AI Industry Shakes Off Jitters as Big Investments & New Products March Ahead
Plus, legal tech again in VCs' sights
The Main Item
Big Tech Stays Bullish on Data Center Demand as Reasoning Models Proliferate
The week began with the news that tech giant Microsoft was voiding two of its leases for planned data centers, an unexpected development that was first noted in a report from TD Cowen and then picked up by Bloomberg.
Was this the much-feared pullback in spending that some in the industry believed would result from DeepSeek’s success with low-cost AI?
Five days later we have our answer: no.
For one thing, it’s not clear whether Microsoft has actually dialed anything back, despite talk that its deal with OpenAI might be fraying amid concerns about AI spending. TD Cowen analysts speculated in a second report that the cuts could be related to OpenAI’s new partnership with Oracle for data centers, as well as its shift to focus on US investment, but the company also reiterated its plans to spend some $80 billion on new AI infrastructure this fiscal year.
OpenAI’s GPT-4.5 launch announcement on Thursday came with a warning that the model’s rollout would be hindered by a dwindling supply of GPUs, indicating that demand for AI compute still exceeds supply. “This isn’t how we want to operate, but it’s hard to perfectly predict growth surges that lead to GPU shortages,” CEO Sam Altman wrote on X.
Demand for GPUs continues to power extraordinary results at Nvidia, whose shares took a beating after the DeepSeek announcement. The dominant AI chip-maker beat its projected earnings as Q4 revenue leaped 78% from a year earlier, though its stock price dipped on concerns about margins and export restrictions, along with a broad market “risk-off” that followed Trump’s latest tariff comments.
Amazon, for its part, launched a new Alexa+ assistant this week to high praise. If it can bring us closer to our promised AI agent future that could also prompt fresh demand, which the company aims to satisfy in part with its own chips.
Meta is in talks to ramp up its own data center investments even further, per The Information, with plans for a project with a $200 billion price tag — it’s even looking into raising $35 billion in outside capital from Apollo to do so. Indeed, none of the largest public companies have seen a pullback in capex yet, with $300 billion still earmarked for this year.
There seems to be an emerging consensus that “reasoning” models like DeepSeek may be much cheaper to build than the LLMs from OpenAI and the other giants, but since they require a lot more processing power to run queries, they won’t necessarily reduce demand for data center capacity.
“For a while we’ve had too much demand for not enough supply,” said FirstMark’s Matt Turck, referring to the need for compute to power training and AI applications models. “It’s possible we may have a phase at some point where have too much supply and not enough demand — my guess is that we’ll have one or two more ups and downs like that before the market reaches some equilibrium.”
Demand forecasting is brutally difficult, since it requires predicting both the evolution of the technology and the adoption rate. Tyler Cowen discussed one dimension of this in Marginal Revolution earlier this week, noting that as technology progresses, sectors that are less efficient — think healthcare, or government — tend to become a larger share of the economy over time. That can hold back the macro benefits of new technology, which in turn can slow investment.
“The more efficient AI becomes, the more this trend is likely to continue, which slows the prospective measured growth gains from AI,” he writes.
But VCs are starting to see fast-growing vertical AI startups even in healthcare, an industry with a massive TAM that is often slow to adopt new technology en masse.
Health and bio AI startup funding last year didn’t surpass its 2021 peak, but so far in 2025 it’s already hit $1.8 billion, well above the pace of the past several years. One investor I spoke with pointed to Sequoia’s recent bet on OpenEvidence, a fast-growing search engine for doctors, as a promising sign. A16z on Wednesday announced their investment into Camber, which is using AI to speed up healthcare reimbursements.
To be sure, there are plenty of AI skeptics out there, worried that AI will develop too slowly or that its capabilities have been overstated. Jeremy Kahn at Fortune argued in a long piece this week that OpenAI’s failure to launch GPT-5 on the promised schedule pointed to a larger issue with LLMs, namely that they are reaching a point of diminishing returns well before reaching AGI.
There’s also the remaking of the world order to worry about. Semafor says business leaders are already turning on Trump, with the ongoing threats and flip-flops on tariffs shaking financial markets. Nvidia and its chip industry compatriots are sweating China export controls that could hit sales.
These caveats, along with the arrival of SoftBank as a major AI investor, could be read as suggesting a market top: SoftBank’s Masayoshi Son was often mocked as the “dumb money” in the frenzy of the last decade. But valuations for companies like Uber and Airbnb that looked insane in the early 2010s turned out to be more than reasonable, and AI investors clearly believe we’re still at the beginning.
Brought to you by Vanta
Compliance for Startups: Download the Checklist
As a startup founder, finding product-market fit is your top priority.
But landing bigger customers requires SOC 2 or ISO 27001 compliance — a time-consuming process that pulls you away from building and shipping. That’s where Vanta comes in.
Join over 9,000 companies, including hundreds of Y Combinator-backed startups like Supabase, Newfront, and Fern who streamline compliance with Vanta’s automation and trusted network of security experts. Start with our SOC 2 compliance checklist, which breaks down the process into clear steps — so you can spend less time on compliance and more time growing your business.
Two Big Charts
Legal Tech Funding Jumps in the New Year as AI Takes Hold
Startups building AI tools for lawyers are seeing an uptick in interest from VCs, new data from Crunchbase confirms. We wrote about the sector back in October, with investors noting that a lot of legal administrative work is ripe for automation.
Q3 of last year saw a big jump in legal funding with over $1.23 billion in funding going to 45 deals in the sector. AI-powered legal tech company EvenUp closed its $135 million Series D led by BCV during that quarter.
Legal tech companies have already garnered $600 million in funding so far this year, nearly double what the sector received in Q4 of 2024. Last year’s total of $2.15 billion was just shy of the $2.35 billion raised in 2021, but topped both 2022 and 2023.
Five Notable Deals
NinjaOne, Eikon Therapeutics, Quantum Machines, Raise, Metronome
Late-stage powerhouses including Iconiq Growth and PSG Equity participated in outsized funding rounds for IT and quantum computing companies.
IT startup NinjaOne raised $500 million in a Series C extension round co-led by Iconiq Growth and CapitalG.
Biotech startup Eikon Therapeutics raised $350.7 million in Series D funding from Lux Capital, The Column Group, AME Cloud Ventures, E15 VC, Foresite Capital, Soros Capital, and General Catalyst, among others.
Quantum computing startup Quantum Machines raised $170 million in Series C funding led by PSG Equity. Intel Capital and Red Dot Capital Partners also participated, among others.
Raise, a crypto platform for buying gift cards, raised $63 million in a round led by Haun Ventures. Other investors include Paper Ventures, Selini Capital, and GSR Ventures.
Metronome, a usage-based billing startup, raised $50 million in Series C funding led by returning investor NEA. Other investors include a16z, General Catalyst, Workday Ventures, Greyhound Capital, TrueBridge Capital Partners, Activant Capital, SineWave Ventures, and Megalith Ventures.
Stories We’re Reading
Thinking Machines Lab Enters the Three Comma Club, Amazon’s Quantum Chip Debuts, Perplexity Raises a Venture Fund, DOGE Gets a Presidential Mandate, Nayeema Raza’s Smart New Podcast Debuts With Mark Cuban
We got a better picture of Mira Murati’s fundraising efforts for Thinking Machines Lab this week, per Business Insider: the startup has been in talks to raise $1 billion at a $9 billion valuation. That puts it squarely in the ranks of her former colleague’s initial fundraising hauls — Ilya Sutskever’s SSI first raised $1 billion at a $5 billion valuation before starting talks to up that to $20 billion. Both companies have yet to launch a product.
Following fast on the heels of Microsoft and Google, Amazon is now jumping into the quantum game. Amazon Web Services launched a new quantum chip on Thursday named “Ocelot,” which sports a novel design developed in partnership with Caltech researchers. The chip incorporates two micro-quantum chips stacked on top of each other, which the company says can reduce the costs for error correction by up to 90%. Big tech breakthroughs over the past year coincided with a bump in funding to quantum startups: venture funding for the sector hit a record high in 2024 with over $1.5 billion in dealflow by November of last year, per Crunchbase.
AI-powered search engine Perplexity is raising a $50 million venture fund to back early-stage AI startups, CNBC reported Tuesday. The startup will be the anchor investor in the fund, but will also seek outside LPs. The two general partners will be Kelly Graziadei and Joanna Lee Shevelenko, who previously led early-stage deals at f7 Ventures.
Elon Musk’s federal cost-cutting project got a boost from President Trump on Wednesday by way of an executive order creating a centralized system for federal government payments — to be monitored by DOGE. At the same time, DOGE is backpedaling on its cost-cutting claims as journalists reveal that many of them are bogus. Musk admitted at Trump’s cabinet meeting Wednesday that his team had “accidentally” canceled USAID funding for Ebola prevention, but that it had been restored, saying the team “won’t be perfect” when it comes to its initial orders. The Washington Post refuted his statement, reporting that the Ebola teams were still dismantled. The Trump administration has also walked back DOGE’s requirement that federal workers send weekly performance and progress reports, calling compliance with the request “voluntary.”
Independent journalist and friend-of-the-newsletter Nayeema Raza is launching a new podcast today called “Smart Girl Dumb Questions.” First up, she asks Mark Cuban if billionaires can save us, how he hopes to get employers like Google and Apple to ditch big health insurers, and what capitalism might learn from the NBA. We at Newcomer would highly recommend that you check it out. Tune in on Apple, Spotify or YouTube.
Newcomer Podcast
$200 Billion For Data Centers... In This Economy?
We break down the week that was in AI news in this episode of the Newcomer Podcast, from the unnecessary panic caused by reports of Microsoft canceling leases for data centers to Nvidia’s quarterly earnings report. Despite tech companies moving full steam ahead on AI, including OpenAI’s launch of GPT 4.5 and SoftBank’s $200 billion commitment to data centers, tech stocks dropped, which our hosts blame on the natural slower tech adoption curve and overall market uncertainty from President Trump’s chaotic policy changes.
Later in the episode, we unpack Jeff Bezos’ increasing control over the Washington Post’s opinion section and tech CEOs pre-empting so-called “hit pieces” on X. The two then shout out UK-based developer tool Lovable’s $15 million funding extension and rapid growth. They close the episode with a call for readers to submit their burning questions and commentary for next week’s Newcomer reader mailbag. You can submit your own question by filling out this Google form.
Chapters
01:40 — Microsoft’s Data Center Cancellations Is No Cause For Alarm
07:27 — GPT4.5 Is ‘Not A Frontier Model’
08:37 — Jeff Bezos’ Control Over The Opinion Section While Journalism Is In a ‘Lose-Lose’ Position
14:45 — Lovable Raises $15 Million Off Of $17 million in ARR In Just 3 Months
16:35 — Reader Mailbag: Callout For Questions