A Visual Guide to How Diffusion Models Work

Stay Ahead, Stay ONMINE

A Visual Guide to How Diffusion Models Work

This article is aimed at those who want to understand exactly how Diffusion Models work, with no prior knowledge expected. I’ve tried to use illustrations wherever possible to provide visual intuitions on each part of these models. I’ve kept mathematical notation and equations to a minimum, and where they are necessary I’ve tried to define […]

Intro

I’ve framed this article around three main questions:

What exactly is it that diffusion models learn?
How and why do diffusion models work?
Once you’ve trained a model, how do you get useful stuff out of it?

The examples will be based on the glyffuser, a minimal text-to-image diffusion model that I previously implemented and wrote about. The architecture of this model is a standard text-to-image denoising diffusion model without any bells or whistles. It was trained to generate pictures of new “Chinese” glyphs from English definitions. Have a look at the picture below — even if you’re not familiar with Chinese writing, I hope you’ll agree that the generated glyphs look pretty similar to the real ones!

Random examples of glyffuser training data (left) and generated data (right).

What exactly is it that diffusion models learn?

Generative Ai models are often said to take a big pile of data and “learn” it. For text-to-image diffusion models, the data takes the form of pairs of images and descriptive text. But what exactly is it that we want the model to learn? First, let’s forget about the text for a moment and concentrate on what we are trying to generate: the images.

Probability distributions

Broadly, we can say that we want a generative AI model to learn the underlying probability distribution of the data. What does this mean? Consider the one-dimensional normal (Gaussian) distribution below, commonly written 𝒩(μ,σ²) and parameterized with mean μ = 0 and variance σ² = 1. The black curve below shows the probability density function. We can sample from it: drawing values such that over a large number of samples, the set of values reflects the underlying distribution. These days, we can simply write something like x = random.gauss(0, 1) in Python to sample from the standard normal distribution, although the computational sampling process itself is non-trivial!

Values sampled from an underlying distribution (here, the standard normal 𝒩(0,1)) can then be used to estimate the parameters of that distribution.

We could think of a set of numbers sampled from the above normal distribution as a simple dataset, like that shown as the orange histogram above. In this particular case, we can calculate the parameters of the underlying distribution using maximum likelihood estimation, i.e. by working out the mean and variance. The normal distribution estimated from the samples is shown by the dotted line above. To take some liberties with terminology, you might consider this as a simple example of “learning” an underlying probability distribution. We can also say that here we explicitly learnt the distribution, in contrast with the implicit methods that diffusion models use.

Conceptually, this is all that generative AI is doing — learning a distribution, then sampling from that distribution!

Data representations

What, then, does the underlying probability distribution of a more complex dataset look like, such as that of the image dataset we want to use to train our diffusion model?

First, we need to know what the representation of the data is. Generally, a machine learning (ML) model requires data inputs with a consistent representation, i.e. format. For the example above, it was simply numbers (scalars). For images, this representation is commonly a fixed-length vector.

The image dataset used for the glyffuser model is ~21,000 pictures of Chinese glyphs. The images are all the same size, 128 × 128 = 16384 pixels, and greyscale (single-channel color). Thus an obvious choice for the representation is a vector x of length 16384, where each element corresponds to the color of one pixel: x = (x₁,x₂,…,x₁₆₃₈₄). We can call the domain of all possible images for our dataset “pixel space”.

An example glyph with pixel values labelled (downsampled to 32 × 32 pixels for readability).

Dataset visualization

We make the assumption that our individual data samples, x, are actually sampled from an underlying probability distribution, q(x), in pixel space, much as the samples from our first example were sampled from an underlying normal distribution in 1-dimensional space. Note: the notation x ∼ q(x) is commonly used to mean: “the random variable x sampled from the probability distribution q(x).”

This distribution is clearly much more complex than a Gaussian and cannot be easily parameterized — we need to learn it with a ML model, which we’ll discuss later. First, let’s try to visualize the distribution to gain a better intution.

As humans find it difficult to see in more than 3 dimensions, we need to reduce the dimensionality of our data. A small digression on why this works: the manifold hypothesis posits that natural datasets lie on lower dimensional manifolds embedded in a higher dimensional space — think of a line embedded in a 2-D plane, or a plane embedded in 3-D space. We can use a dimensionality reduction technique such as UMAP to project our dataset from 16384 to 2 dimensions. The 2-D projection retains a lot of structure, consistent with the idea that our data lie on a lower dimensional manifold embedded in pixel space. In our UMAP, we see two large clusters corresponding to characters in which the components are arranged either horizontally (e.g. 明) or vertically (e.g. 草). An interactive version of the plot below with popups on each datapoint is linked here.

Click here for an interactive version of this plot.

Let’s now use this low-dimensional UMAP dataset as a visual shorthand for our high-dimensional dataset. Remember, we assume that these individual points have been sampled from a continuous underlying probability distribution q(x). To get a sense of what this distribution might look like, we can apply a KDE (kernel density estimation) over the UMAP dataset. (Note: this is just an approximation for visualization purposes.)

This gives a sense of what q(x) should look like: clusters of glyphs correspond to high-probability regions of the distribution. The true q(x) lies in 16384 dimensions — this is the distribution we want to learn with our diffusion model.

We showed that for a simple distribution such as the 1-D Gaussian, we could calculate the parameters (mean and variance) from our data. However, for complex distributions such as images, we need to call on ML methods. Moreover, what we will find is that for diffusion models in practice, rather than parameterizing the distribution directly, they learn it implicitly through the process of learning how to transform noise into data over many steps.

Takeaway

The aim of generative AI such as diffusion models is to learn the complex probability distributions underlying their training data and then sample from these distributions.

How and why do diffusion models work?

Diffusion models have recently come into the spotlight as a particularly effective method for learning these probability distributions. They generate convincing images by starting from pure noise and gradually refining it. To whet your interest, have a look at the animation below that shows the denoising process generating 16 samples.

In this section we’ll only talk about the mechanics of how these models work but if you’re interested in how they arose from the broader context of generative models, have a look at the further reading section below.

What is “noise”?

Let’s first precisely define noise, since the term is thrown around a lot in the context of diffusion. In particular, we are talking about Gaussian noise: consider the samples we talked about in the section about probability distributions. You could think of each sample as an image of a single pixel of noise. An image that is “pure Gaussian noise”, then, is one in which each pixel value is sampled from an independent standard Gaussian distribution, 𝒩(0,1). For a pure noise image in the domain of our glyph dataset, this would be noise drawn from 16384 separate Gaussian distributions. You can see this in the previous animation. One thing to keep in mind is that we can choose the means of these noise distributions, i.e. center them, on specific values — the pixel values of an image, for instance.

For convenience, you’ll often find the noise distributions for image datasets written as a single multivariate distribution 𝒩(0,I) where I is the identity matrix, a covariance matrix with all diagonal entries equal to 1 and zeroes elsewhere. This is simply a compact notation for a set of multiple independent Gaussians — i.e. there are no correlations between the noise on different pixels. In the basic implementations of diffusion models, only uncorrelated (a.k.a. “isotropic”) noise is used. This article contains an excellent interactive introduction on multivariate Gaussians.

Diffusion process overview

Below is an adaptation of the somewhat-famous diagram from Ho et al.’s seminal paper “Denoising Diffusion Probabilistic Models” which gives an overview of the whole diffusion process:

Diagram of the diffusion process adapted from Ho *et al*. 2020. The glyph 锂, meaning “lithium”, is used as a representative sample from the dataset.

I found that there was a lot to unpack in this diagram and simply understanding what each component meant was very helpful, so let’s go through it and define everything step by step.

We previously used x ∼ q(x) to refer to our data. Here, we’ve added a subscript, xₜ, to denote timestep t indicating how many steps of “noising” have taken place. We refer to the samples noised a given timestep as x ∼ q(xₜ). x₀ is clean data and xₜ (t = T) ∼ 𝒩(0,1) is pure noise.

We define a forward diffusion process whereby we corrupt samples with noise. This process is described by the distribution q(xₜ|xₜ₋₁). If we could access the hypothetical reverse process q(xₜ₋₁|xₜ), we could generate samples from noise. As we cannot access it directly because we would need to know x₀, we use ML to learn the parameters, θ, of a model of this process, 𝑝θ(𝑥ₜ₋₁∣𝑥ₜ). (That should be p subscript θ but medium cannot render it.)

In the following sections we go into detail on how the forward and reverse diffusion processes work.

Forward diffusion, or “noising”

Used as a verb, “noising” an image refers to applying a transformation that moves it towards pure noise by scaling down its pixel values toward 0 while adding proportional Gaussian noise. Mathematically, this transformation is a multivariate Gaussian distribution centered on the pixel values of the preceding image.

In the forward diffusion process, this noising distribution is written as q(xₜ|xₜ₋₁) where the vertical bar symbol “|” is read as “given” or “conditional on”, to indicate the pixel means are passed forward from q(xₜ₋₁) At t = T where T is a large number (commonly 1000) we aim to end up with images of pure noise (which, somewhat confusingly, is also a Gaussian distribution, as discussed previously).

The marginal distributions q(xₜ) represent the distributions that have accumulated the effects of all the previous noising steps (marginalization refers to integration over all possible conditions, which recovers the unconditioned distribution).

Since the conditional distributions are Gaussian, what about their variances? They are determined by a variance schedule that maps timesteps to variance values. Initially, an empirically determined schedule of linearly increasing values from 0.0001 to 0.02 over 1000 steps was presented in Ho et al. Later research by Nichol & Dhariwal suggested an improved cosine schedule. They state that a schedule is most effective when the rate of information destruction through noising is relatively even per step throughout the whole noising process.

Forward diffusion intuition

As we encounter Gaussian distributions both as pure noise q(xₜ, t = T) and as the noising distribution q(xₜ|xₜ₋₁), I’ll try to draw the distinction by giving a visual intuition of the distribution for a single noising step, q(x₁∣x₀), for some arbitrary, structured 2-dimensional data:

Each noising step q(xₜ|xₜ₋₁) is a Gaussian distribution conditioned on the previous step.

The distribution q(x₁∣x₀) is Gaussian, centered around each point in x₀, shown in blue. Several example points x₀⁽ⁱ⁾ are picked to illustrate this, with q(x₁∣x₀ = x₀⁽ⁱ⁾) shown in orange.

In practice, the main usage of these distributions is to generate specific instances of noised samples for training (discussed further below). We can calculate the parameters of the noising distributions at any timestep t directly from the variance schedule, as the chain of Gaussians is itself also Gaussian. This is very convenient, as we don’t need to perform noising sequentially—for any given starting data x₀⁽ⁱ⁾, we can calculate the noised sample xₜ⁽ⁱ⁾ by sampling from q(xₜ∣x₀ = x₀⁽ⁱ⁾) directly.

Forward diffusion visualization

Let’s now return to our glyph dataset (once again using the UMAP visualization as a visual shorthand). The top row of the figure below shows our dataset sampled from distributions noised to various timesteps: xₜ ∼ q(xₜ). As we increase the number of noising steps, you can see that the dataset begins to resemble pure Gaussian noise. The bottom row visualizes the underlying probability distribution q(xₜ).

The dataset xₜ (above) sampled from its probability distribution q(xₜ) (below) at different noising timesteps.

Reverse diffusion overview

It follows that if we knew the reverse distributions q(xₜ₋₁∣xₜ), we could repeatedly subtract a small amount of noise, starting from a pure noise sample xₜ at t = T to arrive at a data sample x₀ ∼ q(x₀). In practice, however, we cannot access these distributions without knowing x₀ beforehand. Intuitively, it’s easy to make a known image much noisier, but given a very noisy image, it’s much harder to guess what the original image was.

So what are we to do? Since we have a large amount of data, we can train an ML model to accurately guess the original image that any given noisy image came from. Specifically, we learn the parameters θ of an ML model that approximates the reverse noising distributions, pθ(xₜ₋₁ ∣ xₜ) for t = 0, …, T. In practice, this is embodied in a single noise prediction model trained over many different samples and timesteps. This allows it to denoise any given input, as shown in the figure below.

The ML model predicts added noise at any given timestep t.

Next, let’s go over how this noise prediction model is implemented and trained in practice.

How the model is implemented

First, we define the ML model — generally a deep neural network of some sort — that will act as our noise prediction model. This is what does the heavy lifting! In practice, any ML model that inputs and outputs data of the correct size can be used; the U-net, an architecture particularly suited to learning images, is what we use here and frequently chosen in practice. More recent models also use vision transformers.

We use the U-net architecture (Ronneberger *et al*. 2015) for our ML noise prediction model. We train the model by minimizing the difference between predicted and actual noise.

Then we run the training loop depicted in the figure above:

We take a random image from our dataset and noise it to a random timestep tt. (In practice, we speed things up by doing many examples in parallel!)
We feed the noised image into the ML model and train it to predict the (known to us) noise in the image. We also perform timestep conditioning by feeding the model a timestep embedding, a high-dimensional unique representation of the timestep, so that the model can distinguish between timesteps. This can be a vector the same size as our image directly added to the input (see here for a discussion of how this is implemented).
The model “learns” by minimizing the value of a loss function, some measure of the difference between the predicted and actual noise. The mean square error (the mean of the squares of the pixel-wise difference between the predicted and actual noise) is used in our case.
Repeat until the model is well trained.

Note: A neural network is essentially a function with a huge number of parameters (on the order of 10⁶ for the glyffuser). Neural network ML models are trained by iteratively updating their parameters using backpropagation to minimize a given loss function over many training data examples. This is an excellent introduction. These parameters effectively store the network’s “knowledge”.

A noise prediction model trained in this way eventually sees many different combinations of timesteps and data examples. The glyffuser, for example, was trained over 100 epochs (runs through the whole data set), so it saw around 2 million data samples. Through this process, the model implicity learns the reverse diffusion distributions over the entire dataset at all different timesteps. This allows the model to sample the underlying distribution q(x₀) by stepwise denoising starting from pure noise. Put another way, given an image noised to any given level, the model can predict how to reduce the noise based on its guess of what the original image. By doing this repeatedly, updating its guess of the original image each time, the model can transform any noise to a sample that lies in a high-probability region of the underlying data distribution.

Reverse diffusion in practice

We can now revisit this video of the glyffuser denoising process. Recall a large number of steps from sample to noise e.g. T = 1000 is used during training to make the noise-to-sample trajectory very easy for the model to learn, as changes between steps will be small. Does that mean we need to run 1000 denoising steps every time we want to generate a sample?

Luckily, this is not the case. Essentially, we can run the single-step noise prediction but then rescale it to any given step, although it might not be very good if the gap is too large! This allows us to approximate the full sampling trajectory with fewer steps. The video above uses 120 steps, for instance (most implementations will allow the user to set the number of sampling steps).

Recall that predicting the noise at a given step is equivalent to predicting the original image x₀, and that we can access the equation for any noised image deterministically using only the variance schedule and x₀. Thus, we can calculate xₜ₋ₖ based on any denoising step. The closer the steps are, the better the approximation will be.

Too few steps, however, and the results become worse as the steps become too large for the model to effectively approximate the denoising trajectory. If we only use 5 sampling steps, for example, the sampled characters don’t look very convincing at all:

There is then a whole literature on more advanced sampling methods beyond what we’ve discussed so far, allowing effective sampling with much fewer steps. These often reframe the sampling as a differential equation to be solved deterministically, giving an eerie quality to the sampling videos — I’ve included one at the end if you’re interested. In production-level models, these are usually preferred over the simple method discussed here, but the basic principle of deducing the noise-to-sample trajectory is the same. A full discussion is beyond the scope of this article but see e.g. this paper and its corresponding implementation in the Hugging Face diffusers library for more information.

Alternative intuition from score function

To me, it was still not 100% clear why training the model on noise prediction generalises so well. I found that an alternative interpretation of diffusion models known as “score-based modeling” filled some of the gaps in intuition (for more information, refer to Yang Song’s definitive article on the topic.)

The dataset xₜ sampled from its probability distribution q(xₜ) at different noising timesteps; below, we add the score function ∇ₓ log q(xₜ).

I try to give a visual intuition in the bottom row of the figure above: essentially, learning the noise in our diffusion model is equivalent (to a constant factor) to learning the score function, which is the gradient of the log of the probability distribution: ∇ₓ log q(x). As a gradient, the score function represents a vector field with vectors pointing towards the regions of highest probability density. Subtracting the noise at each step is then equivalent to moving following the directions in this vector field towards regions of high probability density.

As long as there is some signal, the score function effectively guides sampling, but in regions of low probability it tends towards zero as there is little to no gradient to follow. Using many steps to cover different noise levels allows us to avoid this, as we smear out the gradient field at high noise levels, allowing sampling to converge even if we start from low probability density regions of the distribution. The figure shows that as the noise level is increased, more of the domain is covered by the score function vector field.

Summary

The aim of diffusion models is learn the underlying probability distribution of a dataset and then be able to sample from it. This requires forward and reverse diffusion (noising) processes.
The forward noising process takes samples from our dataset and gradually adds Gaussian noise (pushes them off the data manifold). This forward process is computationally efficient because any level of noise can be added in closed form a single step.
The reverse noising process is challenging because we need to predict how to remove the noise at each step without knowing the original data point in advance. We train a ML model to do this by giving it many examples of data noised at different timesteps.
Using very small steps in the forward noising process makes it easier for the model to learn to reverse these steps, as the changes are small.
By applying the reverse noising process iteratively, the model refines noisy samples step by step, eventually producing a realistic data point (one that lies on the data manifold).

Takeaway

Diffusion models are a powerful framework for learning complex data distributions. The distributions are learnt implicitly by modelling a sequential denoising process. This process can then be used to generate samples similar to those in the training distribution.

Once you’ve trained a model, how do you get useful stuff out of it?

Earlier uses of generative AI such as “This Person Does Not Exist” (ca. 2019) made waves simply because it was the first time most people had seen AI-generated photorealistic human faces. A generative adversarial network or “GAN” was used in that case, but the principle remains the same: the model implicitly learnt a underlying data distribution — in that case, human faces — then sampled from it. So far, our glyffuser model does a similar thing: it samples randomly from the distribution of Chinese glyphs.

The question then arises: can we do something more useful than just sample randomly? You’ve likely already encountered text-to-image models such as Dall-E. They are able to incorporate extra meaning from text prompts into the diffusion process — this in known as conditioning. Likewise, diffusion models for scientific scientific applications like protein (e.g. Chroma, RFdiffusion, AlphaFold3) or inorganic crystal structure generation (e.g. MatterGen) become much more useful if can be conditioned to generate samples with desirable properties such as a specific symmetry, bulk modulus, or band gap.

Conditional distributions

We can consider conditioning as a way to guide the diffusion sampling process towards particular regions of our probability distribution. We mentioned conditional distributions in the context of forward diffusion. Below we show how conditioning can be thought of as reshaping a base distribution.

A simple example of a joint probability distribution p(x, y), shown as a contour map, along with its two marginal 1-D probability distributions, p(x) and p(y). The highest points of p(x, y) are at (x₁, y₁) and (x₂, y₂). The conditional distributions p(x∣y = y₁) and p(x∣y = y₂) are shown overlaid on the main plot.

Consider the figure above. Think of p(x) as a distribution we want to sample from (i.e., the images) and p(y) as conditioning information (i.e., the text dataset). These are the marginal distributions of a joint distribution p(x, y). Integrating p(x, y) over y recovers p(x), and vice versa.

Sampling from p(x), we are equally likely to get x₁ or x₂. However, we can condition on p(y = y₁) to obtain p(x∣y = y₁). You can think of this as taking a slice through p(x, y) at a given value of y. In this conditioned distribution, we are much more likely to sample at x₁ than x₂.

In practice, in order to condition on a text dataset, we need to convert the text into a numerical form. We can do this using large language model (LLM) embeddings that can be injected into the noise prediction model during training.

Embedding text with an LLM

In the glyffuser, our conditioning information is in the form of English text definitions. We have two requirements: 1) ML models prefer fixed-length vectors as input. 2) The numerical representation of our text must understand context — if we have the words “lithium” and “element” nearby, the meaning of “element” should be understood as “chemical element” rather than “heating element”. Both of these requirements can be met by using a pre-trained LLM.

The diagram below shows how an LLM converts text into fixed-length vectors. The text is first tokenized (LLMs break text into tokens, small chunks of characters, as their basic unit of interaction). Each token is converted into a base embedding, which is a fixed-length vector of the size of the LLM input. These vectors are then passed through the pre-trained LLM (here we use the encoder portion of Google’s T5 model), where they are imbued with additional contextual meaning. We end up with a array of n vectors of the same length d, i.e. a (n, d) sized tensor.

We can convert text to a numerical embedding imbued with contextual meaning using a pre-trained LLM.

Note: in some models, notably Dall-E, additional image-text alignment is performed using contrastive pretraining. Imagen seems to show that we can get away without doing this.

Training the diffusion model with text conditioning

The exact method that this embedding vector is injected into the model can vary. In Google’s Imagen model, for example, the embedding tensor is pooled (combined into a single vector in the embedding dimension) and added into the data as it passes through the noise prediction model; it is also included in a different way using cross-attention (a method of learning contextual information between sequences of tokens, most famously used in the transformer models that form the basis of LLMs like ChatGPT).

Conditioning information can be added *via* multiple different methods but the training loss remains the same.

In the glyffuser, we only use cross-attention to introduce this conditioning information. While a significant architectural change is required to introduce this additional information into the model, the loss function for our noise prediction model remains exactly the same.

Testing the conditioned diffusion model

Let’s do a simple test of the fully trained conditioned diffusion model. In the figure below, we try to denoise in a single step with the text prompt “Gold”. As touched upon in our interactive UMAP, Chinese characters often contain components known as radicals which can convey sound (phonetic radicals) or meaning (semantic radicals). A common semantic radical is derived from the character meaning “gold”, “金”, and is used in characters that are in some broad sense associated with gold or metals.

Even with a single sampling step, conditioning guides denoising towards the relevant regions of the probability distribution.

The figure shows that even though a single step is insufficient to approximate the denoising trajectory very well, we have moved into a region of our probability distribution with the “金” radical. This indicates that the text prompt is effectively guiding our sampling towards a region of the glyph probability distribution related to the meaning of the prompt. The animation below shows a 120 step denoising sequence for the same prompt, “Gold”. You can see that every generated glyph has either the 釒 or 钅 radical (the same radical in traditional and simplified Chinese, respectively).

Takeaway

Conditioning enables us to sample meaningful outputs from diffusion models.

Further remarks

I found that with the help of tutorials and existing libraries, it was possible to implement a working diffusion model despite not having a full understanding of what was going on under the hood. I think this is a good way to start learning and highly recommend Hugging Face’s tutorial on training a simple diffusion model using their diffusers Python library (which now includes my small bugfix!).

I’ve omitted some topics that are crucial to how production-grade diffusion models function, but are unnecessary for core understanding. One is the question of how to generate high resolution images. In our example, we did everything in pixel space, but this becomes very computationally expensive for large images. The general approach is to perform diffusion in a smaller space, then upscale it in a separate step. Methods include latent diffusion (used in Stable Diffusion) and cascaded super-resolution models (used in Imagen). Another topic is classifier-free guidance, a very elegant method for boosting the conditioning effect to give much better prompt adherence. I show the implementation in my previous post on the glyffuser and highly recommend this article if you want to learn more.

Fun extras

Diffusion sampling using the DPMSolverSDEScheduler developed by Katherine Crowson and implemented in Hugging Face diffusers—note the smooth transition from noise to data.

Stay Ahead

Explore More Insights

Stay ahead with more perspectives on cutting-edge power, infrastructure, energy, bitcoin and AI solutions. Explore these articles to uncover strategies and insights shaping the future of industries.

Broadcom strikes chip deals with Google, Anthropic

Anthropic said this week that the AI startup’s annual revenue run rate has now crossed $30 billion, up from about $9 billion the previous year. “We are making our most significant compute commitment to date to keep pace with our unprecedented growth,” said Krishna Rao, CFO of Anthropic, in a

Google owns the most AI compute, and it built it its way

Google, for its part, will continue to be “one of the largest, if not the largest,” consumer of compute resources, said Bill Wong, research fellow at Info-Tech Research Group. “Its business model drives that global demand, specifically through the widespread use of Google search and Gemini, which it provides for

Cisco to acquire Galileo for AI observability

“We have this opportunity to be a trust layer, not just for … network activity, but actually what’s happening at the application layer, at the workload layer, between agents, between workloads, between data,” Peter Bailey, senior vice president and general manager of Cisco’s security business, told Network World last month.

Lumen: Upstream network visibility is enterprise security’s new front line

“Threat actors can, for pennies, cycle through IP addresses to get fresh positive-reputation IP addresses minute to minute and use them for their wares at various parts of the attack lifecycle, whether it’s brute forcing, whether it’s leveraging known positive credentials, or whether it’s exfiltrating information at the end of

Energy Department Continues Initiating Strategic Petroleum Reserve Emergency Exchanges

WASHINGTON — The U.S. Department of Energy (DOE) today issued a Request for Proposal (RFP) for an emergency exchange of up to 30 million barrels of sweet crude oil from the Strategic Petroleum Reserve (SPR). This action continues DOE’s execution of the 172-million-barrel release President Trump authorized as part of a coordinated release of 400 million barrels from IEA member nations’ strategic reserves. The Department will continue issuing RFPs to deliver on President Trump’s commitment to the coordinated release, addressing global supply disruptions and strengthening energy security for the United States and its allies. The crude oil will originate from the SPR’s West Hackberry site. This action builds on the Department’s two previous Emergency Exchange RFPs, which together quickly moved to award approximately 55 million barrels from the Bayou Choctaw, Bryan Mound, and West Hackberry SPR sites. These earlier exchanges demonstrated the SPR’s ability to rapidly deliver crude oil to the market under emergency authorities. Today’s announcement of the 30-million-barrel exchange leverages the full capabilities of the SPR, alongside the President’s limited Jones Act waiver, to accelerate critical near-term oil flows into the market and strengthen supply resilience. “Today’s action strengthens efforts to move oil quickly to market and address short-term supply pressures,” said Kyle Haustveit, Assistant Secretary of the Hydrocarbons and Geothermal Energy Office. “With two earlier exchanges, we are building on proven momentum. Thanks to President Trump, we are managing the SPR responsibly, returning premium barrels later to bolster long-term readiness at no cost to taxpayers.” Under DOE’s exchange authority, participating companies will return the borrowed 30-million-barrels with additional premium barrels by next year. This exchange delivers immediate crude to refiners and the market while generating additional barrels for the American people at no cost to taxpayers. Bids for this solicitation are due no later than 11:00 A.M. Central Time on Monday, April 13, 2026. For more information on the SPR, please visit DOE’s website. 

DOE’s Office of Energy Dominance Financing Announces Conditional Commitment for a Domestic Medical Isotope Manufacturing Facility

WASHINGTON—The U.S. Department of Energy’s (DOE) Office of Energy Dominance Financing (EDF) issued a conditional commitment for a loan of up to $263 million to SHINE Chrysalis, LLC (SHINE) to support the construction of Chrysalis, a high-volume medical isotope production facility in Janesville, Wisconsin. The production facility will create a supply of reliable and secure medical isotopes made with fusion and fission technology. Millions of patients rely on medical isotopes for life-saving diagnostic imaging and cancer treatments, but most of that supply comes from overseas with limited capacity to meet demand. Utilizing fusion and fission technology, the facility will establish the only domestic commercial supply of molybdenum-99 (Mo-99). Mo-99 is a life-saving medical isotope essential for diagnostic imaging; its decay product of technetium-99m (Tc-99m) is used in over 40,000 medical procedures in the United States each day. “The SHINE Chrysalis project is vital to improving the nuclear supply chain and contributing to a strong next-generation nuclear workforce while onshoring this critical production and improving national security,” EDF Director Gregory A. Beard said. “Using EDF’s loan authority to further commercialize a project long supported by DOE is Trump’s policy at work: ensuring a reliable and secure domestic supply chain while lowering costs.” Thanks to President Trump and the Working Families Tax Cut, today’s announcement highlights the Energy Department’s role in advancing President Trump’s Executive Order, Reinvigorating the Nuclear Industrial Base, by maximizing the use of uranium through an innovative recycling process. This project will support hundreds of construction and operation jobs, contributing to a strong next-generation American nuclear workforce. SHINE has developed and demonstrated its technology through significant support from the National Nuclear Security Administration (NNSA) over the last 16 years. With NNSA funding, the National Laboratories also played a key role in developing SHINE’s technology. “SHINE is the key to

United States and Australia meet for Mining, Minerals and Metals Investment Ministerial

We, the Australian Minister for Resources and Northern Australia, the Hon Madeleine King MP, and Secretaries and senior representatives from the United States, including Secretary of Interior Doug Burgum, Administrator of the U.S. Environmental Protection Agency Lee Zeldin, Chairman of the U.S. Export Import Bank John Jovanovic, and Assistant Secretary of Energy Audrey Robertson, held our inaugural Mining, Minerals, and Metals Investment Ministerial in Tokyo on 14 March 2026, to advance cooperation under the landmark bilateral agreement, the United States–Australia Framework for Securing Supply in the Mining and Processing of Critical Minerals and Rare Earths (the Framework). Under the Framework, Australia and the United States are delivering concrete outcomes to strengthen, secure, and diversify critical minerals and rare earth supply chains. Within six months of agreement of the Framework, we have each taken measures to provide at least USD $1 billion in financing to key critical minerals projects. By mobilising government and private sector capital, these investments support the development of our shared industrial base and strengthen longer term supply for defence, manufacturing, and energy supply chains. To build resilience, enhance stability, and bolster economic security in support of our shared critical minerals interests, Australia and the United States today announce the establishment of the Critical Minerals Supply Security Response Group and commit to deeper cooperation between our key agencies. In line with the Framework, the Critical Minerals Supply Security Response Group, led by senior representatives from the United States Department of Energy and the Australian Department of Industry, Science and Resources, will cooperate on priority minerals and supply chain vulnerabilities and coordinate efforts to accelerate the delivery of processed minerals under the Framework. Australia and the United States also commit to leveraging shared policy and interagency regulatory tools and, where appropriate, investments to secure critical minerals supply, including through cooperation between Australia’s

Energy Security for Indo-Pacific Endurance, a Global Growth Center of the 21st Century

We, the ministers and representatives of Australia, Bangladesh, Brunei, Japan, Malaysia, New Zealand, Philippines, Republic of Korea, Singapore, Timor-Leste, United States, and Vietnam met in Tokyo, Japan, on March 14–15, 2026, to hold the historic Indo-Pacific Energy Security Ministerial and Business Forum. The forum was co-hosted by the Chair and Vice Chair of the U.S. National Energy Dominance Council, Secretary of the Interior Doug Burgum, Secretary of Energy Chris Wright, and Japanese Minister of Economy, Trade and Industry Akazawa Ryosei. We affirm our shared determination to work collectively to ensure stable and secure energy supply in the Indo-Pacific region. To this end, we focused on three key themes: reliable energy for Indo-Pacific growth and security; securing energy supply chains, infrastructure and maritime routes; and enabling trade and investment. To support these goals, Ministers affirm the value of: The necessity of reliable, affordable, secure and dispatchable energy from all sources depending on each country’s situation, in meeting the region’s surging energy demand. Promoting quality as a key procurement mechanism to mitigate risk of operational liabilities. Protecting against rising cyber threats to the security of the energy grid, critical infrastructure, vehicles, and devices. Investment in comprehensive energy infrastructure that encompasses the entire energy supply chain from upstream development facilities to downstream equipment to support an affordable, reliable, and secure energy supply including baseload electricity. Continuing to supply affordable and reliable energy sources in the Indo-Pacific region, including through emergency response measures, to benefit both producers and consumer countries. While maintaining strong relations with current partners, expanding and diversifying energy suppliers and fuel types in order to strengthen energy security. Promoting transparent, long-term energy contracts that reduce market volatility. As the global economy expands, so too does demand for energy driven by AI and electrification, we, as countries committed to a free and

Energy Department Issues Funding Opportunity to Strengthen American Critical Minerals and Materials Supply Chain

WASHINGTON—The U.S. Department of Energy’s (DOE) Office of Critical Minerals and Energy Innovation (CMEI) and Hydrocarbons and Geothermal Energy Office (HGEO) today announced a funding opportunity of up to $69 million for technologies or processes that advance the domestic production and refining of critical materials. Projects selected through this Notice of Funding Opportunity (NOFO) will address the greatest technical obstacles to a stronger critical materials supply chain. “This funding will help establish a more secure and affordable supply of the critical minerals and materials that are foundational to American energy dominance, national security, and industrial competitiveness,” said Assistant Secretary of Energy (EERE) Audrey Robertson. DOE is seeking projects that bridge the gap between bench-scale innovations and commercially viable technologies. Selected project teams will form industry-led partnerships and conduct research and development with support from the U.S. national laboratories. The NOFO, which is part of DOE’s Critical Minerals and Materials Accelerator Program and jointly funded by CMEI’s Advanced Materials and Manufacturing Technologies Office and HGEO’s Office of Geothermal, has three primary topic areas: Production and material efficiency for critical materials including rare earth elements Processes to refine and alloy gallium, gallium nitride, germanium, and silicon carbide Cost-competitive direct lithium extraction, separation, and processing CMEI will host an informational webinar on April 16, 2026, to discuss the NOFO and application requirements. Letters of intent are due on April 21, 2026, by 5 p.m. ET. Deadlines for full applications will be staggered based on topic area, starting in May 2026. For more details on sub-topics and deadlines, visit the NOFO landing page. The Critical Minerals and Materials Accelerator is one of several programs developed through DOE’s Critical Materials Collaborative. This NOFO is part of $1 billion in critical materials funding announced by DOE in August 2025, and follows the Manufacturing Deployment Office’s announcement

Latin America returns to the energy security conversation at CERAWeek

With geopolitical risk central to conversations about energy, and with long-cycle supply once again in focus, Latin America’s mix of hydrocarbons and export potential drew renewed attention at CERAWeek by S&P Global in Houston. Argentina, resource story to export platform Among the regional stories, Argentina stood out as Vaca Muerta was no longer discussed simply as a large unconventional resource, but whether the country could turn resource quality into sustained export capacity. Country officials talked about scale: more operators, more services, more infrastructure, and a larger industrial base around the unconventional play. Daniel González, Vice Minister of Energy and Mining for Argentina, put it plainly: “The time has come to expand the Vaca Muerta ecosystem.” What is at stake now is not whether the basin works, but whether the country can build enough above-ground capacity and regulatory consistency to keep development moving. Horacio Marín, chairman and chief executive officer of YPF, offered an expansive version of that argument. He said Argentina’s energy exports could reach $50 billion/year by 2031, backed by roughly $130 billion in cumulative investment in oil, LNG, and transportation infrastructure. He said Argentine crude output could reach 1 million b/d by end-2026. He said Argentina wants to be seen less as a recurrent frontier story and more as a future supplier with scale. “The time to invest in Vaca Muerta is now,” Marín said. The LNG piece is starting to take shape. Eni, YPF, and XRG signed a joint development agreement in February to move Argentina LNG forward, with a first phase planned at 12 million tonnes/year. Southern Energy—backed by PAE, YPF, Pampa Energía, Harbour Energy, and Golar LNG—holds a long-term agreement with SEFE for 2 million tonnes/year over 8 years. The movement by global standards is early-stage and relatively modest, but it adds to Argentina’s export

OpenAI puts part of Stargate project on hold over runaway power costs

OpenAI has postponed plans to open one of the data centers central to its Stargate project. It announced its plan to open the data center in the UK with great fanfare last September, when it was regarded as a major boost for the country’s nascent AI industry, as well as proving a step up for OpenAI’s international credentials. At the time, Sam Altman, CEO of OpenAI, said, “The UK has been a longstanding pioneer of AI, and is now home to world-class researchers, millions of ChatGPT users, and a government that quickly recognized the potential of this technology.” All of that has been quietly forgotten. The plans for the data center in Northumberland, in the Northeast of England, have been put on hold, with the project ready to be revived when the conditions are ripe for major infrastructure investment, according to a report by the BBC.

Neoclouds gain momentum in a supply-constrained world

And since they used the same hardware, both neoclouds and traditional cloud providers are subject to the same shortage problem. Component suppliers are reporting significant shortages due to demand for AI data centers and Synergy sees neoclouds also experiencing delays just like traditional cloud providers. “Demand is currently outstripping supply,” said Dinsmore. “It will take a while before that starts to come into more balance.” Among neoclouds, CoreWeave stands out as the most direct challenger to traditional hyperscale cloud providers. Meanwhile, OpenAI and Anthropic represent a distinct but increasingly important category, and that is platform-centric providers offering cloud-like access to foundational models and AI development environments. Synergy says that as demand for AI infrastructure accelerates, neoclouds are positioning themselves as focused alternatives to traditional hyperscale providers such as Amazon, Microsoft and Google.

What is AI networking? How it adds intelligence to your infrastructure

The end goal is to make networks more reliable, efficient and performant. Enterprises are already seeing notable results when AI is applied to IT operations, including shorter deployment times, a decrease in trouble tickets, and faster time to resolution. With the help of AI, networks will become more autonomous and self-healing (that is, able to address issues without the need for human intervention). In fact, Tier 1 and Tier 2 infrastructure is moving toward ‘no human in the loop,’ Nick Lippis, co-founder and co-chair of enterprise user community ONUG, recently told Network World. In time, humans will only need to step in for policy exceptions and high-risk decisions. “Layering in AI capabilities makes LAN management applications easier to use and more accessible across an organization,” Dell’Oro Group analyst Sian Morgan said. Gartner predicts that, by 2030, AI agents will drive most network activities, up from “minimal adoption” in 2025. The firm emphasizes that leaders who overlook the AI networking shift “risk higher MTTR [meantime to repair], rising costs, and growing security exposure.” The core components of AI networking It’s important to note that the use of AI and machine learning (ML) in network management is not new. AI for IT operations (AIOps), for instance, is a common practice that uses automation to improve broader IT operations. AI networking is specific to the network itself, covering domains including multi-cloud software, wired and wireless LAN, data center switching, SD-WAN and managed network services (MNS). The incorporation of generative AI, in particular, has brought AI networking to the fore, as enterprise leaders are rethinking every single aspect of their business, networking included.

Aria Networks raises $125M and debuts its approach for AI-optimized networks

That embedded telemetry feeds adaptive tuning of Dynamic Load Balancing parameters, Data Center Quantized Congestion Notification (DCQCN) and failover logic without waiting for a threshold breach or a manual intervention. The platform architecture is layered. At the lowest levels, agents react in microseconds to link-level events such as transceiver flaps, rerouting leaf-spine traffic in milliseconds. At higher layers, agents make more strategic decisions about flow placement across the cluster. At the cloud layer, a large language model-based agent surfaces correlated insights to operators in natural language, allowing them to ask questions about specific jobs or alert conditions and receive context-aware responses. Karam argued that simply bolting an LLM onto an existing architecture does not deliver the same result. “If you ask it to do anything, it could hallucinate and bring down the network,” he said. “It doesn’t have any of the context or the data that’s required for this approach to be made safe.” Aria also exposes an MCP server, allowing external systems such as job schedulers and LLM routers to query network state directly and integrate it into their own decision-making. MFU and token efficiency as the target metrics Traditional networking is often evaluated in terms of bandwidth and latency. Aria is centering its platform around two metrics: Model FLOPS Utilization (MFU) and token efficiency. MFU is defined as the ratio of achieved FLOPS per accelerator to the theoretical peak. In practice, Karam said, MFU for training workloads typically runs between 33% and 45%, and inference often comes in below 30%. “The network has a major impact on the MFU, and therefore the token efficiency, because the network touches every aspect, every other component in your cluster,” Karam said.

New v2 UALink specification aims to catch up to NVLink

But given there are no products currently available using UALink 1.0, UALink 2.0 might be viewed as a premature launch Need to play catch up David Harold, senior analyst with Jon Peddie Research, was guarded in his reaction. “While 2.0 is a significant step forward from 1.0, we need to bear in mind that even 1.0 solutions aren’t shipping yet – they aren’t due until later this year. So, Nvidia is way ahead of the open alternatives on connectivity, indeed ahead of the proprietary or Ethernet based solutions too,” he said. What this means, he added, is that non-Nvidia alternatives are currently lagging in the market. “They need to play catch up on several fronts, not just networking. … I can’t think of a single shipping product that meaningfully has advantages over a Nvidia solution,” he said. “Ultimately UALink remains desirable since it will enable heterogeneous, multi-vendor environments but it’s quite a way behind NVLink today. ” There are plenty of signs that organizations will find it hard to break free of the Nvidia dominance, however. A couple of months ago, RISC-V pioneer SiFive signed a deal with Nvidia to incorporate Nvidia NVLink Fusion into its data center products, a departure for RISC companies. According to Harold, other companies could be joining it. “Custom ASIC company MediaTek is an NVLink partner, and they told me last week that they are planning to integrate it directly into next-generation custom silicon for AI applications,” he said. “This will enable a wider range of companies to use NVLink as their high-speed interconnect.” Other options And, Harold noted, Nvidia is already looking at other options. “Nvidia is now shifting to look at the copper limit for networking speed, with an interest in using optical connectivity instead,” said Harold.

Nvidia’s SchedMD acquisition puts open-source AI scheduling under scrutiny

Is the concern valid? Dr. Danish Faruqui, CEO of Fab Economics, a US-based AI hardware and datacenter advisory, said the risk was real. “The skepticism that Nvidia may prioritize its own hardware in future software updates, potentially delaying or under-optimizing support for rivals, is a feasible outcome,” he said. As the primary developer, Nvidia now controls Slurm’s official development roadmap and code review process, Faruqui said, “which could influence how quickly competing chips are integrated on new development or continuous improvement elements.” Owning the control plane alongside GPUs and networking infrastructure such as InfiniBand, he added, allows Nvidia to create a tightly vertically integrated stack that can lead to what he described as “shallow moats, where advanced features are only available or performant on Nvidia hardware.” One concrete test of that, industry observers say, will be how quickly Nvidia integrates support for AMD’s next-generation chips into Slurm’s codebase compared with how quickly it integrates its own forthcoming hardware and networking technologies, such as InfiniBand. Does the Bright Computing precedent hold? Analysts point to Nvidia’s 2022 acquisition of Bright Computing as a reference point, saying the software became optimized for Nvidia chips in ways that disadvantaged users of competing hardware. Nvidia disputed that characterization, saying Bright Computing supports “nearly any CPU or GPU-accelerated cluster.” Rawat said the comparison was instructive but imperfect. “Nvidia’s acquisition of Bright Computing highlights its preference for vertical integration, embedding Bright tightly into DGX and AI Factory stacks rather than maintaining a neutral, multi-vendor orchestration role,” he said. “This reflects a broader strategic pattern — Nvidia seeks to control the full-stack AI infrastructure experience.”

Microsoft will invest $80B in AI data centers in fiscal 2025

And Microsoft isn’t the only one that is ramping up its investments into AI-enabled data centers. Rival cloud service providers are all investing in either upgrading or opening new data centers to capture a larger chunk of business from developers and users of large language models (LLMs). In a report published in October 2024, Bloomberg Intelligence estimated that demand for generative AI would push Microsoft, AWS, Google, Oracle, Meta, and Apple would between them devote $200 billion to capex in 2025, up from $110 billion in 2023. Microsoft is one of the biggest spenders, followed closely by Google and AWS, Bloomberg Intelligence said. Its estimate of Microsoft’s capital spending on AI, at $62.4 billion for calendar 2025, is lower than Smith’s claim that the company will invest $80 billion in the fiscal year to June 30, 2025. Both figures, though, are way higher than Microsoft’s 2020 capital expenditure of “just” $17.6 billion. The majority of the increased spending is tied to cloud services and the expansion of AI infrastructure needed to provide compute capacity for OpenAI workloads. Separately, last October Amazon CEO Andy Jassy said his company planned total capex spend of $75 billion in 2024 and even more in 2025, with much of it going to AWS, its cloud computing division.

John Deere unveils more autonomous farm machines to address skill labor shortage

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More Self-driving tractors might be the path to self-driving cars. John Deere has revealed a new line of autonomous machines and tech across agriculture, construction and commercial landscaping. The Moline, Illinois-based John Deere has been in business for 187 years, yet it’s been a regular as a non-tech company showing off technology at the big tech trade show in Las Vegas and is back at CES 2025 with more autonomous tractors and other vehicles. This is not something we usually cover, but John Deere has a lot of data that is interesting in the big picture of tech. The message from the company is that there aren’t enough skilled farm laborers to do the work that its customers need. It’s been a challenge for most of the last two decades, said Jahmy Hindman, CTO at John Deere, in a briefing. Much of the tech will come this fall and after that. He noted that the average farmer in the U.S. is over 58 and works 12 to 18 hours a day to grow food for us. And he said the American Farm Bureau Federation estimates there are roughly 2.4 million farm jobs that need to be filled annually; and the agricultural work force continues to shrink. (This is my hint to the anti-immigration crowd). John Deere’s autonomous 9RX Tractor. Farmers can oversee it using an app. While each of these industries experiences their own set of challenges, a commonality across all is skilled labor availability. In construction, about 80% percent of contractors struggle to find skilled labor. And in commercial landscaping, 86% of landscaping business owners can’t find labor to fill open positions, he said. “They have to figure out how to do

2025 playbook for enterprise AI success, from agents to evals

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More 2025 is poised to be a pivotal year for enterprise AI. The past year has seen rapid innovation, and this year will see the same. This has made it more critical than ever to revisit your AI strategy to stay competitive and create value for your customers. From scaling AI agents to optimizing costs, here are the five critical areas enterprises should prioritize for their AI strategy this year. 1. Agents: the next generation of automation AI agents are no longer theoretical. In 2025, they’re indispensable tools for enterprises looking to streamline operations and enhance customer interactions. Unlike traditional software, agents powered by large language models (LLMs) can make nuanced decisions, navigate complex multi-step tasks, and integrate seamlessly with tools and APIs. At the start of 2024, agents were not ready for prime time, making frustrating mistakes like hallucinating URLs. They started getting better as frontier large language models themselves improved. “Let me put it this way,” said Sam Witteveen, cofounder of Red Dragon, a company that develops agents for companies, and that recently reviewed the 48 agents it built last year. “Interestingly, the ones that we built at the start of the year, a lot of those worked way better at the end of the year just because the models got better.” Witteveen shared this in the video podcast we filmed to discuss these five big trends in detail. Models are getting better and hallucinating less, and they’re also being trained to do agentic tasks. Another feature that the model providers are researching is a way to use the LLM as a judge, and as models get cheaper (something we’ll cover below), companies can use three or more models to

OpenAI’s red teaming innovations define new essentials for security leaders in the AI era

Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More OpenAI has taken a more aggressive approach to red teaming than its AI competitors, demonstrating its security teams’ advanced capabilities in two areas: multi-step reinforcement and external red teaming. OpenAI recently released two papers that set a new competitive standard for improving the quality, reliability and safety of AI models in these two techniques and more. The first paper, “OpenAI’s Approach to External Red Teaming for AI Models and Systems,” reports that specialized teams outside the company have proven effective in uncovering vulnerabilities that might otherwise have made it into a released model because in-house testing techniques may have missed them. In the second paper, “Diverse and Effective Red Teaming with Auto-Generated Rewards and Multi-Step Reinforcement Learning,” OpenAI introduces an automated framework that relies on iterative reinforcement learning to generate a broad spectrum of novel, wide-ranging attacks. Going all-in on red teaming pays practical, competitive dividends It’s encouraging to see competitive intensity in red teaming growing among AI companies. When Anthropic released its AI red team guidelines in June of last year, it joined AI providers including Google, Microsoft, Nvidia, OpenAI, and even the U.S.’s National Institute of Standards and Technology (NIST), which all had released red teaming frameworks. Investing heavily in red teaming yields tangible benefits for security leaders in any organization. OpenAI’s paper on external red teaming provides a detailed analysis of how the company strives to create specialized external teams that include cybersecurity and subject matter experts. The goal is to see if knowledgeable external teams can defeat models’ security perimeters and find gaps in their security, biases and controls that prompt-based testing couldn’t find. What makes OpenAI’s recent papers noteworthy is how well they define using human-in-the-middle

Stay Ahead, Stay ONMINE