or established practices to upend by bringing in AI, so why not bring it in from the beginning?
That's all well and good if it works out, but the envisioned AI future is still far from a sure bet.
[14]AI coding hype overblown, Bain shrugs
[15]If you want a picture of the future, imagine humans checking AI didn't make a mistake – forever
[16]AI coding tools make developers slower but they think they're faster, study finds
[17]AI has had zero effect on jobs so far, says Yale study
As we've covered extensively, most big businesses that have poured money into AI [18]aren't seeing much return , and the next big wave of it - agentic AI that promises independent decision-making and worker replacement - is still immature, with Gartner predicting more than 40 percent of such projects will be [19]canceled by 2027 .
Then there's the fact that the entire AI market is [20]blowing up like a balloon so rapidly that it's impossible to overlook the possibility that it's going to collapse soon.
[21]
Take the OpenAI/Oracle deal, for example. OpenAI recently signed a deal with Oracle to [22]buy some $300 billion worth of compute capacity over the next five years. It doesn't have that money, and so will need investors to prop up its ambitions.
Oracle, meanwhile, may need to borrow [23]as much as $100 billion over the next four years to build the datacenters required for its OpenAI deal, according to analyst estimates.
"Growth before profit" was a pervasive theme of the dot-com crash; OpenAI has yet to turn a profit, nor has Anthropic - [24]lots of companies are losing money hand over fist trying to prop up the AI future they're pushing.
If A16z's reading of the state of AI spending is anything to go on, either giant enterprises are going to be left behind as startups innovate by going AI first, or the bubble is going to pop and a lot of small outfits are going to find themselves in trouble. We reached out to the authors to get their take, but didn't hear back. ®
Get our [25]Tech Resources
[1] https://a16z.com/the-ai-application-spending-report-where-startup-dollars-really-go/
[2] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=2&c=2aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D2%26raptor%3Dcondor%26pos%3Dtop%26test%3D0
[3] https://www.theregister.com/2025/07/25/opinion_column_vibe_coding/
[4] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[5] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[6] https://www.theregister.com/2025/07/21/replit_saastr_vibe_coding_incident/
[7] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=4&c=44aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D4%26raptor%3Dfalcon%26pos%3Dmid%26test%3D0
[8] https://www.theregister.com/2025/02/25/fine_sought_ai_filing_mistakes/
[9] https://www.theregister.com/2025/07/08/georgia_appeals_court_ai_caselaw/
[10] https://www.theregister.com/2023/06/22/lawyers_fake_cases/
[11] https://www.theregister.com/2025/02/14/attorneys_cite_cases_hallucinated_ai/
[12] https://www.theregister.com/2025/07/15/sec_risk_factors_ai/
[13] https://www.theregister.com/2025/05/06/ibm_ai_investments/
[14] https://www.theregister.com/2025/09/23/developers_genai_little_productivity_gains/
[15] https://www.theregister.com/2025/07/16/if_you_want_a_picture/
[16] https://www.theregister.com/2025/07/11/ai_code_tools_slow_down/
[17] https://www.theregister.com/2025/10/01/ai_isnt_taking_people_jobs/
[18] https://www.theregister.com/2025/09/17/return_on_investment_for_copilot/
[19] https://www.theregister.com/2025/06/29/ai_agents_fail_a_lot/
[20] https://www.theregister.com/2025/10/03/ai_bubble_watch_markets/
[21] https://pubads.g.doubleclick.net/gampad/jump?co=1&iu=/6978/reg_software/aiml&sz=300x50%7C300x100%7C300x250%7C300x251%7C300x252%7C300x600%7C300x601&tile=3&c=33aOBHfU8oc2Eu9cMkzDyUMQAAABI&t=ct%3Dns%26unitnum%3D3%26raptor%3Deagle%26pos%3Dmid%26test%3D0
[22] https://www.theregister.com/2025/09/11/openai_reportedly_on_the_hook/
[23] https://www.theregister.com/2025/09/29/oracle_ai_debt/
[24] https://www.wheresyoured.at/why-everybody-is-losing-money-on-ai/
[25] https://whitepapers.theregister.com/
zimzam
Very much like blockchain, outside of a few specific instances, "AI" is a solution to problems with existing solutions that are more efficient.
I suspect LLMs will flame out and be supplanted by Small Language Models that are trained for specialised use cases. Essentially what we had before but with newer techniques. Whether startups make any better use of them than established firms...
Anonymous Coward
True, especially when the inevitable price hikes begin as the LLM companies try to recoup vast money spent. Then all of the secondary services built on them will get squeezed.
Help startups avoid
that one in the corner
> "multi-year contracts with expensive providers like lawyers, accountants, etc."
Well, I reckon I can guess what is going to wrong at the end of the tax year.
popcorn startups to rake it in when the startups go pop?
Anonymous Coward
pop pop pop...
Excused Boots
Of course one explanation is that ‘startups’ don’t know any better and latch onto the the latest ’shiny', but more longer lived firms have seen and witnessed all sorts of fads come and go, the vast majority of which turn out to be snake oil.
If I recall correctly, a few years ago, wasn’t ‘blockchain’ going to be the answer to all problems?
From reports here there's a third category: bigger companies using AI as an excuse for redundancies getting rid of staff who can't use it.