DeepSeek & More AI Excitement For 2025

This story features NEXTDC LIMITED, and other companies. For more info SHARE ANALYSIS: NXT

The rapid pace of AI evolution continues, what’s new, and which sectors/stocks are in focus

-Trump flicks the AI switch
-China challenges US innovation 
-Morgan Stanley’s latest AI survey in focus
-The next big thing in AI
-Beyond textual data to the optical data transformation
-E&P’s update on Australian Tech Stocks

By Danielle Ecuyer

At the outset of the first update on GenAI in 2025, a brief definition reset is apposite.

Such is the scope and scale of change in the megatrend that, without defining the different forms of AI, the investor and reader are left proverbially up the technology river without a paddle.

The following definitions should assist:

–Inference AI focuses on applying trained models to infer results or make predictions. Currently widely used with chat bots, generating or predicts outputs from inputs, such as the current large language models

–Generative AI are systems designed to create new content, data, or ideas that resemble human-produced outputs. It uses machine learning models, often powered by neural networks, to generate text, images, music, videos, code, or other content based on patterns learned from existing data.

–Agentic AI acts on its own to achieve goals, requiring autonomy and real-time decision-making. Agentic is in development and includes autonomous vehicles and virtual assistants.

–AGI or artificial general intelligence, would embody human-level cognition and versatility, transcending the limitations of current AI. In development, theoretical, and referred to as singularity,’ a point at which machine intelligence supersedes human intelligence.

Source: ChatGPT

2025 starts with a Big Technology Bang

Trump 2.0 has hit the ground running and seemingly already upped the ante in the race for developing artificial general intelligence, having declared he was softening the regulatory safety guardrails for generative AI development on day one.

Unlike Trump 1.0, America’s tech titans are clearly in the President’s orbit. The Stargate announcement at the Oval Office for a joint venture, where an initial US$100bn will be spent on the construction of the “physical and virtual infrastructure to power the next generation of advancements in AI and this will include the construction of colossal data centres,” further confirmed the President’s newfound enthusiasm for AI.

The joint venture is between Oracle, SoftBank, and OpenAI, with funds from additional investors, such as Microsoft yet to be announced in full. The Stargate project has a target to reach US$500bn in AI projects.

According to Bloomberg, the Chief Executive Officer of Google DeepMind told the agency in Davos, “the administration is getting advice from the people who really understand what’s happening at the cutting edge.”

Nigel Green, CEO of deVere Group, an independent financial advisory and asset manager, said:

“Many analysts argue the AI rally is going to dampen this year, but this news reaffirms we’re merely in a recalibration phase, not a revolution in market leadership. AI is not a fleeting trend, it’s the foundation of the future. Investors who fail to recognise this risk missing out on one of the most transformative opportunities of our time.”

And:

“Trump’s AI initiative is a wake-up call to the immense and ongoing potential of artificial intelligence. The time to build exposure to the defining economic shift of our era is today.”

The AI race continues to heat up

As described by Geninnov founder and investment manager Nilesh Jasani, world powers and corporations are confronting the prisoner’s dilemma: if they do not pursue AI developments, their biggest competitors will.

The Trump Administration’s incoming announcements suggest caution around intelligent AI concerns and sufficient clean energy to power the mega data centres is seemingly swept away in favour of significantly advancing America’s technologies.

The fear of China beating the US in the AI race is a highly motivating factor, see also the DeepSeek development over the Australia Day long weekend.

Both Bloomberg and Jasani lay bare the challenges involved in progressing AI.

Confronted with substantial investment costs and potential challenges in securing the necessary GPUs, China is adopting a different pathway to bypass the “compute constraints,” Jasani proposes.

Instead of projects such as Stargate, where billions of dollars are invested in large-scale infrastructure for large-scale language models, China is moving to use new technologies at a lower cost and lower energy use.

If necessity is the mother of invention, Cathy Wood, the founder and CEO of ARK Invest, emphasises the transformative potential of disruptive innovation through the phrase “cheaper, faster, better, and more scalable.”

DeepSeek, a Chinese start-up, showcased the release of DeepSeek-V3, an open-source AI language model using old Nvidia GPUs, H800s, at a cost of US$5.5m, or, as Jasani highlights, at costs 15 to 20 times less than usually assumed.

The H800 GPU was launched on March 21, 2023, and priced at US$17,800, in contrast to the Blackwell GPUs expected to cost US$30,000 to US$70,000..

ChatGPT underscores the DeepSeek-V3 model outperforms Meta’s Llama 3.1 and Alibaba’s Qwen 2.5 and is competitive with ChatGPT-4.0 and Claude 3.5 Sonnet.

Jasani also highlights MiniMax-01 succeeded in developing a model that can use 20 to 30 times bigger input than the best current models. The model proved it could extract data from a 450-page IPO document and compute DuPont analysis (a method of financial analysis to break down a company’s return on equity into multiple components).

Other existing models are unable to handle the large size of the data input and computation. Jasani proposes there are alternatives to developing compute with lower costs and increased efficiencies.

As the DeepSeek development spread across the five continents, shares in US technology companies and AI beneficiaries sold off with follow-through implications for peers on other exchanges.

Nvidia continues to challenge the status quo

Nvidia’s latest GB10 chip, cost US$3000, was showcased at the January Consumer Electronics Show in Las Vegas. Jasani details how the chip allows for decentralised computation on devices, obviating the need for large data centres with high-cost barriers.

Latency, security, and privacy issues have the potential to be reduced with local processing on an AI-powered device or tool, and the lag between image capture and analysis, for example, is minimised for the medical industry.

Jensen Huang believes the shift to physical AI is a burgeoning new era, Jasani details, where more powerful PCs and physical devices will be supported by Edge AI, new AI internet-of-things platforms, and networks of connected devices.

Although the innovations suggest the death knell of large-scale data centres, Jasani stresses AI-driven solutions will require “robust infrastructure” to handle growing inference loads.

A layperson might visualise the evolution and development as a mix-and-match combination of different solutions and outcomes. The earlier quoted Nigel Green highlights AI is no longer just a tech story; increasingly there is evidence of its application changing healthcare, logistics, energy, and financial services.

“It’s become a critical driver of productivity and efficiency, making it a foundational component of the global economy.”

AI adoption is just getting started

Morgan Stanley’s third AI Adopter survey, with the first in January 2024 and second in June 2024, has revealed a surprising rate at which AI continues to evolve.

The analysts’ coverage of over 3,700 global stocks showed 585 stocks had changed their AI exposure or AI materiality worth US$13trn in market capitalisation.

Materiality refers to the relevance or significance of artificial intelligence (AI) in terms of its impact on companies.

The broker states the capabilities of AI models and costs are developing apace, with adoption by corporations still at an early phase.

Five new findings emerged from Morgan Stanley’s third study:

-275 stocks with a US$5.7trn value changed exposure, and 310 stocks with a US$7.8trn value changed materiality, with 115 stocks or US$2.2trn in value changing both exposure and materiality. Within those findings, some stocks changed categories and exposures either up or down.

-24 stocks (net) moved from Adopters to Enabler/Adopter, and seven stocks were disrupted.

-AI is more material for 17% of Financial coverage, with Financials having the highest AI materiality over the analysts’ global coverage. Consumer Staples, in contrast, experienced a net -3% lower importance.

-139 adopters were identified as possessing “High Pricing Power” and outperformed “Low Pricing Power” adopters by 30% since the release of ChatGPT. Eight companies have been included in the high pricing power and high AI materiality adopters category since the last survey.

-Interestingly, the companies that increased exposure and materiality in Survey 2 outperformed the MSCI World by 25% in 2024, predominantly over the second half.

Morgan Stanley proposes the rate of AI change in adoption will remain a “differentiating factor” for stocks in 2025. Expanding further, the broker details the speed of AI evolution in foundational models, such as ChatGPT.

In the last two years, model capabilities have moved from a reliable output or supplement for five-second or five-minute tasks, the analyst details, to models offering reliable output for 15 minutes to one hour.

The latest models, which require more compute intensity, are transitioning to the inference or “greater thought” stage rather than data “regurgitation.” These models are evolving to four-hour and ultimately five-day tasks being automated.

Morgan Stanley stresses the rate of “evolution” and level of corporate adoption and materiality remains underappreciated by investors.

Agentic AI, on the rise

Both Citi and Morgan Stanley agree 2025 brings forth the new “New Thing,” Agentic AI, in what Citi describes as the “Do it For Me” economy.

Agentic AI is artificial intelligence that can implement autonomous decisions without humans and solve complex problems independently and proactively, Citi explains.

From the broker’s perspective, agentic AI will have a bigger impact on the US economy and finance than the internet era.

Morgan Stanley’s head of software believes 2025 will mark the escalation of the agentic layer onto enterprise software.

The broker highlights Gartner’s research, which proposes by 2028, 33% of enterprise software applications will include agentic AI, up from 1% today. Industry analysts estimate 15% of daily work decisions will be made autonomously.

In Morgan Stanley’s third study, the banking sector showed the highest percentage of stocks evidencing AI materiality. The study illustrated Financials exhibited the highest degree of “low-hanging opportunities” across revenue and costs.

Citi highlights the Financial sector as the second-largest user of GenAI after telecoms and media; historically it is the largest IT spender.

The broker isolates applications across personalised offers, customer engagement, operational efficiency, risk and underwriting, financial forecasting, know-your-client/onboarding, and fraud prevention in the financial sector for agentic AI, specifically the wealth management/retail banking, corporate banking, institutional investor, and insurance sectors.

Morgan Stanley concludes investors should continue to play the followings three themes for AI in 2025:

-Enablers/Adopters will continue to play the most significant role in AI evolution, which is believed to be a decade-long thematic. 2025 will start to reveal “software layer monetisation,” which is viewed as an opportunity. Infrastructure continues to be important. This category is represented by 234 companies with high AI materiality, US$29trn in market cap, and US$5trn in revenue.

-The rate of change. Morgan Stanley views the surveys as providing valuable inputs into tracking the rate of change, and results indicate companies adopting new AI technologies can achieve “structural advantages.” Financials are viewed as potential AI beneficiaries with rising AI importance.

-Adopters with pricing power have a “key advantage.” Sifting through the 1,300 list of AI Adopters, the broker views high-pricing-power companies as in the strongest position to generate value from improved efficiencies.

Changing the industrial complex

The Tesla analyst at Morgan Stanley, Adam Jonas, recently discussed the multiplicity of technological challenges Elon Musk is trying to address.

Jonas refers to them as Tesla’s economy of DREAMS (data, robotics, energy, AI, manufacturing, and space) when GenAI and large language models are employed as the “great unlock” for Musk’s companies’ power, like Tesla.

The analyst depicts how AI will move from the digital world to the physical or atomic world with autonomous vehicles and robotics. Instead of written input to the models, sensory data such as optical data are fed into the models.

Chatbots apply textual data to train foundational models, while physical robots require vision data to train their vision and language foundational models.

Jonas expects a race to collect passive visual data to create a “hyper-realistic digital twin of the physical world.” The evolution of the use of optical sensors in spectacles and vacuum cleaners to the US$10trn global transportation market could be a potential game changer.

E&P stock analysis for reporting season

NextDC ((NXT)) remains one of the broker’s top picks again in 2025, with a valuation of $29.27 compared to the FNArena average target price for daily monitored brokers of $20.117. The analyst at Evans & Partners (E&P) believes the recent weakness in the share price is due to impatient sentiment around deal announcements.

And that was before DeepSeek triggered a sell-off for AI beneficiaries around the world.

The trading pattern aligns with historic trends with a positive setup from a trading perspective for the stock as the valuation multiple has been compressed and significant construction activity in Melbourne is usually a forerunner to deal announcements.

The macro backdrop for data centre demand remains robust. E&P anticipates the prospect of “significant deals” across all three of the company’s facilities in M2, M3, and S3. The 1H25 result is believed likely to bring back investors’ attention to the stock.

Megaport ((MP1)) is highlighted as a recovery story, although there is no specification on guidance from management as to when revenue growth will significantly start to pick up. With a lot of negative news and low expectations already in the stock price, the broker doesn’t envisage a lot of downside to the price, nor is there at this stage any reason for upside risks.

The stock is valued at $14.16. FNArena’s daily monitored brokers’ average target price is $10.142.

Although the backdrop is positive for the broker, Macquarie Technology Group ((MAQ)) is expected to have a “quiet first half year” with guidance already issued and the VMware hypervisor repricing last year which impacted the company’s Cloud Managed Services margins.

As construction on IC3W advances over the year, investors are anticipated to place more attention on the potential news for a contract with a hyperscale client, the analyst explains. E&P’s valuation for the company stands at $109.43 versus the FNArena daily monitored average broker target of $100.

ASX software companies

Amidst E&P’s software coverage, none of the stocks are trading at “cheap” valuation multiples, the analyst highlights. There are near-term risks to share price performance due to the collective outperformance in 2024.

WiseTech Global ((WTC)) is the broker’s top software stock, although the analyst does not have high conviction on any of the software companies covered.

The performance of Wisetech is seen as dependent on the success or not of the new Container Optimisation product which was due to launch in 2024. Post discussions with CargoWise customers, the broker is on balance “excited” about the product and views the upside potential as significant, assuming it works out.

The new product allows the customer to reduce the cost of goods sold, which is easier for a freight forwarder to integrate and achieve efficiencies than CargoWise’s traditional scope of lowering opex-per-container, which results in the forwarder needing to restructure down the track with workforce reductions to achieve efficiencies, the broker explains.

E&P’s valuation is $153.41 compared to the FNArena daily monitored brokers’ average target price of $141.393.

Block ((XYZ)) is one of E&P’s top picks for 2025. The analyst highlights optionality for the company to beat consensus earnings from one of a few levers, including cost control for operating leverage; Square simplifying customer onboarding by unifying the Square App; the build-out of an improved sales effort; Afterpay to potentially launch a Cash App card; paycheck deposit growth; and bitcoin mining facilities. Valuation stands at $337 (not a typo).

Xero ((XRO)) stands as a “low-risk proposition” for investors from an earnings perspective for 1H25, according to the analyst, with last year’s price rises leading to upside risks on consensus earnings.

Customer checks at the end of 2024 and the start of 2025 have given way to concerns the analyst has on the stock due to complaints from domestic accountants and UK users on the price rises, which has resulted in small reductions to subscribers forecasts.

By contrast, there is more positive feedback from the US market, but this has yet to evolve into customer growth.

The valuation stands at $164m versus the FNArena daily monitored broker average target price of $196.15. The broker believes the market will dismiss subscriber growth concerns if the average revenue per user supports revenues. The consensus growth revenue forecasts for 1H25 are quite modest, and forex tailwinds should support revenue growth for FY25/FY26, the broker states.

SiteMinder ((SDR)) is likely to be more impacted by new product launches than the cyclicality of the travel sector, according to E&P. The travel industry is showing signs of momentum returning to more “normal” levels post the exuberant post-covid bounce-back, the broker explains.

From a product perspective, channel checks with the industry have been very positive and “hopefully” are a positive indication the company can meet investor expectations on new product adoption.

Valuation comes in at $7.33 against the FNArena average broker target price of $6.94.

TechnologyOne ((TNE)) comes up squeaky clean on the investment proposition and the financials for 2025; the issue remains the valuation for E&P. The company reports out of cycle and will not release earnings until May.

Post mark-to-market of forex changes, the broker’s valuation lifts 10.8% to $27.27 against the FNArena average broker target price of $29.191.

The author owns Microsoft, Nvidia, WiseTech Global, TechnologyOne, and Xero shares.

For more on these topics and robotics refer to our dedicated GenAI section on the FNArena website: https://fnarena.com/index.php/tag/gen-ai/

as well as: https://fnarena.com/index.php/2024/12/12/the-2024-gen-ai-xmas-special-incl-2025-outlook/

Find out why FNArena subscribers like the service so much: “Your Feedback (Thank You)” Warning this story contains unashamedly positive feedback on the service provided.

FNArena is proud about its track record and past achievements: Ten Years On

Click to view our Glossary of Financial Terms